I had the pleasure to visit this year’s Entwickler Summit 2026 in Berlin. It’s held by the German-speaking, well, let’s call it “software developer learning platform” entwickler.de (and yes, Entwickler basically means “developer,” so you get the idea).

Consequently, most talks were in German, although many slides were in English. I am not sure if we should really focus on only the German speakers in our community, but on the other hand, it was refreshing to listen to the talks in your own language.

It took place in a cinema, the Colluseum, and which is a great place for such an event. Just the wifi was lousy, and the lunch was out before I could grab some. Bummer. But who needs food when there is AI?

Because not surprisingly, most of the talks and topics contained AI either partially or fully. The talks I visited went from criticism over pure pragmatism to curiosity and pure excitement. Pretty much the mixture I experience among fellow developers today. Plus maybe even some doomerism.

Local LLMs

Starting the day with a good mug of filter coffee served from big thermos cans (a continuous delivery pipeline that just works!) and the talk Lokale LLMs: Sind sie den Aufwand wert? (Local LLMs: Are they worth the effort?) by Rainer Hahnekamp (Soverius AI).

Highlighting the benefits (sovereignty and cost control) over the drawbacks (investment), I found this talk highly entertaining, since Rainer managed to convey his content with nicely animated slides and the ability to put complex concepts (like quantization) into simple words. Great!

And yes, starting with 5k to run decent LLMs locally is something most of us can’t afford as individuals, but organizations should definitely do the math for themselves. As technology progresses, we can expect more capable models to run on existing hardware, hopefully, and Rainer draws the comparison to how video compression algorithms improved over time.

I missed the fact that using local LLMs decouples you from the big AI datacenters, which are often pure greenhouse gas-spitting monsters.

It’s all been there

Alistair Cockburn (where the “ck” in the name is silent, which is good to know; otherwise, it could be embarrassing to announce him wrongly) gave the first keynote, Simplifying Software Design: The Genius of Bureaucracies. Did you ever wonder how bureaucracy helps in software engineering? Me neither, but it all makes a lot of sense!

By saying “Not my job” and “No need to know”, it helps improve your architecture in terms of separation, encapsulation, and so on … and works extremely well with, of course, LLMs (because AI has to be in every talk, doesn’t it?).

Mr. Cockburn spat out the fanciest slides (I mean, look at that monster on the right!) of the whole day, often holding far too much information to even read, let alone process, before proceeding to the next. Nevertheless, a very informative talk from an absolute software engineering legend.

Jobs change

Markus Andrezak (überproduct) gave a talk from the Product Managers perspective: When Code Becomes Free: The Organizational Bottleneck of the AI Age. Setting the stage with comparing AI-driven development with the idea of continuous deployments back then, when people were skeptical or even repellent.

And although using AI in development decreases the amount of time needed, he pointed out that code actually never was the bottleneck. Yes, we heard that before, but it’s not less true.

Bottlenecks now are, for example, too much code, PRs by humans, useless features being released, or the slow feedback loop, which tells us eventually if a feature is a success. Or did you know that only one out of three features being released is actually beneficial to the user?

That is also why, in his opinion, the job of a PM changes: less gatekeeping, less ticket prioritization, and rather testing a feature in 5 different AI-generated versions before deciding which one to build properly.

Finally, Markus stated that management decisions that are relevant for software development should be written down and stored in the repository on GitHub so that AI can take it into account when building stuff.

Otherwise it’s just “building more rubbish in a shorter time”, which is not the only time I heard that quote that day.

APIs and AI

Next up: APIs für die AI Nutzung: Wie geht das? (APIs for AI Use: How Does It Work?) by Erick Wilde (OpenAPI Initiative) and Thilo Frotscher, where it was pointed out that LLMs will benefit from well-documented, well-versioned APIs. So if you haven’t, better document your API in some useful format like OpenAPI.

Next steps in the API AI evolution could be to improve the definitions through backward-compatible changes (no customer impact), set up an entirely new API (customer impact), or, the cherry on top, add an orchestration layer, where multiple API calls are hidden behind a wrapper.

Humans and AI

Menschen und KI im Feedback-Loop der Softwareentwicklung (People and AI in the Feedback Loop of Software Development) from Eberhard Wolff (SWAGLab) came to a similar conclusion as Markus before - without clearer guardrails, we’re just building slop faster. While AI will speed up the development loop (build, deploy, test, code), it currently won’t help with getting faster customer feedback.

Furthermore, companies also struggle with unclear business values, excessive planning, long times for decisions, and so on. So eventually, although Big Tech is currently firing people to finance their hunger for data centers, software development will always be a team effort. Developers are wise to become domain experts and work on their soft skills (which, as he acknowledges, can be challenging for some of us who chose computers for a reason…).

We are all doomed

After that somewhat uplifting talk, Dirk Specht’s keynote Die Zukunft lesen - ohne sie zu verpassen (Predicting the Future—Without Missing It), was like a slap in the face. Humans are particularly bad at grasping the concept of exponential growth and thus underestimating the colossal shift we are about to experience. As an illustrative example, he showed the share of electric vehicles over the last decades, which, after going slow for many years, is now starting to grow fast.

And whilst in Germany politicians – and, surprisingly, some business leaders too – are trying to cling on to their beloved internal combustion engines, even though these are already technically and economically inferior, they will soon be phased out. The same applies, in Dirk’s opinion, to people who deny AI as the technology of the future.

In times of disruption, it is important to not stay with “boundary value optimization,” e.gm trying to speed up the software delivery pipeline by a few percent rather than unlocking the full potential. Humanoid robots vs. highly specialized robots are another example here: You can still improve the machines to a certain degree, but humanoids will unlock whole new opportunities. AI might even be capable of skipping programming languages entirely and writing assembler code directly.

No word on how this exponential growth can be handled from a social perspective, like, how do we compensate people who lose their jobs because of AI? And where the energy should come from, not to mention the environmental impact of building massive data centers.

What the bottle-heck

Code no longer is the bottleneck is a quote you probably often hear these days. I would like to refine it a bit:

Shipping meaningful code is still the bottleneck

Everyone can blast out lines of code like nothing these days. Shooting features at customers like a shot gun might lead to feature fatigue very soon. I think we will only learn in a few years what that approach will cost us eventually. Currently, many people even seem to be in some sort of AI psychosis, or, like Markus coined it, “it’s like a drug”.

What I missed completely among the topics that have been discussed: sustainability. The price humanity pays for AI is already very high, and I can’t imagine how it will hit us if Dirk is right with the exponential growth. Yet, most professionals don’t seem to care how many new gas turbines have to be built to create features that no customer will probably ever see.

At least some critical thoughts about sovereignty or ethical problems, like the Huggingface hack, were discussed during the panel discussions that took place on this day, too. Of course I couldn’t see all the talks, so I might have missed these thoughts elsewhere.

All in all, this was a fantastic event, which left me with plenty of thoughts and inspirations. That’s what makes these conferences so unique and important to me.