While rarely seen yet in the wild, a new occupant will appear in our clusters: the Autonomous Agent.

I think we have the right engine, but we are using the wrong blueprints. Kubernetes remains the only platform capable of hosting the agentic future, but the design patterns we’ve inherited, the “best practices” of 2020, are actively hindering the growth of autonomous systems. To serve an agentic ecosystem, we must stop treating agents like static web apps and start treating them like the dynamic, cognitive processes they are.

This post is the extended version of my recently at the Agentic AI Foundation published blog: https://aaif.io/blog/agentic-ai-infrastructure-on-kubernetes/

The Problem with Predictability

The core of the conflict lies in the nature of the workload. Traditional Kubernetes patterns are religious about “Desired State.” A human defines exactly how many replicas should run, how many it must be at least, how many it can be at max and what their limits are.

Agents, by contrast, are inherently non-deterministic. They don’t just respond to traffic; they generate logic, spawn sub-tasks, and mutate their own requirements based on the complexity of the goal at hand. (Keep this thought for a moment in you rhead, we will need it later).

Look at a tool like Goose. When Goose is tasked with a migration, it autonomously decides which tools to invoke. If it needs a temporary Python environment to process a dataset, it creates one. Our current GitOps flow, requiring a human-vetted PR for every infrastructure change, is a brick wall in these cases.

Our current infrastructure treats a “crash” as a failure. But an agent might purposely spin up a temporary environment to test a hypothesis and then tear it all down. This “intentional instability” looks like a failure to a traditional orchestrator, but it is actually the heartbeat of autonomous reasoning.

Building the Four Pillars of the Agentic Runtime

And why there might be much more than that.

I initially wrote “To bridge this gap”, but I think that’s wrong. It’s new requirements, not a wrong development. Companies are looking for stability, not chaos. But Agentic means chaos, for the good and for the bad.

1. Safety through Isolation and Sandboxing

When an agent writes and executes its own code, standard Docker containers (sharing a host kernel) are a liability. They can become a risk in themselves, intended and unintended.

We have to consider at least things like Kata Containers or gVisor as a default and ensure by policies that there is no other way. Google identified this risk early and donated a project called Agent Sandbox to the Kubernetes SIGs for further development.

That requires an isolation for every tool call. If an agent “hallucinates” a destructive script, it occurs in a throwaway environment that vanishes in milliseconds.

2. Stateful Memory Fabrics aka a Cognitive Storage Class

We are moving past the “stateless sin” of early AI. The state, memory, context, indexes, and structure all require persistence, which is fast and efficient but lasts longer than any chat.

Agents required an integrated, tiered Cognitive Memory layer directly into the Kubernetes storage class. This moves from local KV-caches for active reasoning to shared memory for agent swarms (via protocols like MCP), and finally to vector-native storage for long-term “experience.” Besides, classic object storage is in high demand.

One of my thoughts would be the memory sidecar. Instead of every time telling the agent what, how, and where to connect, have a sidecar that deploys alongside the agent and handles memory connections. This way, access management for certain data could also be managed.

Think I’m going to vibecode something :D

3. Dynamic and Fractional GPU Scheduling

We are entering the age of “Bursty Reasoning.” An agent might sit idle while waiting for an API request, then suddenly require a massive burst of compute to plan its next moves. Or you have OpenClaw running…Then it is more like drum fire…

Anyhow, static GPU reservation is an economic disaster. We need fractional vGPU scheduling and spatial multitasking, where the scheduler slices compute power in real-time based on the “importance” of a reasoning task.

We see here a lot of development, especially in the training and fine-tuning context, and the same methods are partially used also for inference. But I think in the future companies will have dozens, if not hundreds, of differently sized models with different requirements. This going to be a challenge to optimize.

4. Observability as Trust

An HTTP 200 status code no longer equals success.

When your monitoring tells you a request completed without error, but the agent’s underlying reasoning was incoherent or simply wrong, you have a false sense of security that is more dangerous than an outright failure. Traditional observability was built for deterministic systems: did the CPU spike, did the endpoint respond? These metrics remain necessary, but for agentic systems they are profoundly insufficient.

The shift is from system logs to reasoning traces. The natural extension point is OpenTelemetry, already deeply embedded across modern stacks, it’s the right foundation to build on rather than replace. What changes is the span payload. Alongside execution telemetry, we need to capture the chain of thought: intermediate reasoning steps, tool-call justifications, decision branches considered and rejected.

But what a challenge it is, please? LLMs don’t emit structured reasoning by default, and a verbose trace from a complex multi-step agent can easily overwhelm conventional trace storage. Sampling strategies and semantic compression will become part of the observability conversation in ways they weren’t before.

5.th? Policy as Reasoning (The Guardrail Layer)

In a world where agents spawn their own infrastructure, we can no longer rely on static RBAC.

The traditional access control model assumes a stable population of known actors requesting a bounded set of known operations. Agentic systems break this not because they are adversarial, but because they are generative. An agent operating autonomously can formulate infrastructure changes no human anticipated at policy-writing time. Static rules cannot cover a dynamic action space.

The shift is toward intent-based governance. The question is no longer just “is this principal authorized?” but “does this action align with what this agent was actually sent here to do?” That requires semantic reasoning to answer which is where the LLM-as-a-Judge sidecar comes in. A lightweight inference process intercepts proposed actions before execution, receiving the agent’s declared mission, its budget, and the proposed action itself, then produces a reasoned verdict on alignment. Only if the verdict is affirmative does the API call proceed.

Time for an Intent-based access control: IBAC.

The companies that treat this as an afterthought will find themselves exposed in ways that are hard to recover from. Intent-based governance needs to be designed into the runtime from the start, because the agents it governs are already reasoning about what they’re allowed to do.

Because of all of these challenges, we need, maybe, finally, Kubernetes v2. A chaos empwoering, secure environment for agents to go wild, without burning a company.

The Dimensions of the Agentic Shift

To truly understand the “Kubernetes v2” era, we have to look beyond the code and into the economics, the network and the human element.

I. Reasoning Economics

In the microservices era, costs were tied to traffic. In the agentic era, cost is tied to Reasoning Depth. An agent might get stuck in a logic loop, consuming $50 of GPU time to solve a $0.05 problem.

We need new patterns. For example, Infrastructure must act as a financial governor, providing Token Budgeting at the namespace level. We are moving from monitoring Cost per Request to Cost per Goal. This might require (not an infrastructure problem though), to have agents maybe questioning what other agents are doing. But especially with agents with autonomous freedome this will become highly relevant.

II. Semantic Networking

Standard networking is built on IPs and DNS. But agents don’t care about “Service A”; they care about “the service that can process a 2024 invoice.”

We need new patterns. Like a Semantic Knowledge Mesh. Using the Model Context Protocol (MCP, as long as it exists), the infrastructure “advertises” its capabilities semantically. The network routes agent requests based on intent and provides the necessary schema context automatically.

III. Infrastructure as Rationale

For the last decade, the goal of SRE was “invisible infrastructure.” For agents, invisibility is dangerous. If an agent changes a cluster config, the human “owner” loses the mental model of the system.

We need new patterns. Think of it like a transparent introspection. When an agent modifies a manifest, the “commit message” isn’t a hash, it’s a link to a reasoning trace. The SRE’s role shifts from “debugger of code” to “editor of intent.” The key issue will be that agents might change their ideas and implementations more often than humans can answer…

From GitOps to AgentOps: The New Kernel

We are moving away from static manifests toward the Agentic CRD (Custom Resource Definition). This doesn’t just define an image; it defines a goal, a budget, and a set of tool permissions.

The “Somehow” Era

For the last decade, a whole industry followed the mantra of: “It must run exactly the same on every machine.” In the agentic era, that shifts to: “It will somehow work on every machine.”

Does that feel scary? Yes! Do I think we are ready for it? No!

But if we don’t solve the issue of unpredictability and raise standards in quality of agentic-generated conditions, we will hit some brutal years.

By embracing this “somehow,” we gain immense power and speed. But we also surrender the comfort of the “frozen” production environment. We are trading the safety of the “known state” for the efficiency of autonomous reasoning.

Kubernetes has won the battle to be the platform. Now, it is time to build the brain.

Sovereignty is a question of perspective

“We”, as Europeans, discuss Sovereignty as it would be a globally settled term that describes the technical possibility of freedom of choice while respecting all of our thousands of rules. But sovereignty is universal. USA is sovereign because they have since decades a booming digital ecosystem, and where the market lacks funding the government and military jumps in. China is sovereign because they don’t want to be dependent on the US, and has invested billions of … Dollars… in creating its own, and open source, ecosystem.

Europe is not sovereign, because we just don’t invest, but also because we have to acknowledge that we are many. Many countries, many ideas, many personalities. Comparing the US and China with Europe is so stupid, like talking about Africa as if it were one country.

(And yes, the EU, and local Governments spend some millions here and there.) What you need is for the market to spend enough money too. Every large enterprise in Europe takes money, takes free and open source. But they don’t give back. At least not in a relevant manner. And if they do, like SAP with their project donated via ApeiroRA to the NeoNephos Foundation, people are screaming like “you can’t decide alone that this is the right way” or “how you dare to tell me how to do it” or “it’s not open if it just comes from one company”. Well, then just do it too, contribute, work, open source, invest money and time into it. Complaining is easy; doing something by yourself isn’t. Huh?

Perspectives

Back to the topic. What I wanted to say is that first, we need to learn to understand and express that sovereignty looks different from every country you are looking at it.

Subscribe now

I always like to think about it this way: Imagine Sovereignty is this Ivory Tower standing in the middle of the country. The country I’m talking about is very bio-diverse. Now, around this tower, far away, are standing people looking at it. Each person represents a country. Some people are standing closer together (Europe), others are farther away from each other (USA, India, China). On our way to the ivory tower, we can see the valley of a thousand rules, the desert of no money, and the swamp of burocrazy. On the other side, there is USA standing. They are not far away from sovereignty, but they are very tightly connected to the ivory tower by the money chain, the political mind eater, and the VC bros pushing you from hype to hype. For every country, this perspective looks different. For some, it is easy and flat, dictating a rule and spending money. Others have no chance than to ask for a helping hand. And for some more, this topic is completely uninteresting.

The point is, if we discuss Sovereignty, we can do that from our EU perspective. But we shouldn’t make the mistake of defining ideas that mean that solutions will become less Sovereign for others. Then we are not better than what we try to get detached from.

Sovereignty is not an ON or OFF; it is a path

In those sovereignty discussions, you often have those naysayers, “We will never become sovereign because of reason A, B, C”. Like many things in live, its not a binary option: Yes or No, You are or You aren’t.

Sovereignty is a path. You don’t ask “Are you sovereign?”, you ask “How sovereign are you?”. That’s the only way you can discuss this. You have many different missions on your way to max out your sovereignty. Some have a focus on legal actions, others on encryption and the protection of IP, and for others on the software.

Sovereignty in the context of IT refers to having control over an organization's technological assets, data, and operations, which can mean ensuring data stays within national borders and laws while respecting those set up.

However, sovereignty is also not free. You have to work for it, you have to spend time on it. Somehow, the tenor is getting louder that every software must be free and open. Where does that nonsense come from? How do you want to develop something enterprise-grade if you don’t pay for it? Because of the goodwill of someone? And don’t come with Linux, that’s the one in a million shot.

Yes, but hardware

The hardware issue in the sovereignty discussion is real. But I think we put to much weight on it. Why?

Hardware is a truly global product. The design for the components, like the chips, is US-made. Most components are produced in Taiwan and China. The raw materials for the production come from around the world, but many mines are owned by China. The chemicals needed for the production often come from Japan and South Korea. Besides the chips, you have a couple of other microcontrollers and components in a server. While they are produced in the same places, the software for it comes from many countries. The machines used to produce chips and other fine-mechanical components come from Europe (the Netherlands, Switzerland, and Germany). The logistics to transport those around the globe are primarily done by Switzerland, Denmark, France, China, Germany, and Taiwan (the US practically doesn’t play any role in it).

Therefore, a couple of large providers are building their own servers, yet, they need the global supply chain.

The point (1) is, yes, we have on paper a heavy dependence on the hardware. But we don’t depend on the hardware. You don’t like server A, then buy server B, you don’t like that, get server C. Most hardware is exchangeable. (2) If one country, for reasons, wants to cut out others by supplying hardware, that boomerang will come back fast by getting cut out from other required supplies. And to stop this machinery isn’t that simple. Because (3), the digital supply chain is way easier to turn on and off, which is, in these politically interesting times, way simpler to do.

I know, people will still cry on this point, because it is “not our hardware”, but it is also not your Linux or your DNS service.

What Europe has to do

To turn now to the sovereignty perspective in Europe, we do not have many options, but some.

We have to invest in building our own computing capacity and capabilities through EU-based providers. AND we need to follow a second path to build an abstraction layer, that doesn’t care which provider it is running on. I think we don’t consider this abstraction layer enough. Think of it like “if it can run everywhere, and it can run everything, you can build anything, anywhere you need, to the same conditions, or change the infrastructure as you need if conditions are not met”.

Innovation

Aiming to establish something someone has been doing for over 10 years is a nice race, but it's nothing we need to own. We need different approaches to succeed:

• Photonic Computing

• RISK-V

• Neuromorphic Computing

• Analog In-Memory Computing

None of them “replaces” classic CPUs/GPUs et al, but either enhances them, or are useful in the edge. So two scenarios are relevant here: 1. Europe's future is maybe less in the cloud and more “on the edge”, in the intersection between Robots, AI, and interacting with the world. 2. Creating and strengthening the partnership and dependency on GPU/CPU providers. While we want to reduce dependencies, providing a unique selling point that is not possible to ignore might make one act more in your way. (Being naive here, but you know, all politics will have an end, in the end :D)

Open Source

The CNCF showed at the last KubeCon + CloudNativeCon in Europe that more open-source contributions come from the EU than from the US. But at the same time, we spend almost no money on this.

• If we want more sovereignty, we have to force or create environments that make it appealing for companies to invest in it. Tax reduction, bonus points in public tenders, whatever is needed. BUT, it must be true and valuable contributions. Not just a bot commenting under PRs.

• Additionally, individuals contribute a lot to open source for free. We should give them tax benefits for this, too, and in my opinion, this shouldn’t be small.

• We need to battle the narrative that Open Source must be European. Like it has a citizenship… Open Source is global; the thing we can do is give it our values. Most people who state this just want to push their own business, or are hardcore open-source believers (which we all love, but are in their own bubble).

When I started contributing to Kubernets 1.17, I pushed my company to Top 100 contributors. When we were three people contributing, we were in the Top 30, alongside all the global players. If we can do that, why can’t you?

Open source works that way: The more you do for an open source project, the more you are involved and engaged, the more responsibilities you get, and the more you can shape its future. In other words, if you don’t have competencies and don’t move your 🍑 for it, what do you want to influence then? Alternatively, throw dozens of millions on those projects every year.

Politics and Businesses

I don’t want to discuss policy-making; it sucks, yes, but we also need it to make it attractive to companies to invest in such innovative and/or open source topics.

European companies need to find their way to become a little bit bolder and braver to test new things. All of them are wasting money on a lot of useless stuff. But having money to invest in open source, migrating to another cloud provider, or supporting innovative ideas falls short. In my opinion, we can see a huge mismanagement currently happening. Without spending heavy money on those topics, it will not be possible to evolve for local providers.

We somehow lost the capability to co-create. In the past, it was common to work closely with vendors to improve products and make them fit your needs. Now, you just manage contracts and complain if the SLA is not met. Very innovative.

Stop drawing Devils everywhere

In the end, I wanted to highlight that we should stop drawing the devil on everyone’s wall. US big tech has pushed us so far in tech innovation; it’s nothing but pure stupidity to make them what they aren’t. Yes, they might collect your data, but you also use a hell of a lot of stuff for free. And nothing in life is for free.

Also, EU, African, South America or Asian companies don’t wear clean white gloves. It might look better from our perspective, but again, that’s a question of perspective.

Look, I spent a lot of my last 6 years in open source. The one thing I can tell you: without the big players like Google, Amazon, Microsoft, Nvidia, Red Hat, Intel, Apple etc. we wouldn’t have such a rich, stable and solid developed open source landscape. And I already see the hardcore free and open-source people commenting under this: “they do it just for their purpose, for their products, that’s not free”. Yes, that’s true, and now? Almost no open source solution would have made it to an enterprise-grade ready state without someone believing in and investing in it. Also not Linux.

For Europe, I hope we learn to invest more, be more flexible, just do more, without discussing less. You learn when you fail, not when you debate about potential failure you never experience. And innovation happens at the same place where you fail a lot.

Something quietly remarkable happened in 2025: the most advanced AI models in the world started to look alike. Not identical, but close enough that picking between them began to feel less like a strategic technology decision and more like choosing a brand of coffee. The beans are roughly the same (and I know, beans are never the same, roasting etc., but let’s keep it simple); you’re really choosing the cup, the barista and the “café ecosystem” around it (how easy to get there, nice area, weather, etc.).

This is the commodity paradox of generative AI. After some wild competitive years to build ever-larger, ever-more-capable models, the frontier has converged. Benchmarks cluster. Prices have collapsed. And the companies building these models are now competing ferociously on something entirely different: the platforms, tools, protocols, and ecosystems that surround them.

“Models by themselves are not sufficient for a lasting competitive edge. OpenAI is not a model company; it’s a product company that happens to have fantastic models.” — Satya Nadella, Microsoft CEO, 2025

I want to give a simple but consequential argument: in the GenAI era, the model is the CPU. The platform is the computer. Nobody buys a CPU anymore; they buy a system. And the same logic now applies to artificial intelligence.

Models Are Converging

Benchmark Parity at the Frontier

ChatGPT, Claude and Gemini trade positions across benchmarks, same for Kimi, Qwen, DeepSeek and so on. Each shows dominance in some dimension, none dominant across all.

One 2025 LLM comparison study summarized the situation plainly: “With a few exceptions, Anthropic and OpenAI’s flagship models are essentially at parity, meaning your choice of any particular LLM should focus on specialized features rather than raw intelligence.”

So, benchmark scores are no longer a reliable strategic differentiator.

And open-source models have accelerated this convergence dramatically. DeepSeek-V3 matched frontier proprietary models at a reported training cost of under $6 million. Llama 4 and Qwen3 compete credibly at the top of public leaderboards. The existence of capable, free, self-hostable models has altered the power dynamic. No lab can maintain a lasting performance lead when its architecture and training recipes can be approximated by well-resourced competitors within months.

The question is, why then still use commercial ones? Well, with open-source models, you have to build and integrate your own ecosystem. Tools like Claude Code or Gemini's all-in-one solution are too easy to use and too effective at solving your problem.

More voices see it like this, too

The commodity thesis is no longer a fringe contrarian view.

In December 2025, during Microsoft’s AI Tour in Mumbai, Satya Nadella said: “There are lots of capable models available today. The question is how you bring those capabilities together.” He highlighted the concept of ‘context engineering’ as the real competitive edge. The performance of an AI application, he argued, is “directly correlated to how well your data is brought into context.” But what he meant is the ability to ground AI systems in enterprise data, workflows, and relationships.

Thomas Wolf of Hugging Face has articulated the same transition differently: the focus is shifting from models to the systems that leverage them. He draws an analogy to the early internet: in the beginning, building a website was itself the business; today, companies build internet-native businesses by integrating web capabilities into broader systems. The underlying technology became nearly free. The value moved to the application layer and how it plays with other systems together.

I attended the Bits&Pretzels in Munich 2025, where Mistral AI's CRO, Marjorie Janiewicz, was talking in an AMA about how the platform and the services, in addition to the model, provide the most value to customers. Rather than everyone having to build and invent the whole system, it makes sense to come up with the entire tool stack with all batteries included.

Subscribe now

Building the Platforms is now the Strategy

If the model is no longer the moat, what is? The answer is playing out in real time across every major AI lab: they are racing to build platforms and ecosystems of tooling, integrations, agent infrastructure, and developer experiences that lock in users at a layer above the model itself.

OpenAI: From Model API to Agentic Operating System

OpenAI’s headline for the year was not a model milestone, it was: “2025 wasn’t about a single model launch. It was the year AI got easier to run in production.”

The platform buildout was comprehensive. The Responses API, launched in March 2025, gave developers built-in tools for web search, file search, computer use, all accessible through a single API call. The Agents SDK (available in Python and TypeScript) provided building blocks for multi-agent orchestration, handoffs, guardrails, and tracing and was deliberately made provider-agnostic, able to run non-OpenAI models. AgentKit added higher-level capabilities: an Agent Builder, ChatKit, a Connector Registry, and evaluation loops for teams wanting to ship faster.

The Codex CLI brought agent-style coding into local environments. So developers could run AI directly over real repositories, apply changes iteratively, and keep humans in the loop for long-horizon tasks. By mid-2025, reasoning capabilities were no longer a separate product family but folded into the mainstream GPT-5 line. The strategic message was clear: the model is a component; the platform is the product.

To put a big black bold line under this agentic operating system thinking, OpenAI acquired OpenClaw/MoltBot (whatever its name now) and its inventor. OpenClaw is an autonomous agent that can execute tasks via large language models, using messaging platforms as its main user interface.

Anthropic: Protocol as Platform

Anthropic’s most consequential strategic move in 2024–2025 had nothing to do with model capability. It was the release, in November 2024, of the Model Context Protocol (MCP), an open standard for connecting AI agents to any data source, tool, or application.

Anthropic’s Chief Product Officer Mike Krieger called MCP the “USB-C interface of the AI field”: a single standard connector that eliminates the need for custom integrations between AI systems and the countless databases, APIs, and services they need to interact with. The protocol spread with extraordinary speed. Within one year of release: over 97 million monthly SDK downloads, more than 10,000 active MCP servers deployed, and first-class support integrated into Claude, ChatGPT, Gemini, Microsoft Copilot, Visual Studio Code, Cursor, and dozens more.

“The main goal is to have enough adoption in the world that it’s the de facto standard. We’re all better off if we have an open integration center where you can build something once as a developer and use it across any client.” — David Soria Parra, MCP co-creator, Anthropic

In December 2025, Anthropic donated MCP to the Agentic AI Foundation (part of the Linux Foundation, the same institution that stewards Kubernetes, PyTorch, and Node.js).

This was not only an act of charity. It was a declaration that MCP is infrastructure: the neutral plumbing layer on which the entire agentic AI ecosystem should be built, owned by no single vendor.

Google: Agent-Ready & Hardware Independent by Design

Google’s platform strategy is grounded in its unique position: it controls more entry points to end users (Search, Gmail, Android, Chrome, Workspace) than any other technology company on earth. The AI strategy follows naturally: make all of Google’s services accessible to AI agents, and make Google’s AI ecosystem the place developers build.

In December 2025, Google launched fully managed, remote MCP servers for its Google and Cloud services like Maps, BigQuery, and more. Steren Giannini, Product Management Director at Google Cloud, framed the strategy in plain terms: “We are making Google agent-ready by design.” The servers are protected by Google Cloud IAM and Model Armor, with full audit logging, recognizing that enterprise adoption requires governance as much as capability.

In my opinion, it's also important to mention that Google is, in comparison to the others, very robust and uniquely placed against the AI hype, too. With its own hardware designs especially for AI workload (TPUs) and its own excellent AI forethinker lab with DeepMind, it doesn’t need to participate in this huge money shift around games.

Last, Google’s NotebookLM is perhaps the clearest product-first model: built around Gemini from the beginning, with a 1-million-token context window, audio and video overviews, grounded document search, and tools that make the AI output inherently more useful than raw model output. It illustrates the ‘LLM as long-context research engine’ pattern that will define the next generation of AI products.

Mistral AI: Sovereignty as a Platform Differentiator

Mistral AI, the Paris-based startup founded in 2023 by former Meta and Google DeepMind researchers, offers perhaps the most instructive case study in the platform-over-model thesis. It entered the race without the deep pockets of its American rivals, and has been forced to compete on platform strategy from day one.

Mistral’s CEO Arthur Mensch says: the company’s goal is to enable enterprises and governments to “customise the technology and make it their own so they can control it without external dependencies.”

In October 2025, Mistral launched Mistral AI Studio — replacing its earlier La Plateforme offering with a production-grade environment that directly addresses the gap between AI experimentation and enterprise deployment. Studio integrates four layers in a single platform: a Runtime for executing models and agents; an Observability layer for monitoring and auditing outputs; a Registry for governing model versions and tracing outputs back to source components; and a Workflow Builder (in beta as of December 2025) for building multi-step, automatable AI processes without custom scripting. MCP support was announced as an upcoming integration, connecting Studio to enterprise systems and custom tools.

“The real bottleneck is software. What’s missing are the tools required to supervise, verify, and secure the agent’s actions. We have issues of observability, compliance, and control.”

— Timothée Lacroix, CTO & Co-founder, Mistral AI

The strategic logic Mistral embodies is the clearest expression of the platform thesis from a challenger perspective: when you cannot out-spend OpenAI or Google on model training, you compete on the layers they find harder to replicate:

• deployment sovereignty

• European regulatory alignment

• on-premises control

• applied integration expertise

Mistral is not selling a model. It is selling the assurance that AI can be deployed on your terms, in your infrastructure, under your governance.

Where Models Still Differ: User Taste

To be clear: models are not entirely identical. Meaningful differences remain, in style, personality, and areas of specialization. But these differences have become matters of preference rather than capability gaps.

Claude is consistently described as having “the most soul” in writing with producing vivid character development, immersive world-building, and narratives with emotional depth. ChatGPT’s GPT-4o is praised for its comprehensive toolkit, persistent memory across conversations, and superior instruction-following for creative and visual tasks. Gemini 3 Flash is favored for speed and cost efficiency, particularly in multimodal workflows deeply integrated with Google services. Grok from xAI wins on access to real-time information. DeepSeek wins on price-to-performance for cost-sensitive deployments.

These are not trivial distinctions. They are analogous to choosing between television brands. The underlying panel technology is essentially the same. The brand experience, ecosystem integration, and software capabilities determine the choice. The competition has moved from the chip to the system.

What This Means for Enterprises and Developers

The platform-over-model reality has concrete strategic implications that many organizations have yet to fully internalize.

Evaluate Platforms, Not Just Models

When selecting an AI vendor or infrastructure, benchmark scores should not be the primary criterion. Instead, evaluate:

1. the richness of the platform’s tooling and integrations

2. the maturity of its agentic capabilities

3. support for local and on-premise deployment

4. fine-tuning and customization options

5. governance, compliance, and audit capabilities

6. the openness and interoperability of its ecosystem

Finally, your Data Is the Moat

The single most important message from industry leaders is consistent: competitive advantage in the AI era comes from how effectively enterprise data and context are brought into AI systems. The model is a commodity; the grounding of that model in proprietary knowledge, workflows, and relationships is not.

This means the strategic priority should be building robust data infrastructure with clean, accessible, well-governed enterprise data, rather than chasing the latest model release. The team that can most effectively bring its institutional knowledge into AI context will outperform the team with the nominally better model.

The PC Moment for AI

History has a pattern for this. The transistor commoditized. The CPU commoditized. The operating system commoditized. The cloud server commoditized. In each case, the initial scarce resource became abundant, value migrated up the stack, and the winners were those who built compelling systems on top of the now-cheap substrate.

We are, already, in the PC moment of artificial intelligence. At least for this round of the AI hype. The raw intelligence of frontier models is becoming abundant and interchangeable. The race is now for the platform: the agentic infrastructure, the integration ecosystem, the developer experience, and the enterprise data layer that turns a capable model into an indispensable system.

The question isn’t ‘which LLM is best?’ but rather ‘which combination of tools best serves your specific workflow?

Necessary Intro

I think it is important to explain shortly how markets and businesses work. Usually, if you buy something, money is distributed and allows others to spend money. However, if we zoom very far out, money always runs in circles. But just on such a big scale, we will practically never see the same coin again.

So let’s say you go with a friend to drink a coffee. 20 bucks gone. The cafe owner pays from this the coffee beans, water, electricity, someone for marketing, a credit, the employees, a nearby grocery store etc. And from here your 20 bucks keep distributing. E.g. the employee is saving money to buy a bicycle. A new breakdown starts. And so on. The entire market is connected, and the scale is difficult to imagine, but it is huge. Therefore, money that flows is good for everyone.

Now, what we are going to talk about today are very short cycles. So short that you will see the same coin multiple times.

The Money Flow

OpenAI has made crazy infrastructure commitments totaling approximately $1.15 trillion through 2035, broken down as follows:

• Broadcom: $350 billion (custom chip design)

• Oracle: $300 billion over 5 years (4.5 gigawatts of capacity)

• Microsoft Azure: $250 billion (cloud services/credits)

• NVIDIA: $100 billion (planned investment partnership)

• AMD: $90 billion (GPU purchases, gaining up to 10% stake in AMD)

• Amazon AWS: $38 billion over 7 years (cloud infrastructure)

• CoreWeave: $22.4 billion (data center capacity)

Now, in US you have to live on a big foot. (idk if that make sense, it’s a german saying). OpenAI’s 2025 revenue is projected at only ~$13 billion annually. While the company remains unprofitable with losses of approximately $5 billion in 2024.

The company expects to spend more on inference (running AI models) than it earns in revenue.

Morgan Stanley Research provided a nice overview for this.

What is in it for everyone?

Microsoft

Microsoft is very complex. They invested into OpenAI, are provider, own a stake and get 20% of the revenue of OpenAI and so on.

In other words → Microsoft both invests in OpenAI AND is OpenAI’s largest cloud provider, creating a self-reinforcing loop where Microsoft’s investment helps fund purchases of Microsoft’s own services.

NVIDIA

NVIDIA is somewhat the center water pump for a circular flow. NVIDIA invests in AI companies → Those companies use the money to buy NVIDIA chips → NVIDIA reports strong revenue → Stock price rises → More investment capital available. The investments are going into OpenAI, Anthropic, CoreWave, xAI, Mistral AI, Intel and more.

NewStreet Research estimates that for every $10 billion NVIDIA invests in OpenAI, it expects $35 billion in GPU purchases back. A 3.5x return that raises questions about whether this is an investment or subsidized demand.

Oracle

The big OpenAI Stargate partner comes with an initial equity for the Stargate project and a long-term contract from OpenAI, while building multiple gigawatt-scale data centers in Texas and other locations.

Oracle is borrowing heavily to finance this buildout, expecting OpenAI to generate enough revenue to pay for services. If OpenAI fails, Oracle is left with massive debt and underutilized infrastructure.

SoftBank

Goes big into the Stargate project but also has to borrow against Arm Holdings, liquidating T-Mobile stakes and using its cash reserve.

SoftBank has until year-end 2025 to secure $22.5 billion, and has sold its NVIDIA stake to help fund the venture.

CoreWeave

Despite the contracts and investments with and of OpenAI, Meta, Microsoft, and NVIDIA (and many more), it is just a paradox. CoreWeave spent $1.9 billion in capex in Q3 2025 alone, with projected annual capex of $30 billion in 2026. The company has a capex-to-revenue ratio of 2.3-2.8x, meaning it spends $2.35-$2.77 for every dollar earned. It remains deeply unprofitable despite massive revenue growth.

Amazon, Meta, Google

Also, they are part of investments with OpenAI, they are also driving the diversification in the market.

Amazon invests in Anthropic, and they, in return, commit to using AWS services.

Meta works with CoreWeave, invests heavily in Scale AI, and develops its own model (Llama) as open source.

Google has its own strong ecosystem, data center capacities, its own chips like the TPU, its own AI experts like DeepMind, and so on. Yes, there are also deals here and there with other providers etc, but we could say Google is the most independent from that whole Ring Around the Rosie.

The Bubble Mechanics and Risks

1. Circular financing creates artificial demand

The pattern is remarkably simple and troubling:

1. NVIDIA invests $100B in OpenAI

2. OpenAI uses that money to buy NVIDIA chips

3. NVIDIA reports record revenue from “customer demand”

4. Stock price rises

5. NVIDIA has more capital to invest

6. Repeat

During the dot-com bubble, companies like Global Crossing engaged in “revenue roundtripping” paying other companies for services, who then paid them back in equal amounts for equipment. When the bubble burst, Global Crossing went bankrupt.

2. Massive debt accumulation

Unlike the tech giants (Microsoft, Meta, Google, Amazon) who can fund AI from cash flow, the AI startups are heavily leveraged:

• OpenAI: Unprofitable, burning $5 billion/year

• CoreWeave: $30 billion capex planned for 2026 vs. ~$5 billion revenue

• Oracle: Issuing massive debt to build infrastructure for OpenAI

• SoftBank: Borrowing against assets to fund Stargate

• (by the way, ALL AI companies are unprofitable, doesn’t matter which one we are talking about)

If AI revenue growth doesn’t materialize, these companies face debt defaults that could cascade through the entire ecosystem. It’s like rock climbers who are chained to each other to prevent them from falling. The problem is, the more that fall, the more it drags others with it.

3. Revenue concentration

The entire ecosystem depends on a tiny number of players. A CoreWeave made 77% of its 2024 revenue from just two customers. OpenAI is burning through investment faster than it grows revenue. And the cloud providers are competing to servce the same AI companies.

If any major player fails or slows spending, the ripple effects would be, well, “interesting”.

4. Valuation disconnect

Ok, so we know that this whole market is not profitable. That’s how US market works. Bigger, louder, better. Fake it until you make it. Everything on Green.

Bain & Company estimates the AI industry needs $2 trillion in annual revenue by 2030 just to justify current infrastructure spending. No problem? This is more than the combined 2024 revenue of Amazon, Apple, Alphabet, Microsoft, Meta, and NVIDIA.

5. The Inference cost problem

Recent leaks show OpenAI may be spending more on inference (running models) than it earns.

This suggests the more customers OpenAI gains, the more money it loses. A fundamentally unsustainable model.

6. Limited outside revenue

Despite billions in investment and trillions in commitments, actual revenue from paying customers remains modest.

• OpenAI ChatGPT Plus: 11 million subscribers at ~$20/month = ~$2.6 billion/year

• Enterprise customers: Growing but insufficient to cover costs

• Most AI products still searching for viable business models beyond subscriptions

MIT Media Lab research found that despite $30-40 billion in enterprise GenAI investment, 95% of organizations are getting zero return.

What do the “Experts” say (warn)?

Michael Burry a “Big Short” investor, is betting against NVIDIA saying it is “Almost all customers are funded by their dealers”. And questions OpenAI’s accounting → “Can anyone name their auditor?”.

Dario Amodei, Anthropics CEO, even as his company participates in circular deals, warns about “players who are YOLOing”. Not sure if this should be a reference to Sam Altman who lives in a world of “Trust me bro”.

Investman analysts like Jay Goldberg, Stacy Rasgon or Ben Inker calls NVIDIA-OpenAI deals “bubble-like behavior”, “The action will clearly fuel ‘circular’ concerns” and “We’re certainly seeing lots of evidence of bubble-like behavior in the AI space”.

The problem is that due to the high stakes and investments a single issue can trigger a collapse. OpenAI can run bankrupt, and I believe that is very likely as the OpenAI CFO in a recent interview danced very long around an asnwer to finally tell us that they are fully open for governmental investments. If circular financing is exposed as artificial demand, NVIDIA’s stock could crash, eliminating the capital fueling the ecosystem. Frankly speaking, imo that is the most obvious thing on the table…

Anyhow, there are multiple scenarious analyst have in sight. From energy constraints, to technical disappointments, regulatory interventions (Europe), or competing technology will promise more (e.g. that LLMs are not the solution for AGI, as promised).

Counterarguments

Ok, we also have to take a look in the other direction. The most discussable point is that AI is delivering actual value, unlike many dot-com companies with no product. To some extent, yes, it does. I personally use it daily, but I can’t tell you, if that makes me really more efficient or not. And then we look at the research of MIT which clearly shows, the investments from companies doesn’t turn around…

OpenAI keeps growing and projects a $20 billion run rate by year-end 2025, potentially $100 billion by 2027…

Microsoft, Meta, Google, Amazon can absorb losses and continue investing. For them it’s a spiel. If OpenAI doesn’t make it, then someone else. And as we see with for example Google, they literally would profit from a failure of OpenAI. This would completely blast away Microsoft from the AI radar. Besides I think unlike worthless stock certificates, data centers and chips retain value. The infrastructure has a residual value, and cloud demand (besides AI) is growing.

Maybe, also the bubble has to do its bubble thing. Because, the internet boom eventually paid off after the bubble; AI could follow the same pattern.

Finaly, the Federal Reserve Chairman Jerome Powell argues AI differs from dot-com because corporations are generating large revenues and driving economic growth. Yet, again I think this is highly discussable. With over 80% AI projects failing, and those who succeed seeing no return on their investment, it might be just a lot of hot air. Reality is what we make from it.

Conclusion - A dangerous Game

The AI boom represents the largest technology investment cycle in history, dwarfing even the dot-com era. But the circular nature of these investments, where the same dollars flow in loops between a handful of players, creates a questionable instability.

The ecosystem works perfectly as long as:

• Everyone keeps investing

• No one demands profitability

• The promise of AGI justifies infinite spending

• Stock prices keep rising

But if any link in the chain breaks like OpenAI can’t pay Oracle or CoreWeave’s debt becomes unsustainable or NVIDIA’s revenue is revealed as artificially inflated, the entire structure could collapse with devastating speed.

The question isn’t whether AI will transform the world. It almost certainly will.

The question is whether the current financial architecture supporting that transformation is sustainable or whether we’re building a trillion-dollar house of cards that will leave investors, companies and the broader economy picking up the pieces.

As Anthropic’s Dario Amodei warned: Even if the technology fulfills all its promises, players who get the timing even slightly wrong face catastrophic consequences. With over $1.4 trillion in commitments based on companies that aren’t profitable and revenue models that remain unproven, the margin for error is razor-thin.

The AI revolution may be real. The AI bubble is looking increasingly real as well.

Here is a breakdown of the tools I’m using, the money I’m saving, and the few hurdles I’m still trying to jump over.

by the way, this post is for personal setups not for companies. Yet, I think most tools can be used for teams and companies too.

The Backbone: The Proton Ecosystem

The biggest impact on this migration has come from Proton. They have matured from a simple encrypted email provider into a (almost) full-suite productivity powerhouse.

before we continue, yes Proton is not in the EU, but the Swiss data protection is aligned with EU requirements, are on some cases even stronger and implicit, while on others being more open and flexible to the end user provider…

My subscription now covers:

• Mail & Calendar: Encrypted and clean.

• Drive: Secure cloud storage, having docs for a while, and now also tables.

• Proton Pass: A solid password manager with 2FA integration.

• VPN: Fast and reliable.

• Standard Notes: My go-to for note-taking, supported by the Proton team.

• Lumo AI: A privacy-first GenAI, I’m not yet frequently using it, but we will talk about it later.

Everything is integrated and the user experience is superb. I’m also eagerly awaiting the release of Proton Meet to complete the suite.

This entire setup replaces for me my Google Drive and Gmail ecosystem. Plus NordVPN, Notion, 1Password and Authenticator.

Curious? I will drop you my referral link here.

My mixed AI Strategy

AI is the hardest thing to decouple from Big Tech, but improvements are happening fast here, too.

For privacy-first GenAI tasks within my workflow, I’m using Lumo AI. It’s great for quick, private queries.

However, sometimes you need raw power. For that, I started using Mammouth. I use this less for privacy reasons and more for the sheer value and flexibility. Getting access to every major AI model (including image generation) for just €10 is incredible value.

My default for Mammouth looks like this. You can sort them in your favourite order; whatever is leftmost is your default one.

The two models I use most are Mistral Medium 3.1 and Flux 2 Pro or Fast. For simple coding tasks, I also use Mistral, but when playing around with a larger codebase, I have to admit I’m using Claude Code.

For research, Mistral does its job. But if it becomes complex, I often find Gemini's results to be the best.

At last, Flux for images is fantastic. Yet, in my opinion, it comes with a major downside: You have to give a lot of good instructions. Where a Nano Banana seems more creative, even from short, simple inputs, Flux needs precise orders.

Browsing, Search, and Language

• Browser: I moved to Vivaldi Technologies. It’s highly customizable and respects user data.

• Search: My default is Ecosia. It works well, and planting trees while searching adds a nice layer of purpose to mindless scrolling. Yet, every now and then I have to use Google :/

• Translation: I’ve been using DeepL forever. In my opinion, it is still miles ahead of Google Translate in terms of nuance and quality.

• Spell Check: … Haven’t found a good alternative, so I stay with Grammarly. Hoping for them to return to Europe with their Superhuman Platform.

Infrastructure: Hosting & Domains

For my website and domains, I moved everything to Scaleway.

If you are technical, you will appreciate this switch. It is lean, simple, and provides everything you need as a cloud provider without the bloat of AWS or Azure. Plus, it’s cheaper.

Creativity & Task Management

Canva… what shall I say.

Thanks to the initial discussion I had on LinkedIn (original post), I found Superlist, to which I transitioned from Todoist. I also had a mini stop at MeisterTask, but that was a complete waste of time.

MeisterTaks feels like a kindergarten playground, too much colorful design and too little actual functionality. It is 100% not user-intuitive for any workflow.

So, I’m now on Superlist, and very happy with it.

The Economics: Privacy is Actually Cheaper

We often assume that “boutique” privacy tools cost a premium. Surprisingly, my migration proved the opposite. Initially, I wrote only about a small saving, but I missed the costs of Notion, Todoist, 1 Password, Claude and Canva as I focused on the “Office” suite.

My Old Stack cost ca. 83€ per month

My New EU Stack cost ca. 39€ per month

I’m saving over 528 € a year while owning most of my data.

Where There is Light, There is Shadow

I want to be transparent: you cannot escape everything, and some things are just harder to use.

1. The “Guilty Pleasure”: As a techie, I have too much fun playing with Claude Code. It’s a luxury I treat myself to, so I frequently turn the subscription on. Yet, if you don’t have that use case → Mistral, Lumo

2. The Social Web: You simply can’t get around LinkedIn, GitHub, YouTube, Medium, Substack and so on if you want to stay connected.

3. Convenience: I miss a good Google SSO. It is everywhere, and losing that “one-click login” friction does make life slightly more annoying. I migrated every Login to Proton Pass using MFA and Passkey wherever possible.

4. Office Suite: I am struggling to get used to LibreOffice and Collabora Online. They feel similar to MS Office, but “not quite.” Since I don’t create documents or spreadsheets every day for personal use, the learning curve feels steeper than it should. Screw that → Proton Docs and Sheets are doing it. Worst case for slides, I’m going for Canva, I’m paying for it anyhow.

5. Blogging, Newsletter & Co.: Well, as you can see, I’m writing on Substack. There are no alternatives except to host it entirely yourself, but that doesn’t make sense to me right now.

The Unexpected Gems I found

I also found a couple of positive side effects. The Proton platform itself has everything you need for your day-to-day life.

In addition, I migrated to a Duo plan with my wife. Together we have 2TB on storage for Mails, Files etc. Before we had 30GB on Gmail, which cost a little more.

Proton Pass creates anonymous email addresses in case you don’t want to use your real email address. This plays entirely on Proton's privacy aspect.

Superlist is free for the same features Todoist provided me for a little subscription.

Lastly, I’m always fighting with note-taking. I tested everything and started building something complex on Notion. I’m glad that I could delete it and just use Standard Notes from Proton. In simple terms, it is like notes from Apple, yet E2E encrypted and you can back them up offline. However, most helpful for me is that I finally found a way to work effectively with notes, keep it clean, lean and functional.

The Takeaway

With Proton, Scaleway, Mammouth, Vivaldi, Superlist and DeepL, I have built a useful toolset that, in my opinion, surpasses what I used before.

The apps are cleaner, the UI is often more user-friendly, and the migration was surprisingly simple. Best of all, I can do more with my tech stack for less money.

If you’ve been on the fence about migrating to EU-hosted solutions, take the leap. It’s worth it.

First of all, I have to say that the open approach is a refreshing change. Perhaps this will finally allow modern and innovative technologies to be brought into play. Even if I don’t quite understand what the Federal Ministry for Digital and State Modernization wants to work on for 2-3 years. It’s not rocket science. But hey, #Neuland, right?

I would also like to say that if I come across as too sarcastic, speak badly, or criticize harshly, it is mainly because this is “just another stack” that, at first glance, defines or says absolutely nothing new and is essentially an alternative version of existing landscapes.

Structure, landscape, and more

To maintain the tech stack, it helps to open up the map and consult the stack documentation.

First, the stack is divided into certain groups and layers (those in bold/italics are not yet included in the map):

• Infrastructure

• Development, security, and operations (DevSecOps)

• Platform

• Basic service

• Applications and services

• Interface and access

• Strategy, architecture, and governance

Infrastructure

Interestingly, infrastructure is actually treated as a physical element here, including data centers, protocols, networks, etc.

This should be removed from the map. On the one hand, listing protocols is reminiscent of the subject of computer science fundamentals, and on the other hand, this is a sub-area that may only be of interest to a few.

And in my experience, those few do not need a list of these technologies.

Development, security, and operations

This layer clearly shows that the people working here have not been actively involved in IT for some time, or have never been.

First, I would separate development from operations. I would then reintroduce the topic of observability (monitoring, logging, tracing, audit logs, etc.) into the operations layer—this was already present in the CNCF Landscape—which is already very relevant for operations (!?).

The security group is a joke... Pretty much everything is missing: key management (OpenBao), policy engines such as OPA or Kyverno, OpenFGA for ReBAC, for security scans and runtime security – Falco, Trivy, Kubescape. At the very least.

Also, “commissioning,” now known as CI/CD and GitOps... I don’t understand the map here. GitHub Actions? Cool, GH Actions is great, but how does it fit in here? Argo (CD, WF, etc.), Tekton, and Travis CI are missing from the list of minimum tools. And in practical terms, it could be expanded to include other solutions in the area of fleet management.

App definition, integration, and build should be included again in the new “development” layer. The list of programming languages is, well... cute. But dapr, Helm, Artifact Hub, Buildpacks, Packer, Tilts, etc. add value to a platform.

As a platform engineer, I might even add Backstage, but honestly, that’s too complicated for the target end users.

In development, you can also include development environments, testing solutions, etc.

Platform

Actually a bit more mature than the rest of the map. But not really good either. The integration group is very confusing. Perhaps it would be better to divide it into orchestration and integration? There is a lack of relevant tools and solutions around Kubernetes. This applies to the basis for storage, for example, as well as the container runtime area and automation. The integration (or perhaps better network) layer can then be filled with modern things such as mTLS, APISIX, CoreDNS, and the like.

The absence of NeoNephos projects is disappointing here. We already have a cloud read toolset that is completely OSS and is already funded by the EU and IPCEI-CIS...

For me, the platform should also include elements such as kubevirt, Knative, and KCP—solutions for providing workloads in different forms and colors.

Basic services

On the other hand, I find the area of basic services very exciting: the focus here is on notification systems, payment, wallets, etc. I would have liked to see something on this. What are the proposals for identity management? Where does geodata come from? Which certificate providers should be used?

Applications and services

Sounds a bit like: https://european-alternatives.eu/de/alternativen-zu ?

Strategy, architecture, and governance

Unfortunately, nothing. A fundamental strategy would at least give the whole thing some direction. Are we OSS first? If so, are companies encouraged to invest in OSS? Or are there compromises, such as purchasing pseudo-sovereign solutions?

Summary of the map

It would have been easier to simply keep the CNCF map and, for example, remove all purchased software solutions and integrate another filter that distinguishes between 100% OSS and OSS with enterprise price packages.

Overall, unfortunately, many important solutions are lost.

There is a fundamental lack of innovation. Sovereignty offers more possibilities than just using the widely known open source. Where are the specifications and ideas for confidential computing? Why are there no requirements for WebAssembly? Where are the calls to action for profound technologies that rely on eBPF?

At the Handelsblatt TECH conference, Karsten Wildberger called for sovereignty and innovation. I miss that here.

Criteria and maturity levels

It will be interesting to evaluate the individual solutions that appear on the map. This may be one of the reasons why people don’t start with hundreds of OSS projects. As I said, I would first make a fundamental distinction here: How much OSS is it? Are there any licenses that are no-gos? Is there a paywall? Is there an enterprise license?

(I would like to say here that I have nothing against paid OSS enterprise solutions. Most companies are often on their own, without millions in funding to copy a CNCF landscape.

The stack offers the following criteria:

• Digital sovereignty

• Interoperability

• Future viability

• Market relevance

• Trustworthiness

• Sustainability

And an evaluation could then look something like this.

The evaluation is always carried out in 5 stages, which are defined differently.

However, some criteria and their stage definitions are more than questionable.

Under digital sovereignty, for example, there is “Stage 3 (up to 60%): In addition to Stage 2, design capability is evaluated, i.e., the possibilities for using one’s own skills and activities in the design process.” Can someone explain this to me? If I understand correctly, it’s about using one’s own employees/time/money to “design” OSS, i.e., to further develop it? -> If it’s OSS, that’s always possible. But not every change makes sense.

Also good: “Level 5 (up to 100%): In addition to Level 4, full sovereignty is assessed, i.e., the possibilities for full influence.” -> If someone can exert FULL INFLUENCE, for example, then it is not sovereign.

Sustainability is another interesting point. The following two levels only have a remote connection to sustainability. “In addition to Level 2, research investment is evaluated, i.e., the amount of investment in research and innovation or contributors.

Level 4 (up to 80%): In addition to Level 3, StartUp is evaluated, i.e., the number of existing startups.

I understand where this comes from. But the OSS landscape is rarely fed by research, nor by startups. In some cases, a large number of companies involved in an OSS product can be harmful, causing forks and variations that are no longer viable for the future.

The part about trustworthiness is understandable. But it also shows once again that someone has not understood how OSS works. Let’s take the Linux Foundation and CNCF as examples. They invest millions of dollars in improving and qualifying the security aspects of OSS solutions. Conversely, this means that Germany must create a similarly uncomplicated offering. Otherwise, this combination of wishful thinking will continue to fail.

Open source is all well and good, but in the end, it still costs money to develop. No matter how much we hope that someone will create tools out of goodwill and interest.

OSS control

In my opinion, one thing is missing here: an OSS project should have appropriate control mechanisms, and yes, the word sounds really bad in German. So, something like a Technical Oversight Committee or a Steering Committee, for example, with clear principles, terms of office for chairpersons, etc. This can be used to control the influence of individual parties. This is probably not familiar in Berlin.

However, this is one of the most important elements for controlling the sustainability, future viability, and influence of OSS.

There is also this crazy discussion about “US/China OSS vs. EU OSS.” No, there is only OSS. How much influence each party has depends on how many companies invest in it. In fact, very few companies invest time, money, and personnel in open source. That’s where we need to start. Not by developing something else that no one uses. (Let’s take a look at the K8s contributor list https://k8s.devstats.cncf.io/d/9/companies-table, for example. How many German companies are in the top 100? Just 4-5.)

Chisel procedure and next steps

Personally, I think the open procedure is very good. However, it is unclear what further steps there will be. There is talk of workshops with the participation of associations and stakeholders. That certainly makes sense, but I would advise caution. Some, let’s call them alliances, only have their own solutions in mind, which are often not good. And many associations are simply 10 years behind. Industry associations also do not usually produce good content.

International experts, i.e., real experts, not those who sit on the Digital Council or its successor, the German Digital Strategy Advisory Council, would be the minimum requirement for guidance.

Because in the end, we are lagging behind technologically. And that is a social problem, not a technological one.

What would I change?

I would change the approach. We keep talking about stacks, but they don’t usually help. Why? They are mostly complex, there are 2-3 answers to every problem, and in reality there are not enough IT experts who can deal with them (just as there are not enough AI experts...). Stacks repeat the mistakes of the past. The problems that need to be solved are cultural, in the processes, and in people’s minds.

Therefore, my line of thinking must naturally be platforms. Cloud native platforms, internal development platforms, whatever you want to call them. The point with platforms is that I start at the top with the user and their problems, not at the bottom in the basement.

When implemented correctly, platforms as a product, not as a project, can be adapted to the situation. For AI, for edge, for distributed. Basically, it doesn’t matter what application a platform needs to be adapted for. It’s not just about providing some computing capacity, but about a solution that organizations want to use.

Find my original in German on LinkedIn: https://www.linkedin.com/pulse/ein-erster-kommentar-zum-deutschland-stack-max-k%C3%B6rb%C3%A4cher-zjbff/?trackingId=%2BtAljn3aTzGT0Q1g3q9eHA%3D%3D

Platform Engineering, AIDPs and the future of AI Engineering

The future of AI is unpredictable. While celebrated as the next industrial revolution, the reality check shows that most AI (primarily GenAI) introductions are failing, for now. As companies are hunting for the sexy use cases, the successful projects are telling a different story: replacing and automating the boring back-office tasks is the way to win.

And while we are waiting for the next awesome LLM taking over LinkedIn and YouTube, before we will forget about it within two weeks, this trend pushes us to new limits. Systems are reaching new levels of complexity and scale, and require running applications in practically auto mode.

This is a good point to talk about a recent observation, paired with the actual statements of the big AI companies: It will become harder and harder to improve the performance of an AI model. LLMs are reaching their limits, tied up in a shortage of quantitative and qualitative data, computing power, and the current state of technology. As Marjorie Janiewicz from Mistral says: you get more out of your LLMs through the systems and ecosystem it runs in, than to wait for the next 1% improvement by training. Aleph Alpha directly shows this by becoming more and more a platform provider rather than having its own good models. And surprisingly, Thinking Machine Labs, the new company founded by the former CTO of OpenAI, Mira Murati, has launched a product called Tinker that provides an API for fine-tuning open models.

Since I have heard this once, I also don’t get tired of repeating that: It’s not about the model, it is about the platform. The next best model might perform 1-2% better in the evaluation. However, to make it useful and valuable for you and your company, everything depends on the platform —the system it runs on.

An AIDP, an Internal Development Platform for AI Engineering, is the way to go for building environments that can meet any business demand, while providing a solid tooling to avoid high cognitive load from the very rare individuals who are actual AI experts.

Entrepreneurship & Pretzels

The Bits & Pretzels is for me a yearly reminder about the number of young people working on their ideas, products and dreams. It is an engaging atmosphere that calls for action, also thanks to the MCs who are moderating the stages.

I liked the insights and learning more about their history with stable diffusion, particularly from German pioneers like Black Forest Labs. While the LLM companies are close to hitting the end of available data, BFL has a practically unlimited source of relevant data. Images and videos come every time with new perspectives, focus, and details, even if it is a hundred or a thousand times the exact same animal. Paired with their research, they are just standing at the beginning of their journey.

With Mistral and Hugging Face, two industry leaders shared their perspectives on the market and their current journeys. Again, it’s the platform, not the model ;)

However, I also sometimes just wonder why. On the stages, the big “startups” of the continent, in the hallways 40 ways to get some money, and in the caves new startups presenting their products. For me missing are the end users, the people who use those products. Why? A startup without users typically doesn’t receive funding.

On the other hand, nothing new to learn from the big startups; those are enterprises with millions and millions in debt. And the same questions from the moderator always lead to very similar answers: What does AI do to your business? How do you see AI? What are your challenges with AI? …

I think the bits need to change the format. It’s just about celebrating yourself.

Sovereign Cloud Community

I can’t stop making advertisements for our sovereign cloud community. A regional yet international community for sovereignty in data, cloud technology and AI. We just planned the next session until the end of the year, and we are lucky to have found hosts for Poland and France.

We aim to provide a place where we can have a valuable exchange about sovereignty, without the political or commercial bullshit. What works, what's your experience, what are your learnings? That’s what we want to talk about. Join us at https://www.sovereigncloud.community. We now have a Zulip server, and our latest videos are up on YouTube.

Sources of the week

• Stanford Free Online Courses - Overview of Transformers

• From IDP to AIDP: Evolving Your Platform for the Machine Learning Age - my talk from Container Days

• How AI is changing open source development

Tools to check out

• TmuxAI - AI-Powered, Non-Intrusive Terminal Assistant

• Mammouth AI - The “all AI Models in One” solution

• Presidio Anonymizer - Python based module for anonymizing detected PII text entities with desired values

We see a change in the mindset of the people.

So people expect to get a comparably high salary for comparably less work.

They expect to be promoted faster to senior positions, even though they most often are not even really an expert in one of the subjects they are working in.

And since COVID, the number of sick leaves has also risen drastically in recent years.

We have a very interesting work situation ahead of us.

AI Engineering and the impact of AI on IT

The IT world is undergoing a big change. When we start looking into how the future with AI can look like, we already can see that it provides impact on our current market and our current environment.

Now, the big question is, into what will this all develop? And what will these roles look like in the future?

In the IT space, we frequently hear about AI engineering, and the term is gaining traction for several reasons. It also makes sense to create this bracket (AI Engineer) around the job that makes GenAI usable for companies.

But on the other hand, there are many specific aspects that you cannot put into a single role:

Imagine you have someone who is an expert in training models and who is an expert in building infrastructure and who is an expert in building a product, creating a piece of software that goes in front of a user and operates them while maintaining safety and security, well, that is unlikely.

Does every IT person need to be an AI engineer?

I don’t think so. Who else is going to keep the banking systems and legacy environments running?

But many have the chance to become an AI-X-Engineer and find a role or define a role that suits their personal interest in the AI age, in the AI era.

Understanding AI engineering roles and capabilities

Again, the role of an AI Engineer is not completely nailed in stone, but in short, it is to leverage GenAI/LLMs to integrate into products, services, and workplaces.

This can cover many disciplines, including prompt engineering, evaluations (evals), RAGing, fine-tuning, providing inferencing, data engineering, and developing middleware to use managed models as a service. (I can recommend here the book of Chip Huyen - AI Engineering, which covers this very well, OR Anthropics Academy course on “Claude with the Anthropic API”, also not called AI Engineering, it covers practically all of its aspects).

Fine, but what is the AIX Engineer?

Well, think about it in a vertical and a horizontal domain.

The AI Engineer role is currently putting a very wide and huge bracket around multiple roles. If you're looking up job definitions, they are everything and nothing.

But we can deconstruct them into specific roles.

Vertical roles would be, for example:

• AI software engineer

• AI testing

• AI platform engineer

• AI infrastructure engineer (integrating hardware, networks, tune performance)

• AI cloud engineer (utilizing cloud services)

• AI ops engineer

Those roles can be persistent for themselves and can be a standalone proficiency.

In addition to that, we have horizontal AI engineering roles, like for security and data protection, safety, observability, or cross-cutting jobs like AI engineering manager, AI product manager, or AI user experience designer.

I think we will see a mix of AIX Engineers and AI Engineers.

Due to a huge skill gap, lack of budget and time to educate people, the first will be easier to achieve. However, the AI Engineer is, as we call it in German, a jack of all trades. But very rare.

AI's impact on junior positions and organizational readiness

Many discussions are currently on AI challenging junior positions. But without juniors, who is going to do the work in one generation?

So, you have to educate juniors, and for sure, their job will look different; they will do it differently, and they will approach it differently. They will become AI-native.

I need to say, this discussion has taken a wrong turn in some way.

Yes, junior positions might get other responsibilities.

They need to learn and use and have AI on hand 24/7, but I think that's nothing that has changed so far from a digital-native with all-time access to the internet. You just don’t have to search; you ask for answers.

I think, actually, the question is more on the employer side.

Are you ready for that generation? Are you ready to have and provide the right AI tools for your juniors so that they can perform at their best?

Or are you the blocker? Are you the problem maybe in the hiring process and the job and the execution of the job?

That's a play we have to consider that goes in two directions.

Interestingly, we often discuss how we're short on people entering the market to fill current roles. So maybe it's not a problem that we have fewer jobs available on the market. And that the people who are going to take up jobs have maybe reconsidered the way they are doing it.

As AI improves at handling routine tasks, people may have more time to focus on other aspects of their job. But let's be realistic.

We see numerous useless office jobs that involve little more than pushing paper, numbers, and PowerPoints from left to right.

Those are things you can establish autonomy around, and with enough time and experience, you can also build an agentic network around this. It's not a question if it makes sense. It makes sense, but the real question is how organizations are going to adapt it and whether you can adapt to it too.

And are you willing to pay for the tech and the knowledge needed?

Nevertheless, I think we still haven’t seen more than the peak of the challenges that are awaiting us.

In my opinion, AI is a tool for experts. A fool with a tool is still a fool, right?

Doesn't matter how good or bad an AI becomes. It’s the person in front of the screen, or microphone, that uses it.

Will all the AIs we're going to work with be general-purpose and generally intelligent about everything? Or are they subject matter experts?

Having a tech team of subject matter experts would likely make us better at everything we do. Maybe even faster. But it's not that they are doing everything for us.

EXCEPT we are talking AGI or even ASI, right, right???

Ok, first let me give you some more context.

Current Context and Challenges

Over 80% of Europe's digital technologies are imported, creating systemic vulnerabilities and several critical dependencies:

• Cloud Infrastructure: Three U.S.-based services – AWS, Azure, and Google Cloud – dominate almost 70% of the European cloud computing market, while Europe's largest cloud provider holds a mere 2% share

• Semiconductors: Europe consumes about 20% of the world's microchips but manufactures only 9%

• AI Development: Since 2017, 70% of foundational AI models have been developed in the United States and 15% in China - in Europe, well nothing really relevant, except BFL with Flux

• Operating Systems: Microsoft Windows powers more than 70% of desktops, while Google's Android commands 72% of Europe's mobile market; the only hope „Linux“ is open source, yet the development of its derivates are US led

What‘s the problem with that?

Well we can look at it from different perspectives. Political, sociotechnical, economical etc. All of them can come up with shared issue: Europe sucks in its digital competency.

(And yes we have a few cool products - Spotify, Lovable…)

EuroStacks Vision

The initiative proposes building a "common digital stack" - an integrated ecosystem of technologies organized in interconnected layers. And yes, that‘s nothing new so far:

• Resources: Critical raw materials, energy, and water

• Chips: Semiconductors and processors

• Networks: Digital and physical connectivity infrastructure

• Connected Devices & IoT: Smart devices and sensors

• Cloud Infrastructure: Data storage and computational power

• Software: Operating systems, applications, and platforms

• Data and AI: Processing capabilities, AI models and intelligence systems

To make this vision come true, the EuroStack introduces seven core components:

1. EuroChips: Focus on advanced semiconductor manufacturing, particularly through open-standard RISC-V architecture

2. EuroConnect: Pan-European network infrastructure for 5G/6G

3. SovereignCloud: Decentralized, GDPR-compliant cloud services

4. SmartEurope IoT: Interoperable IoT platform for smart cities and Industry 5.0

5. DataCommons: Federated data exchange platforms

6. EuroOS: Including the Digital Identity Wallet and Digital Euro

7. SovereignAI: AI-as-a-service infrastructure

As it’s slightly complicated to remember all this, just think of it like: we do everything by ourself.

But how?

In a world where the best AI engineers gets a 100m$ signup bonus, Europes money spending approach will not any price. (And yes, that payments are insanely stupid, but that’s how it is)

A Disillusioned Implementation Strategy

To get the things rolling the consortium defines a different initiatives.

With the creation of a European Sovereign Tech Fund with initial €10 billion investment, and which should be increased to €300 billion over the next decade, the Tech Fund should invest heavily in the right ideas.

Now, money isn’t everything, you need ideas. The EuroStack Challenge should run competitions to develop Minimum Viable Products (MVPs). The idea of fast prototyping isn’t that bad. But I can see already the biased approach prioritizing settled enterprises rather than smart people with maybe even no company.

Besides, they will start with a phased roadmap starting immediately with mature technologies. So, let‘s start with the old but gold stuff.

Last but not least, the paper proposes to set up the governance structure model after successful EU institutions… I just don‘t know which successful EU institution they see. The latest prominent EU institution defines Microsoft products as risk-free in regard to GDPR, just a few weeks after the French legal representative of Microsoft said they can‘t assure that confidential data is going to the US…

To guide this strategy the paper provides further principles and strategic approaches. I don’t want to go through them here in detail. They are obvious and decently ok.

However, three things are interesting though:

The paper proposes Data as a Common Good and that it should be treated as a shared resource. I have no idea how it should work together with all the “high” data protection standards.

Furthermore, there is a principle about Decentralized Infrastructure - combining edge and centralized computing. This decentralization, edge continuum theme has somehow become the cornerstone for every European initiative related to cloud computing.

The last one is the strong focus on open source. I’m happy to hear that, but the problem is not the usage, it is the investment and the development. In fact, most EU companies use OSS massively but don’t spend a penny on it. Maybe we should introduce an OSS tax…

Are there Innovative Aspects?

No.

There are many thoughts around openness, the different layers, treating such tech more as an open good rather than a commercial asset.

But frankly speaking, that just lets us use the €10b and invest it into ICP. That practically does all this…

Ok, you see I’m slightly critical here so let me take some more points.

Downsides and Critical Issues

Where to start?

The timeline is unrealistic. Also, 10 years is looking realistic, it is way too long a road. This timeline, in my opinion, stems from the tone you sometimes read that we should build competitive alternatives. Also, the report acknowledges Europe's late start but heavily underestimates catch-up challenges.

We also have cultural challenges. Work your ass off is unfortunately not the modus operandi people are looking for. This leads to a drastic slowdown, and already there is a talent gap. Look, working your ass off is for sure not the best thing, but even working hard, the hours you work have become a rarity.

(If you have a full-time employee in Germany, the standard is a 5-day work week with 36-40 hours; they have, on average, 30 days off, around 10 public holidays, and since COVID, the average is 19 days sick per year. In other words, an FTE just works 3/4 of the year.) - Again, that’s in some way good, but we need to respect that in such plans.

Besides, other countries like the US or China, have been investing for many years, 10x the amount of money in the industry than Europe. And those are two countries, Europe has 27…

Now, the biggest issue I would call structural concerns, which are:

1. Over-reliance on Public Sector: The initiative heavily depends on public procurement and funding. But a greater emphasis on creating genuine market demand and private sector leadership could improve sustainability.

2. Insufficient Focus on User Experience: The report focuses heavily on infrastructure but less on creating compelling user-facing products that can compete with existing solutions. This doesn’t sound so important, but people only use what they like to use. If you have a bad time using those stacks, you won’t use them, and with that, there will be no adoption.

3. Underestimating Network Effects: The power of existing platforms comes not just from technology but from network effects and ecosystems that took decades to build. We have strong networks in Europe, but again we have to respect 27 countries, at least.

How to Improve This Paper?

Rather than abandoning the vision, we need strategic pragmatism:

1. Pick Our Battles: We can't “win” everywhere, but we have to gain competence. Focusing on areas where Europe has genuine advantages - privacy tech, industrial IoT, sustainable computing, could be an approach, but this isn’t a sellable asset.

2. Search for Innovation: Also, it is anticipated that with the MVP procedure, I think we can search and find innovative space faster without the MVP procedure. IT is done the same way the last 40 years, time to change that.

3. Embrace Strategic Partnerships: Complete autonomy is neither feasible nor desirable. Define what's truly critical and partner wisely elsewhere (South Korea, Japan, South America, …).

4. Fix the Fundamentals: Before building new platforms, address why our startups sell to American buyers and why our talent moves to Silicon Valley.

5. Learn from Success: Estonia's digital government and Finland's education system show European digital excellence is possible - let's scale what works.

Are there Alternative Approaches?

The report sounds very confident, almost like a “that’s the plan, there is no other way”. Well, it’s an idea. So I searched for alternatives and comments on the approach and have found a few proposals/alternatives. In Italic, my thoughts.

1. Selective Sovereignty: Focus on truly critical infrastructure (payment systems, identity, core government services) while partnering strategically elsewhere. I think that could be a strong starting point. Identity solutions and core government services can be based on Ukraine's DIA app, for example.

2. Regulatory Innovation: Use Europe's regulatory strength more strategically - not just to constrain but to create new markets (e.g., mandating interoperability standards that level playing fields). Can we please stop with that bullshit that regulatory is a sellable innovative asset? No one wants to buy rules.

3. Bottom-up Innovation: Support grassroots open-source communities and startups rather than top-down industrial champions. Yes, yes, but it also needs the investment of large private enterprises that massively use open source but usually don’t spend on it, or at the wrong places.

4. Regional Specialization: Allow different EU regions to specialize in different stack layers based on existing strengths. I've seen this proposal a few times, but it lacks details on what it could look like.

The Bottom Line

The EuroStack represents necessary ambition, but success requires more than funding and regulation. We need to create products people want to use, ecosystems developers want to join, and companies investors want to fund.

Europe's digital future shouldn't be about recreating Silicon Valley but creating something uniquely European - combining our strengths in privacy, sustainability, and social responsibility, and finding our past in being innovative.

The question isn't whether we need digital sovereignty - we do. The question is whether the EuroStack's top-down, infrastructure-first approach is the right path, or whether we need a more agile, market-driven strategy that plays to our strengths while acknowledging our limitations.

What do you think? Is the EuroStack the bold vision Europe needs, or should we pursue a different path to digital independence?

The European Tech Paradox

We have in Europe 27 something markets, almost every country has its own language, regulatory frameworks, and as I believe, we have to acknowledge and understand the different cultures too. This often leaves companies struggling, as they are challenged by this fragmentation and the varying cultures. A few weeks back, I read a post from a German startup that stated, “We launched our product in the UK, and it allowed us to be successful. If we had started in Germany, we most likely wouldn’t make it.” In the same thread, another founder wrote that he feels the pain, but needs to report that after they entered the German market, their product has proven its reliability, and the changes they had to make organizationally to become solid. I think this accurately describes the cultural aspect.

And while we have many very great startups and companies, we have lost for most of them the first place in the race. We have become good, to be the second or third.

This fragmentation and being nutorically late to the party isn’t only a bureaucratic inconvenience. Also, it would be easier with less paperwork, and time wasted for paying random bills to local authorities.

The question also isn’t whether this fragmentation really exists, but whether we as Europeans can overcome this, transforming this apparent weakness into a competitive advantage.

Fragmentation Reality

France is accelerating in digital sovereignty, while Estonia has become a digital-first nation, Germany, as usual, focuses on Industry 4.0 (still), and in the Nordics, we have our fintech and gaming champs.

This diversity creates multiple centers of excellence, including significant barriers. Building a startups in Barcelona will face different privacy regulations than in Berlin, different tax structures, and different language requirements. In the US, a company scales across the continent with more or less the same conditions. Sure, there are some local adjustments, but the frame is the same.

And this we can see in the numbers. Andrew MacAfee from MIT published a simple chart: The US has 682 Unicorn companies, while Europe just makes it to 112. And, in US 35% of them are not older than 50 years. EU? Only 12%.

We are clearly missing something…

In this discussion, and I have to admit I had the same position, it often came up that US has a very bad money spending habit, coming with huge debts and high risk. While this is true to some parts, here is a thing:

• US external debts equal the external debts of the first 6 EU countries with the highest debts.

The other argument is that Venture Capital in the US is more effective and allocates more funds. This gap is killing us:

• Europe spends via VC only 20% in comparison to the VC in US

And yes, we have a gap in the funding of R&D. Andrew has further numbers from CB Insights, which show that if US spends 100 “Units” for R&D, Europe “just” spend 88 units. (They use units because people have no understanding for money). The fact is, even though there is a gap, it is smaller than the outcome (unicorn companies). If this is broken down to a per capita number, EU even spends more on it.

Sorry for the short, hipster startup VC excursion.

Is there any strength in fragmentation?

I think this fragmentation puts us in a neutral position, like in a car. You can push the gas paddle; it roars, but nothing moves. However, if you step back and push it, it will start rolling. It doesn’t give us an advantage, but it also doesn’t hold us back.

Some leading points I find in this discussions are:

• Regulatory Innovation: Different countries are making market access for different technologies easier. Crypto → Switzerland, Digital Products → Estonia, International Trade → Netherlands. However, I wouldn’t call it regulatory innovation. Looking at many EU acts they increase the level of burden for companies more.

• Resiliency through diversity: Due to that fragmentation, if one economy struggles, one market has an issue, others can compensate. EU doesn’t depend on a single tech hub.

• Privacy and values leadership: An argument that I found very often is, that Europe has created a leadership in privacy. We have put privacy so high that it hurts us more than it brings. It actively prevents digitalization and innovation. And no, privacy is not innovative. You get robbed on the streets? Sorry, no cameras.

• Specialized Excellence: We try not to compete with Silicon Valley, or do we? London is a fintech hub, Stockholm Gaming, Munich Automotive…But somehow also everyone else makes in fintech, France and Germany are in the Space Race, every other EU automotive competitor gets bought by the big fish, every country has its own telco ecosystem, etc.

I would say, companies have learned to live with the circumstances. Companies find strength somewhere else.

Sovereignty Imperative

Unfortunately, the geopolitical tensions increase the demand for digital sovereignty. For the last decade, we have built on partners, who have become unreliable. And this is great, because we finally got some mind back for our own strength.

But European sovereignty can’t simply mean building a European version of an American or Chinese platform. Because of the fragmentation challenge we have, this means that European sovereignty must be architected differently, new, innovative. It must work with, rather than against, our inherent diversity.

Some companies are successfully doing that. Spotify not only revolutionized the music industry, but it also successfully navigated complex licensing in Europe and globally. Adyen solved the payment fragmentation. Europe’s Fintech industry is not strong, it’s fragmented, but the apps are lean, modern, and have useful features. It even caused so much pressure that traditional banks had to follow.

Becoming successful in the European market means securing a spot. Because it's challenging to enter our markets from the outside with a new product due to fragmentation, it serves as a form of protection.

Additionally, this represents a distinct segment in itself, overcoming fragmentation.

For me, sovereignty doesn’t mean “oh they are all bad”. It means for me “hey, we can do that too”. We have to embrace the diversity of the European countries, and we have to help each other. Making the German stiffness a little more risk-affine, bringing some structures to chaotic countries, implementing social systems as in the Nordics, and in the end, standing in for everyone.

Looking Forward

Europe's tech future depends on embracing fragmentation as a feature, not a bug. The continent's diversity creates challenges, but also opportunities for companies that can navigate complexity, respect local preferences, and build bridges between different markets and cultures.

The fragmentation challenge is simultaneously Europe's greatest obstacle and its most distinctive asset. The companies and countries that learn to harness this complexity—rather than fight against it—will likely define Europe's role in the global tech ecosystem.

In an increasingly multipolar world, Europe's fragmented approach to technology may prove to be perfectly suited to a future where diversity and resilience matter more than simple scale. The question is whether European entrepreneurs, policymakers, and investors can execute on this vision before others capture the opportunities that fragmentation creates.

We should change our mindset, from the hunt for an infinitely growing company, to the hunt for an infinitely healthy company. The first requires growing markets and population, the second is that we are just happy with what we have at some point.

A “perfect” platform engineer needs to combine 3 qualities:

• 40% on being an expert on the tools and technologies that are playing together to become a platform

• 40% on managing developers and operations demands, understanding their needs and applying that feedback into your platform product

• 20% on whatever is relevant to your organization

The last 20% can sometimes be even more important: Consider the situation where you work at a financial institute. You might focus on a heavily isolated and secured environment. As a contrast, you are working for a game company, latency and availability are now your key drivers. Any of those aspects will influence drastically how you gonna build platforms.

Tools & Tech

I think a good guide to the tools and technologies that you should master is the platformengineering.org landscape. However, it provides tons of options that you might not need. So I will give you a very opinionated view on how I would do it, if I would need to start by 0 again.

The following order of the segments are also the order I would learn. Recommended certifications are just like checkmarks and might help for your future employment.

Resource Plane

I would start with a cloud provider, ideally AWS or Azure, yet I would go with AWS. The benefit is that you can get started easily, there are plenty of free materials (e.g. FreeCodeCamp) around it, and once the cloud concept is understood, you can apply it to any other provider too.

For a platform it will become very important to have a deep and fundamental understanding of networking, storage, and computing services, especially for Kubernetes.

In addition, you should spend a good amount of time to learn and understanding container technology. You will meet this everywhere: in the resource plane, CI/CD, it’s important to understand for security and observability.

But what about VMs and Bare-metal?

Honestly, those are harder to do and to learn. Generally, I’m not a huge fan of VMs; to me, they don’t make sense to have. There is at least one critical CVE for hypervisors every year, and I believe thanks to closed source (ya ya except OpenStack…) there are many 0-days we don’t even know. Yet, there are many cases where containers are also not the right thing to do.

If you have the budget and time to go down to metal, you should do, it is worth it, but if not, don’t hunt for it.

So, somewhere between the resource plane and the integration & delivery plane, we also have to place Infrastructure as Code (IaC). Technically, it lives in the Integration & Delivery Plane, practically it handles the resources…

My take: go with Terraform (or OpenTofu), it is most widely used and can be applied on any provider. Some cloud provider-specific implementations might come with some benefits, but often lack wider understanding.

Qualification track:

1. AWS Certified Solutions Architect Associate

2. AWS Certified SysOps Admin Associate

3. AWS Certified DevOps Engineer Professional

4. Certified Kubernetes Administrator (CKA)

5. Terraform Associate

6. (optional, would make you stand out a little more) Terraform Professional

Unfortunately, the AWS certifications to qualify you a little for deep platform knowledge. Therefore, you absolutely should spend time on the AWS EKS Workshop.

Integration & Delivery Plane

In Platform Engineering, we might have overcomplicated the automation part. It is the core of every platform and the most relevant part to learn and understand.

For the basics, start with GitHub and GitHub Actions, it is widely used in enterprises. Microsoft provides a good learning path.

I wouldn’t worry too much about container registries at that point. You can partially cover it by digging into AWS ECR as part of your resource plane journey. But there are also more powerful solutions like Nexus and Harbor. But I wouldn’t waste time here. More important are the concepts of GitOps.

For GitOps I’m slightly opinionated: use ArgoCD. Also, there is Flux, and thanks to good marketing, it continuously pops up and is simpler than Argo; I didn’t find it that often in the wild. ArgoCD has many getting-started guides. But, handle it with care. To get started ArgoCD is enough. In your later career Argo Rollout, Events and Workflow might become interesting, too. Important to understand is that you also need to understand the concept in depth.

I think neither Platform Orchestrator nor Infrastructure Control Planes are very much needed at the beginning of a Platform journey. BUT, if I could spend money on one, I would opt in for Infrastructure Control Plane. Most companies don’t use it and have pure chaos, drifts and outdated infra because they don’t want to spend money on it.

Qualification track:

1. GitHub Actions Certification

2. Certified Argo Project Associate

3. (just if you want to burn the money) Certified GitOps Associate

At this point, I’m always unsure about digging into the topic of what is not shown in the reference, which is what you should add to your CI/CD and GitOps pipelines.

What do I talk about?

• Security scans e.g. before/after container builds

• SBOMs

• or infrastructure cost estimations

Monitoring & Logging Plane (Observability)

The observability stack can be a fast one, but you can be also lost in it. Defacto standard is today to apply Open Telemetry (OTEL) and use it with an open source stack like Grafana & Prometheus.

Grafana Labs provides a couple of hands-on workshops where you can get started with some ideas of what to do. Now, observability is easy to start with, but in platforms at scale you will have to master it all of its complexities.

Qualification track:

1. OpenTelemetry Certified Associate (OTCA)

2. Prometheus Certified Associate (PCA)

Security Plane

As already said before, security will find place on many spots. How to configure your resources correctly and securely? Where to apply security scans? How to manage the access? Those elements are scattered and integrated in every domain.

Let’s start by an important one, secrets and certificate management. Also, technically it’s two things, we see them play along together frequently. To handle secrets we have to take a look at vault or key management solutions. The AWS KMS you might find on your way over at your cloud journey. In addition, Hashicorp Vault has become widely used, also due to license changes, the adoption might have dropped. A pure open source alternative, OpenBao is in development.

At this point, we will hit a problem of the reference architecture. It shows the foundation of the platform, but it doesn’t show the inner guts of a platform. What do I have to install on Kubernetes?

Policy engines and management like Kyverno or OPA work best from within the platform. Network encryption is ensured by the resources you use and from within the platform via mTLS configured on the container network interface. Last, we talked about security scans in the CI/CD; best practices would be to go further ahead and bring it into the platform as something called admission controller, that continuously scanns the system.

And we didn’t touch the basics about Role-Based Access-Control (RBAC).

In short, security is everywhere, in many layers, and you, as a soon-to-be Platform Engineer, have to know it too!

Qualification track:

• Vault Associate

• (optional, would make you stand out a little more) Vault Professional

• Certified Kubernetes Security Specialist (CKS)

• Maybe a CompTIA Security+ (very generic) or a cloud provider-specific security certifications (very specific) comes on handy too

Developer Control Plane

The last category is the Developer Control Plane. Many people who are looking at Platform Engineering and Internal Development Platforms (IDP) are understanding this as a Developer Portal, similar to Backstage.

The most successful platforms I have met, don’t use Backstage. And many teams also hate the solution itself. In my opinion, there are many misunderstandings. Backstage is very good in showing you what you already have and can, in a single place. But what you don’t have, you can’t show off with.

Until this point, if you have followed along the track, you should have mastered version control in the one or the other way.

I usually see at this point in companies that there are two paths taken:

• Providing unified development environments

• Start working with a Developer Portal

Both ways are a good choice, but they massively depend on the corporate problems you want to solve. (That’s why the next chapter is so important).

Qualification track:

• Certified Kubernetes Application Developer (CKAD)

• Certified Backstage Associate (CBA)

(The missing child) - The Capability Plane

For me, there is one part missing in the reference and learning path, and that’s what is on the platform. I call it the Capability Plane. Because many things that we are implementing for the platform, are around and under the platform. Those have to be made available in the platform too. Those are reflected as features or capabilities of the platform provided to the platform user:

As you can see we can find again many pieces that we talked before about in security, networking or observability.

There are some aspects that are important to delve down on it:

Scale and Scheduling are core features of Kubernetes, but they need to be fine-tuned to your environments. On a cloud provider, a Karpenter comes in handy for optimizing the infrastructure. An application or serverless-aware framework might be useful, too. And a batch system like Volcano is in times of massive data and AI also very much needed.

All those need also the right integration of resources, and especially a good management and access control.

For me personally, the capability plane is what brings the value to the user. Because a platform consist of two layers:

1. the infrastructure, automation and integrations that are forming the platform

2. the services and capabilities that make the platform useful for the end user

People, Culture and Processes

A shameless plug at this point for my book Platform Engineering for Architects - Crafting modern platforms as a product. We aimed to write a book that is timeless, focusing on every element of the platform engineering that will ensure its success, but without sticking to a specific technology.

Therefore, I will give here a rundown on the important parts:

• A platform's purpose - don’t build a platform because it’s cool. A platform should fulfill a purpose, helping your organization to become better in its engineering journey and solve problems your developers, operations teams, DevOps etc are facing.

• Defining principles will help you stay on track and not start with “Randomeering”

• Before you target for an whole aircraft carrier size of complexity, start small and simple with the thinnest viable platform (TVP). Develop the one feature that will help one of your end user groups a lot.

• You have to fight Conway's Law. Break it, change it, make it work for you.

• Target Developer Experience and the reduction of Cognitive Load. Today, we hear often “shift left”, I think we shifted to much left, throwing more and more on the plates of Devs. But not alone on their plate, but also other users are facing drastic increase of complexity. For example: I haven’t met in the last 10 years a corporate security that was up to the technology and time we were working with. As a platform engineer, you have to help them.

• Security is everywhere. Many security issues are created by people and processes. It’s often not the technologies fault.

• Learn to manage technical debts. Open source and cloud native are moving fast, and there is no way that you can take a technical decision without causing technical debt.

• You will need to learn how to optimize your platform usage, shrinking down costs, and leverage the utilization, for example, for expensive hardware.

• Build golden (opinionated) paths with escape routes. Your standardization platform 1 (Kubernetes API) and 2 (cloud API) will still keep the user on a controllable field.

Besides all those aspects, as a platform engineer you have to learn to work with people together, get their feedback, be open for their contribution and show them how to be successful. A good success factor is inner open communities. Be welcoming to their contribution and give frequently introduction and updates to what is new.

As a platform engineer, you, your management, and your executives need to learn to understand the value of product thinking. No matter how good a platform is, it dies if it is treated like a project and loses budget, people and priority after the project is done. Internal Development Platforms deliver a lot of value to your organization. But you have to make this transparent. For this, measure different metrics, from DORA and Space to DevX, before the introduction of an IDP. Measure it during and from there on, continuously after the IDP introduction. This will show the continuous value of the system, of your team and your work.

A Platform Engineering is many, it is not a new role, it is a new level

I’m always very proud of my people and team. Those are the best people you can find. But why?

A Platform Engineer is not a new role; it is an entirely new level.

To become a Platform Engineer, you need to master the role of a Cloud Engineer, a DevOps, an Requirements Engineer, and depending on the direction of your platform also of an SRE, Security Engineer and for sure in the end as Solutions Architect.

To be a professional Platform Engineer means you are at least 80% better in every one of those domains than the rest of us.

And yes, that’s maybe a bold statement. But if you don’t understand this statement, you might haven’t met yet a professional platform engineer.

See you out there, cheers!

But the danger is that “AI Engineering” will become a term like DevOps, meaning everything and nothing.

A term where we have included every part of an AI lifecycle, and still, everyone understands something different.

Whatever the way will be, we (all, from IT to Business, from hospitals to logistics) should work on ourselves and our capabilities, leveraging AI. AI is inevitable. On the other hand, I would like to call for more realism. Call you to think twice, or maybe even more often than that, before starting to run into the rabbit hole of AI.

As the latest paper from Apple shows, AI is not there yet that it can “think”. It is very good at mimicking the process of thinking, or as it is called, reasoning. And, yes!, it is often good enough for many of the paper jobs we have to do. Sprinkling some more context, tools and information on top and it can also do way more complex tasks.

I honestly think it just shows with which stupid and useless time-consuming bull horns… we waste our time. People are still afraid of AI taking their jobs. Yet, spending 30 hours a week in meetings, I wouldn’t consider it work. Sorry, corporate live…

But for all physical workers, I see that AI comes with a huge relief. For the same reason why “knowledge” worker… Sorry I just need to laugh about that wording… are afraid of it: taking away the paperwork that no one needs.

Enough of the socio-technical aspects, let’s go into the topic.

Every role needs to find its own way to include AI Engineering

The way how and where we use AI can’t be more different. Some industries and jobs are not affected yet at all. Others are running full headfirst into it, without a single second of reflection.

Prompt Engineering - the minimum effort to get started

I don’t see prompt engineering as a future job or a crazy profession. It might be an interim functionality as long as people didn’t wrap their heads around it. I see prompt engineering similar to professional communication training in a real-life setting.

To interact and get results of an LLM as you want to have it, means you have to learn to talk to it in the right way. (Unfortunately, in most companies, comms training is coming short, too)

For the more techy people among you, Anthropic has a nice interactive tutorial on Prompt Engineering. It’s a year old, but still absolutely relevant. You can find it on GitHub. If you'd like to start a little bit less hacky, there are many good Prompt Engineering starting points: Anthropic, Google, Andrew Ng DeepLearning.ai, and if you want to go 1000% into prompting, learnprompting.org.

The message should be clear.

Everyone should learn and understand the basics of a good prompt. Maybe even forget about the engineering in Prompt Engineering.

As we “googled” the last decades, we now “prompting” it. And you should learn how to do that in the right way.

Do I have to change my IT Engineering profession?

To be honest, who knows? But this doesn’t mean you shouldn’t learn how to include AI in your work (at least prompting), and even better, how your job serves the integration of AI for the company you are working for.

If you are in Software Development

Learn how to integrate LLMs into your software products, which features and functions do the different provider give you through their APIs? How can you turn your software API into an endpoint for an AI via MCP?

MCP, introduced by Anthropic, stands for Model Context Protocol.

You should stay informed and in an frequent exchange with your product and application manager, to discuss what’s possible and what make sense. Don’t let disillusionment cause false hopes. There are enough managers, CxOs, and whatnot out there who are very good at repeating settled business/AI language; usually lacking the understanding of what can be done. Become the voice for that.

If you are in Platform Engineering or DevOps

You should learn the lifecycle of models, AI products and how to best support those developments. This includes observability capabilities that need to be adjusted for AI, new Security approaches, and most importantly, new scaling and scheduling approaches.

I like to differentiate here between AI-enabling and AI-enhancing.

AI-enhancing means you need to learn how to advance existing systems and development flows with AI. Those can be from code generation, review support, documentation to supporting onboarding processes and answering questions on documentations and git repositories.

AI-enabling means providing infrastructure, platforms, and environments that help to provide, train, and run AIs in all their flavors. (A hint, a one-platform-fits-all will not be an option). As a Platform Engineer, I like to think about how we can bring developers and AI researchers closer together? How can we use our decade of experience in automation to automate the AI foundation? And how can we enable people to serve themselves with the tools and access to infrastructure that they need?

I call the solution to this an AiDP (where IDP stands for Internal Development Platform), the AI part you will be able to guess in this name play. Others, like Pini Reznik, call it AI-native. In the end, it moves the two disciplines together. I don’t want to say merge, that might be false, but create a beneficial relationship between both disciplines. Data, ML, can utilize pattern from Dev and Ops, become way more effective.

Looking into the tech stacks of providers for such dedicated AI-enabling platforms like Crusoe, we can see in their high-level description how deeply they utilize Kubernetes and all cloud native elements around it.

As Cloud Engineer

I’m honestly not sure what the best way is here. Learning the different approaches to serving infrastructure and services for AI is for sure the way to go. However, I also feel that the role of “the cloud engineer” will become a commodity for every other role. This is because cloud providers have become “just” yet another service provider that facades tons of default services. What I mean by that is that system components are meanwhile look everywhere the same; whether it is on premise, in the cloud, on Kubernetes or somewhere else virtualized. Also, the applied logic in between is almost the same. What I mean by that is that the cloud itself is not the challenge nor the problem; it is mature enough to be boring.

I think, what we will see are two things:

• A new generation of “cloud” providers, focusing on serving AI, like Crusoe

• A split point of traditional cloud providers, serving in one domain, traditional computing capacities, and providing new dedicated services, not like it is at the moment, but with a full-blown AI platform service

The Infrastructure Engineers

I believe in this time, this job is the real deal. No joke.

Many new initiatives are being launched to build extremely large AI-focused data centers. The tech stacks are focusing on providing the best and fastest access to GPUs, and you're gonna work with all that hot stuff. But I think delving upwards to Kubernetes and then back down to how to provide cloud native GPU integrations is where even more fun is waiting.

But please throw away that legacy virtualization trash.

No one needs this anymore.

Which AI Engineer will you become?

So far, we just talked about the groundwork, and we haven’t even touched models, RAGing, agents, etc. For the most part, I would leave it like this and would recommend you check out sources like SwirlAI from Aurimas. In his latest newsletter, “Breaking into AI Engineering in 2025,” he gives you a nice, clear roadmap.

In the end, how you and where you will let AI take place in your career is up to you. And again, the topic might seem annoying, and almost impossible to keep up with the speed at which it develops, applying it to your niche and profile will make you go with the times. AI reminds me a little bit of Thanos…

More sources

With the focus on our cloud native environments you can find some dedicated talks around serving AI on KubeCon, Cloud Native Days/Summits and KCD.

• KubeCon London 2025

• Data on Kubernetes Day EU 2025

• Kubeflow Summit EU 2025

• Cloud Native & Kubernetes AI Day 2025

• Understanding Platform Engineering → Platform Engineering Day EU 2025

I found an interesting source: the channel of AI Engineer, which are recordings from the same-named conference. If you guys ever read this, can we bring your event to Europe too?

Tell me about your journey

Where are you currently in your engineering journey?

What are other challenges you see and face?

Let me know!

Usually, sustainability posts start with how important it is to be sustainable, how much screwed we are, or that your consumer behavior kills the planet. But in short, it is not your problem. The majestic enterprises who are responsible for it make you believe so. In the end, it might still be about your energy demand, but who am I to judge you?

The momentum of sustainability, which is a very old topic, is gaining speed in IT. Maybe driven by Gen Z, maybe because people wonder if they could do something more valuable than useless. Who knows. For me, it turns out to be the question:

What could I do that utilizes my skills and knowledge to impact this planet positively?

Because picking up trash in the forest or shutting down every unnecessary light in the apart might do its thing, but it doesn’t feel like it can make a difference.

So, what are your skills, knowledge and interests you can utilize to make a positive impact?

This was the birth of the idea to bring sustainability to open source and, especially, to cloud native. Soon, after finding some fellows, we drafted a work group proposal for the Cloud Native Computing Foundation, that eventually, some months later, became a Technical Advisory Group — Environmental Sustainability. That’s where my FOSS sustainability journey all started.

This makes me realize that without FOSS, we will not have any sustainable IT.

The Problem within the Roots of IT

I often have the chance to meet people who hate it if I call their IT traditional, classic, or legacy. “We have to do it this way; we have done this for decades; there is no other way to provide it with that reliability.”

Excuses.

No one wants to invest money to throw out all that old stuff, start from scratch and do it a new way. I don’t want to call it a better way; that would be pretentious. But a change is urgently needed.

You need tons of people to keep the old stuff running; many people don’t want to learn something new, too, and there are too few new people joining the IT market. Through that spiral of death, you drastically have increase the payment. Because people who could do something new are becoming rare.

Gif by BhumiPednekar on Giphy

To acknowledge and adress this problen is crucial to understand that change doesn’t come from just bringing new tools. But that you actively have to seek, argue, fight for getting this things moving.

What is Sustainable IT?

Sustainable IT, also called Green IT or Eco-friendly IT, combines technology with environmental awareness to reduce the impact of information technology on the environment.

While there are many aspects to it, the three pillars are:

Energy Efficiency: Reducing the energy consumption of hardware and software is one of the key goals of Sustainable IT. However, this works best if you have modern, flexible systems. #legacyIsHarmful

E-/Waste Management: This is not only about ensuring that your data center trash is recycled correctly but that you enter a circular economy. Also, good starting points are to use hardware as long as possible, which could be contra-productive to energy efficiency.

Sustainable procurement: It doesn’t help if you want to change the planet. You also have to demand this from your vendors. Commercial solutions generally aren’t interested in becoming sustainable, except it is a KO criteria for not buying them.

Nothing of this works with the right awareness and measurement to implement accountability.

How Open Source Contributes to Sustainability

Open Source is clearly leading that space.

While we are talking, or reading, there are more than two hand full of open source organizations with hundreds of contributors working on specific fields to make IT more green. I summarized recently some of the bigger ones here on the Linux Foundation Europe blog.

But why does FOSS leads this space?

Social contributions. People work on those topics because they love it, because they have the feeling to do something, and because they can really provide some change.

Overall, FOSS has a very large lever: multiplication. Over 90% of all kind of software runs with FOSS (as programing language, including libraries, middleware, runtime environemnts and so on).

Improving only one component just a little bit will be mutliplied, changing many components, you can start feeling the change. On top of that, bring a new mindset, a new principle to design solutions, and even new technologies and all of the sudden you have a meassurable change.

Gif by ctmagazin on Giphy

My favorite part in it is, that sometimes it is enough just to raise the awareness to drive someone else thoughts and innovative thinking to new solutions.

When we started the working group for environmental sustainability a guy showed up at one of these meetings, telling he loved the initiative and proposed to an open soruce tool he contributes at to propose making it carbon aware. It didn’t took that long that the community has developed this feature, extending its capabilities and providing an additional purpose to be carbon aware. If you are interested you can read more about the KEDA carbon aware operator.

You need FOSS for a sustainable IT vision

That’s the reason why crafting a sustainable IT vision requires FOSS. And exactly this seeding, that growth, is what I would like to see in any project.

A vision to become carbon aware, to become carbone efficient and to max out possibilities will be build on solutions that are not there yet, that need your commitment and time. All the given foundations digging into a specific field offer you a wide range of options to start.

The community gives this commitment, and you can build on top of that, or even better, become by yourself a contributor and investor to accelerate the innovation and change, in a sustainable IT.

The Treasure Hund

If you visit

https://terraform.tf

you will be redirected to the OpenTofu website. Just a big love and high five for that fun move, whoever it was ✋

I hope you liked what you read, see you later 👋

Max

p.s. if you want to sign up for the Zeitgeist of Bytes newsletter or share it with a friend or colleague, that’s the way to go:

Share

I had two primary funny discussions the last week. First, that cloud is the only solution we need in the future. I couldn’t hold my laugh. If cloud means matrix and we have a universal distributed cloud that doesn’t know provider-specific software-defined specifications, then yes. Otherwise, hell no.

Second, one was trying to convince me that having ca. seven months of “experience” as a Scrum Master makes one already a veteran in this field. We couldn’t find a joint agreement, but it's clear that with practically no experience, you haven’t even left the diapers stage of your new career.

Why we still do cloud like it’s 1999?

Calm down, I know cloud is not that old. But IT et al, and especially the way how we build, host and maintain software is - and even older. But I wonder why we still run it/IT like decades ago? Observing cloud migration projects failing, which then becomes a cloud transformation project, which fails; and then you establish a Cloud Center of Excellence (CCoE), who should do the cloud work for you. Bad news: in most cases, your organization still fails.

We have an actual shortage of new professionals on the market, while the old generation is looking for retirement and not motivated to learn something new. And many middle-aged people don’t enjoy using the cloud. However, the technologies also don’t get more complex; they just get more. And for each layer, you need another expert. We practically cut down our available pool of experts through the amount of layers we stack together. And yes, I’m with you; if everyone would try to build their own layers, we would even need more people. But that’s not the answer. We need a renaissance of IT throwing all mid layer tech out of the window, ideally together with the organizations saying, “We have done this always like this” and are not open to new concepts.

All credits to Equinix https://deploy.equinix.com/blog/bare-metal-bear-metal/

Web 3.0 (and later Web3) and distributed ledger technology were promising directions. Simple, safe ways to run software. Yes, your 2000’ windows software doesn’t run on it, the mainframe also has hard times to migrate and all that uselessly collected data, the new gold, would be to expensive to store on it. All of it is right, but you are looking at the public parts of that tech. Why not combine them? Why not make your software easier to run? And most likely safer…

A funny anecdote about the safety of your data. Many years ago I had the advisory to a publishing company. They defined their knowledge as a digital asset of high value and must be better accessible, so a cloud project it was. After some time, the company's owner received the list of potential security risks our favorite cloud provider might have, and the project was done… The beauty of family businesses. The reason? He believed that their own data center is better protected and has fewer risks. On the way out, I had to walk across the courtyard and passed by an old little rusty door that was half open. I was curious because, normally, any door on that property was closed, as some parts of the property were publicly accessible. So I took a look and called to see if someone was in. There was no answer, just the loud sounds of hundreds of cooling fences for the servers. Now I know where the data is resting, and now I know that protecting that data is just an old rusty door. I wasn’t even sure it was waterproof.

Back to business. Web 3.0 and its descendants were screwed by scams, its early adoption with a not yet ready-developed solution landscape, and many people who doubt that this tech is serious. ICP, Algorand and co might drive a new renaissance. I hope so.

What do you think? What needs to change?

Why are so many organizations doing Scrum, SAFe and co wrong?

It’s fascinating how wrong you can do a thing when you believe it’s simple and just because you want it you are able to do it. The best agile projects I was in are at least 5 or more years ago. But today, as it has become a commodity, the projects are just degrading in their quality, acceptance of the method and the overall education of the frame you are working in.

Gif by stonehampress on Giphy

Scrum or SAFe is not an organization form; stop it. When your whole company becomes an “agile" powerhouse” that is “dynamically deciding,” you are on the wrong track. But why is it that those methods have become so horribly wrong used?

From my observation, it’s all about experience. People call themself Scrum Master or RTEs after doing training and a certificate. What we are missing is real training and experience sharing. Run first with some projects alongside and learn how to do it right. Then, get your first team and slowly, with time, grow into the role. Some good 2-3 years will get you rolling. That takes to long? That’s how it takes when you want to become professional in something.

Another example is SAFe. SAFe requires work. You are not SAFe if you meet every few months to drink coffee and pep-talk with your team. You need to have an actual Product Owner, like a real one. Not someone who has never managed a “product”. You also need a meeting culture that is not eating your time because you have actual work to do: plan features, define dependencies, align with the business, etc. When you go into PI planning, then this is about explaining what's on the way, why it's needed, and what support you require. But if you start writing epics and features just in the PI and think that something “might be a good idea”, you are too late doing your job.

All of those wrongly done frameworks and career wishes are leading to constantly failing, depressed and demotivated teams. And so the death spiral is on.

My learning of the week

People can’t take criticism and hard talks professionally anymore

This week, I understood that we are getting soft-washed. We have to respect everything; we can’t be too direct, as it might hurt one, and from tons of training, you shape your speech so that it doesn’t offend anyone. Work is not your family. If it is, it means unconditional love, loyalty and no payment. Work is like a professional sports team. Every team member wants to be the best at what they do. Together, they want to be the best team. Their leads are the coaches, and the directors of the club are similiar to the managing directors of a company. And perhaps they all play in the same league with many other teams. This is, how work should be.

If someone tells you your shooting technique is horrible and you're never going to make it, it is not a critique of your feed, your personality, your character, or anything else. It is purely a critique of the way you do the thing. But this is exactly what young people today no longer have. Critiquing your technique is the same “hurtful” as personally attacking one. But where does this all start? In the elementary schools where you don’t get marks, or you don’t win a school cup and everyone gets a medal for participation? Or university where our speech gets so much polished that we express ourselves in complicated not to piss others off? Or maybe in our classic corporate feedback burger training. Please if you have such, change your trainers, that’s outdated!

On the other hand, it is interesting to talk to other leaders who almost all share the same viewpoint. Go back to more constructive feedback: No cushimushi, that's the situation, that's the problem. Here is an idea of how to fix that. As you see, it's not about what we want to improve; it is about how. Super polite, positively packed feedback doesn’t find its way.

Just wanted to say that I dislike footbal, but for some reason it fits well.

A good approach is the CORE method:

• C — Context (What happened in the specific situation?)

• O — Observation (Tell what was done or said.)

• R — Result (What is the result of that action?)

• E — Expected next steps (What are the expected next steps?)

Example: “You were asked to write two simple Terraform modules for our customer project X (context). Both modules that you delivered were not well received by the customer (observation). The team had to spend additional hours improving the code quality and reliability of the modules so that they could pass the tests (result). For the next modules take your time and ensure that they adhere our quality standards. If you need further assistance please ask your colleagues for early feedback or take some time to educate yourself more on it.(expected next steps)”

Now, that’s not precisely “hard” feedback. But even such feedback has become rare because the more often you have to give that, the less well people feel. But its a question of the mindset.

Whats your feedback on feedback?

Gif by hyperrpg on Giphy

The Treasure Hunt

• Platform Engineering is for Everyone: Platform Engineering isn’t only focused on developers, it brings actual business value and a competitive advantage https://thenewstack.io/platform-engineering-is-for-everyone

• Google data centre in Ireland blocked after NGO calls out ‘greenwashing’: https://www.edie.net/google-data-centre-in-ireland-blocked-after-ngo-calls-out-greenwashing

• Public clouds are 'dirty' about VMware's on-prem push, says Broadcom CEO Hock Tan: https://www.theregister.com/2024/08/29/hock_tan_vmware_private_cloud

I hope you liked what you read.

See you later ;)

Max

Are Virtual Machines finally dead yet?

The whole license changes around VMware and the insanely increased costs are like a horrible accident. You know you shouldn't look at it, but you also can't look away. tl;dr After the acquisition of VMware by Broadcom, the customers experienced a drastic price change, making many challenging their position. On the other hand, multiple corporations report that they can't just use something different as they are too dependent on the technology stack. Well, if you just had to invest five more minutes to be a little bit more cloud native, I'm not saying that it is the solution for everything, but at least it decreases the number of chains you are putting on your IT.

We hopefully can move on now to overcome VMs and focus on more lightweight solutions. Better for your pocket money and for the planet. But it doesn't look like people think at least one step ahead - the common discussion out there in the WWW is which hypervisor to choose instead of thinking about how to build and ship your software so that you can use whatever you want...

Thinking of the 1:1 replacements:

• I see here a few chances for cloud providers and their rapid adoption. If AWS, Azure, and Co. are too expansive, you should test Scaleway, OVH, and DigitalOcean! (no advertisement)

• The FOSSlers, on the other hand, are screaming PROXMOX, but I think that is not an option except if you want to double or triple your current infrastructure team.

• OpenShift or OKD would be more elegant than that. Here, you would get a full powerhouse for your transition. Run VMs, run containers, and host a Serverless platform on-prem or in the cloud. You see where this is going.

Whatever option you choose, first, you should invest in the way your teams develop, build, and ship software. As an example you could use Packer to create from the same software VM images or container images. With the focus on the pipeline you are going to be prepared for future changes and enable your organization to take on any challenges without having such minor blockers.

Co-Pilots and why you have to stay the captain

The age of co-pilots is here. But maybe we are just at the level of kinder garden. As BusinessInsider (yeah, I know BI...) reported about a Pharma company throwing out Microsoft Co-Pilot because it is only able to create PowerPoint slides, which also could be from a child. How much meat we have on this stake is difficult to say, but to find relevant success stories is also almost impossible. I think we should turn down the heat of the AI flame to not burn all the fuel at once and see the whole industry going down because, surprisingly, the technology isn't ready yet.

What most people forget is that we are in the middle of a huge, big-fat hype. And the wave we are riding is still strong, constantly pushed by new companies, new products, and billions of dollars invested by people who actually play Monopoly.

Obviously, technology will need some more years to become really helpful. The results of the report from BCG and Harvard are good. "Completed 12% more tasks; completed 25% more quickly; increased performance by 17% with AI".

Looking into most of the tasks like creating lists or defining steps for a process I can say, not every company is just working on power point slides, some have real work to do OR if you use dead simple tasks, you are not going to replace your experts, but your trainees who should become experts in some years.

In the end we have to stay the captain, define clear ways to work with AI support but don't let young people lose track because we can now abstract away basic tasks. And, we will lack creativity. Finding creative and revolutionizing solutions is what brings a company forward. Use the tool but don’t let the tool use you as fool.

our young talents getting carried by co-pilot

Learning of the Week

In the last couple of months, I have spent my time writing a book together with my two co-authors. It is an exciting project and something I have been looking for for many years. It will be released in the next months, the heavy writing part is already done and the final editing and technical review is fully in progress. Our book "Platform Engineering for Architects" tries to balance between the technical standpoint and a more methodological, timeless approach for designing platforms and internal development platforms. I will keep you posted.

However, what I learned this week is about creativity. While writing the book for the first chapters, I felt unstoppable. My thoughts shaped like magic, and the more I wrote, the more thoughts and ideas came up. But also, the more I wrote, the longer I needed to put my thoughts into words. And towards the end of the book, it felt like ages passed by until one page was done. So I did my research on it what others do when they feel stuck with their creativity, and I found many saying they are losing their creativity. But I think that is wrong. The motivation and ideas are still there, they just felt tethered to the ground. Creativity doesn't vanish, she just get to many dependencies, to many other ideas that are connected which keep holding it down. I continuously had questions in my head like - didn't we write about this before? Haven't we taken an opposite statement in chapter x? Is it still the right topic?

So, either one of two things helped me a lot:

• Let it flow: I completly ignored the questions I had and just let the words flowing out. Unbiased it became more what I wanted to say then continuously double check it while writing. A similar effect is given when you do a brainstorming session without rules and sharing thoughts and then consolidate it afterwards. But it is very difficult to ignore those inner voices that continously wants to redirect you.

• Build on ideas of others: Sometimes it just wasn't possible to ignore given content and later clear out your writing and making it fit together. So you have to understand it, study it, and then build on it. Building on someone else thoughts triggers your brain to think more deeply about it and provide actually nice new ways of thinking.

Both don't seem to be crazy solutions. But having a walk outside to get the head free or doing something else and return to writing didn't help out, and maybe decluttered the thoughts for some sentence before the rest of the thoughts jumped in to ring your creative part of the brain down.

The Treasure Hunt

• Black Forest Labs, a new German AI Startup, yes buuh another one, launched their new image-generating model Flux.1. The model comes with a <3 for developers with a clean, clear API and examples of integration. You should check out the demos. I'm sure you will be surprised by the outcome.

• While talking about AI, NVIDIA has released new free training on its website and Coursera, like the "Generative AI Explained". And don't forget about evergreens like the AI Infrastructure and Operations Fundamentals.

• In this, The New Stack podcast, Anne Currie talks about creating green software, energy proportionality, and DevOps.

• The NIST published their recommendation on a Post-Quantum Cryptography, saying FIPS 203, 204 and 205 are solid options. FIPS 203 and 204 contain algorithms from the Cryptographic Suite for Algebraic Lattices initiative. However, the choices are already under criticism by Dan J. Bernstein, a cryptology and open-source legend, pointing out stupid calculation issues. Maybe they need a Quantum computer to do better math ...

Don’t miss the party

Some upcoming events of the next month you shouldn’t miss

• Civo Naviagte EU - 10.-11.09. - talking about Cloud, Kubernetes, AI and Co

• Open Source Summit EU - 16.-18.09. - Open Source, Containers, Digital Trust & Security

• Container Days - 03.-04.09. - Microservices, Cloud Journeys, DevSecOps

I hope you liked what you read.

See you later ;)

Max

Tech YouTube is exploding… Some call it magic, others AI. Now, the Rabbit R1 brings both to your pocket, and it drives techfluencers insane. The intelligent bunny should change your future behavior in interacting with smart devices. What could possibly go wrong when you combine a camera, a screen, and an AI in a fancy 80s-vibe orange box?

Rabbit R1 Product Page - No paid ad here!

I think the people are not ready for it. The fear of missing out and the addictive need for moving pictures will keep people on their phones. Looking into the future, and yes, maybe Star Trek kicks in too much; without augmented implementations (AR), such a device might not catch the people. The first reviews also don’t leave room for hope, calling it slow, with a poor connection, very generic answers, and often just wrong. I found it astonishing that with all that GenAI is going on, you can deliver such poor products. What do you think? Will the rabbit in the pocket be a game-changer?

In signs of sustainability - Green Platform Engineering (!?)

Gif by jonhanlan on Giphy

In the last months, I had the chance to talk more often about how cloud-native platforms, internal developer platforms, et al., are the foundation for a more sustainable IT. The complexity of architecting a system with an environmentally sustainable principle in mind is much more challenging than for static conditions like costs. Also, I see here much traction and correlation; it’s more a coincidence than a lever. In the live discussion with the Platformers Community and The Good Guy I highlighted the element of complexity and abstraction. Platforms will lower that issue, and looking into the future, I believe we will have systems that completely ignore the underlying infrastructure and schedule workload based on our requirements, for example, the CO2e footprint.

At the Platform Engineers Meetup I gave a short summary on that topic.

I’m curious: does this topic play any role in your organization?

Talking about becoming greener, for the last two days, I participated in the Eco-Compute in Munich. Which was a fantastic event, with very different talks and chances to network and exchange ideas. You should keep an eye on it for next year! It’s definitely worth your time and support.

My learning of the week/month

What does it mean to have “a career”? My learning for the last month is that a career doesn’t mean growing within a company or hunting for the next higher title. A career means mastering your profession. And you can do this over and over again. For some reason, we humans are so stupid to believe that anyone cares if we are the super vice president of Yadda. It is good to have ambitions, something that drives us. And we should be compensated for our experience and talents. I had the luck of meeting some fantastic individuals at KubeCon, and even before in some open source projects, who have proven this multiple times. A career is what you make from it, not what others believe it is.

Gif by 60secdocs on Giphy

The treasure hunt

• 📕 Kill It with Fire by Marianne Bellotti is a fantastic read on modernizing legacy systems and code. It is also a great source of inspiration for non-software engineers.

• 💻️ CNCF TAG Environmental Sustainability - in the context of Sustainability and Cloud Native, more than two years ago, I proposed the idea of starting a working group to open up this field and provide guidance. Have a look!

• 🦜 https://internaldeveloperplatform.org/ - While researching for a project I stumbled on this open community/page about IDPs. They provide a good starting point to better understand what IDP do and bring you.

See you later!

Max

p.s. if you want to sign up for the Zeitgeist of Bytes newsletter or share it with a friend or colleague, you can find us here.

I must conclude that AI is far more hyped at a management level than in the engineering corner. Looking back to KubeCon in Paris, it was opened, how it can be different, with a mini AI demo. To round this up, talk after talk, which was not from the community or updates, underlines that AI is a thing. Don’t get me wrong, it is a thing, but people are stuffed with AI. You see it in every corner, at every conference, and in every video. It is great an fantastic what can be achieved with it, when it really helps people. But for many use cases presented, it’s right now more a toy. I hope to see a change soon. However, when talking with engineers, it’s just yet another technology that needs to be implemented. Maybe it is because it is too far off or because the challenges in integrating GPUs and providing environments to make inferences are not that difficult.

I believe we will see this year already the hype turn into reality; we will see a market consolidation and many dreams bursting. Similar to previous years with NFT and Crypto, the whole thing will calm down. That is a chance to then finally talk about serious business and products that are not looking into making fast some money.

European Innovation Council

Talking about innovation, the EIC impact report was just released and the numbers are strong. In 2023, the EIC Fund concluded over 100 investments in deep tech companies worth around €1.2 billion.

Purple: Health Tech; Black: Clean Tech; Red: Digital Industry and Space

Also, Europe often lacks on investments and funding capabilities for startups and innovative ideas the EIC and its predecessors have created so far:

• over 150 “centaurs” (valuation above of €100 million)

• 15 of whom have a valuation above €500 million

• 8 unicorns of whom have a valuation over €1 billion

Most of those are more like the hidden champions. That reflects very well the European market: we don’t have that huge big brand awareness (except in cars etc.), we have the technology experts, niche down on a single topic and no one else can compete there. But no one knows them, either. Who will find the next truly revolutionizing product/service from the heart of Europe?

Twitter/X - and the future of social networks

For 1.5 years, the bird has been running under the command of Master E. While this pushed the growth of users on Mastodon, Bluesky and later Meta/Instagrams Threads; none of them really can replace Twitter.

Are you dead yet? - Children of Bodom, Metal Band

Despite all the critics of Twitter’s new strategies and approaches dealing with questionable content and providing space for those who usually hide under the darknet - the Olymp hasn’t fallen. Many user and usage statistics around it are vague, but if we believe some of them, Twitter keeps growing its user base and monthly active users.

Is it because the unregulated space opened its doors for all the trash you don’t want to see? 🤷‍♂️

Bluesky, Mastodon & co do not really catch up. After an initial spike and scandal-driven shift of users, the number of posts and daily active users remained steady or dropped.

I’m actually tired. I like the short form of those tools, posting content that doesn’t fit a 100% into your professional account and other things. It would be great to have a solution that combines all platforms, such as a short message apps federation. What do you think?

The Graph

I recently learned about the Web3 project, The Graph, primarily developed by Edge & Node. Which problem does it solve? Without The Graph, you’d have to rely on basic read operations built into the blockchain. This works for simple scenarios, but becomes difficult for searching, aggregating, or meaningfully organizing data.

Imagine wanting to query NFTs in a wallet and filtering by a specific characteristic. You’d have to manually process countless events, read metadata from IPFS, and aggregate it yourself. Even a simple process like this could take hours or days to finish. Plus, you’d have to account for the fact that new blocks holding new transactions are constantly being added to the blockchain!

The Graph solves this problem and many more by indexing the world’s blockchain data in a decentralized way.

Essentially, The Graph organizes web3 data.

The future of Web3 is questionable, but I believe we need to evolve how we serve applications. Doing it the old way doesn’t seem to be right anymore. Companies focus and invest (wasting) a lot of money in basic infrastructure services rather than focusing on delivering true value.

But maybe also just our market is broken …

I hope you liked what you read.

See you later 👋

Max

Giphy

Happy New Year!

Enough of the celebration back to business.

What is the next hype you are looking for? Are you betting on GenAI, or which new kid on the block is hiding in the corner?

Believing Gartner Hype Cycle for Emerging Tech we have a bunch of good options. WebAssembly, Cloud Sustainability, Internal Developer Platforms and some Open Source Program Office are waiting next on the list. Yet, I believe outside the server (cloud) room, no one will become that much hyped. I personally can’t wait to see more adoption in those topics. In my work, with my team, and with the open-source community, we have all been working on this topic for a while. It is less a hype and more to be expected. But yet, it misses the adoption.

A section of the Gartner Hype Cycle for Emerging Technologies

Sustainability, Wasm & IDPs

Handle this with care; I’m biased on those topics! However, I have the feeling those are the go-to topics for 2024 (besides GenAI).

Cloud Sustainability

There are multiple topics we have to take into account. Green Coding, Sustainable Infrastructure (hardware, cloudware), Eco Architecture, also Change Management, and overarching principles - such as setting requirements within the procurement process.

I recently published at the Linux Foundation Europe an overview of open-source foundations, which all tackle sustainability from different ankles. From that, it becomes pretty clear we have a movement. Events like the Green Coding Summit by SDIA or dedicated tracks for sustainability at the Open Source Summit underline this.

What is highly interesting in this field is the complexity of the moving parts and the speed at which new tools, methods, and specifications are developed. Measuring the IS situation isn’t as easy as with other static components; optimizing it not only requires better power utilization but also changing the system's architecture, maybe refactoring some code, scalability, and so on.

I find this topic so interesting not only for the greater good, but because it can become engineering wise complex and challenging. Does Sustainability in IT play a role in your organization?

WebAssembly (Wasm)

Wouldn’t a more secure, faster and leaner container alternative be great? No worries about the OS, just pure code. It's a Devs heaven.

Well, almost. Wasm isn’t going to erase containers, it’s more like a synergy partnership. However, the possibility and functionality to replace those parts of the systems that contain actual business logic is an absolute real use case. Platforms such as Fermyon Spin or Wasmcloud are even going further, providing a pure feature-oriented view.

Matt Butcher wrote a nice article at the New Stack about his 2024 Wasm predictions:

• Wasm is the perfect partner for AI: efficiency, speed, and security are the drivers for inferencing, and its portability also to GPUs allows its usage everywhere.

• The WASM standards will be finally defined this year; the WASI is crucial for the general availability, and the component model is a true game changer. “That means that for the first time in the history of computer science, libraries from arbitrary languages can work together. Your Rust app can import a Python library that in turn uses something written in Go.”

• Wasm will find its home and true adoption on server side.

• Running wasm apps in duality, Server Side and Client Side, outperforming any other common framework.

I really recommend you to read the article.

Gif by producthunt on Giphy

Internal Developer Platforms (IDPs)

I truly hate this name. It is the best try to describe the purpose of such platforms, and it might also be the best fit founded, but it sounds like a sneaky project that a bunch of people run under their table.

With the shortcomings of IT specialists, growing prices, and knowledge gaps, IDPs have become inevitable. Right now, most of them feel like a garage plug-and-play with tons of tools, but my guess is that we will see some holistic solutions this year. Maybe some of the commercial onces will do a big step forward, maybe we will see a FOSS alternative striving. RedHats Developer Hub could be such.

Placing DevOps practices in an organization-wide end-to-end delivery platform that welcomes devs, ops and other departments will empower future IT projects, faster time to market and highly reliable almost autonomous solutions.

What about Quantum Computing?

Last month, the IEEE Spectrum published an article on a realistic view of Quantum Computing. In short, it will take way longer to be useful because the use cases are not large and complex enough. Also, the communication on quantum computing isn’t clear, open and transparent. Most things tested failed, in some fictional experiments it opened up that running on an GPU will provide fast results then on a Quantum Computing. Industries such as banking, for example, are always highlighted as the big winners of quantum computing, but they are actually not. Natural science & life science are the sectors which will be able to do useful with it.

Time will tell, but I think we will have way more interesting changes in IT the next year, and quantum will most likely be only exciting in the Avengers movies.

Gif by becausescience on Giphy

See you later ;)

Max

p.s. if you want to sign up for the Zeitgeist of Bytes newsletter or share it with a friend or colleague, you can find us here