AI Engineer Europe 2026: reflection

#ai #conference #opinion

Now that the dust is settling, it’s time to reflect a bit on what I’ve heard at the conference.

My takeaways

This was my first AI Engineer conference and also the first AI conference I attended. It was a nice way to gauge where both I and we as a company stand. This is especially true when I realise that mostly (only?) companies at the forefront of these developments were there. AI is hot, but as we have heard, only 12% of the (big) companies are “AI achievers”.

What I think the speakers agreed on:

  • The role of the engineer is shifting. Less coding, more writing specifications, planning and reviewing
  • Codebases must be designed for humans and agents to read and understand
  • Guardrails are essential. If we want to give agents more freedom to solve problems, we need to give them a certain amount of freedom, but we need guardrails first
  • We need to have feedback loops. The quality of the feedback loop determines the quality of the output
  • The human stays in the loop (at least for now)
  • Introducing AI is a process. Start with non-critical, well-scoped tasks and let the system “prove” itself and earn trust.

There are also things the speakers do not agree on:

Other takeaways (that were not explicitly mentioned by multiple speakers):

  • Agents are bad at self-evaluation
  • Only the first ~100K tokens of context are useful. More context will only make the agent dumber
  • Agents are consuming APIs, documentation and websites and might even already make up a large portion of their users
  • I should probably use more skills

In general, I think we, as an industry, are still figuring out how to work with AI. This is also directly related to the rapid pace at which things are changing. Something may work today, but may be obsolete next month.

The conference

Overall, I liked the atmosphere and energy. The organizers managed to get a great bunch of speakers on stage who delivered insightful content.

I’m a bit on the fence about the workshop day. I really liked that the speakers had more time to go in depth (which is hard in the 20-minute timeslot of the breakout sessions). On the other hand, I was expecting to do more “work” in a workshop. However, most talks I attended were just that: longer talks. And if it were not for taking notes, I would not have needed to bring my laptop.

Having said that, I had fun, learned a lot and would love to go again next year.

The trip

This wasn’t my first trip to London, but it was my first conference there. The conference was held in the Queen Elizabeth II Centre, which is in the center of the city. And especially since we stayed in a hotel close to the venue and could walk over there each day, I was very aware of where we were, which was great. (I like London, can you tell?)

This was also my first time going to a conference with such a big group. Sixteen people from Schuberg Philis were there! This also meant that for most (perhaps even all?) of the sessions I joined, I was not the only one from our company in the room, which means I could discuss what we’ve heard and how it applies to our company and customers. This added a lot of value for me.

And it’s also nice to get to know my colleagues a bit better, especially the people I don’t work with on a day-to-day basis. And while we talked a lot of shop there, there was also time and space to bond on a more personal level.

Conclusion

I think I can say that we, as a company, are in a good place with how we are adopting AI and helping our customers. Having said that: AI engineering is a field that is still very much developing. So we’ll continue to learn and adapt.

As for myself: I’ll work, together with my colleagues, on integrating the takeaways into my everyday work.