AI Engineer Europe 2026: Takeaways from London

I volunteered (associate) at AI Engineer Europe in London last week. Three days: workshops on Wednesday, presentations on Thursday and Friday. I had my family with me, so I didn’t go for any drinks / side events.

Workshops (Wednesday)

Anthropic

I arrived early for the Anthropic “How to Build Agents That Run for Hours (Without Losing the Plot)”. I assumed we would go through https://github.com/anthropics/claude-quickstarts/tree/main/autonomous-coding and ask questions, but it was more of a presentation on the whole “Adversarial Agent” pattern.

The big idea (as I understood it) was to have a Generator (not good at critique) <-> Evaluator (possible to be critical) aka the “adversarial eval”, two agents running in tandem.

It was interesting how they mentioned this is “Agile pilled” with familiar development ceremonies like:

• Sprints
• Definition of Done
• Scope clarifications (kick offs)

Though I couldn’t help think this is NOT Agile as I know it, nonetheless this AI usage pattern neatly fits with the way a lot of software is produced up until now in “Jira pilled” Enterprises.

@AshPrabaker had a curious recommendation of using JSON since the agent (Claude) is less likely to manipulate it over Markdown. He recommended Playwright MCP (I was a little incredulous about this) and reading traces. I noticed the nomenclature Anthropic were using, they use the word rubric a lot which I understand as a “prompt”.

Observability and team collaboration isn’t quite figured out and this approach doesn’t make sense on brownfield projects. A lot of food for thought and the whole experience helped my FOMO, since I feel my own AI journey was pretty close to what they and others were doing.

Pydantic

I drifted into @samuelcolvin’s workshop and I was glad I did. I’m a big fan of Pydantic model validation, and I had no idea about Logfire and these other AI ventures. This was a proper workshop where we effectively refined a prompt to achieve better results with a model.

I learned about GEPA - Genetic Pareto and associated Pydantic AI tooling, though tbh this use case was a bit specialised to me. Since I’m a frontier model addict, I don’t need to optimise the use of existing models. Useful nonetheless to know in case I work with a client who wants to for cost / scale reasons.

Loved how hands-on Sam was.

Google - Building conversational agents

I attended this because of legend @thorwebdev who I know from Singapore. It’s a bit frightening how good the Gemini AI audio stuff is with its multilingual support.

Tbh I was exhausted after these sessions and the weather was awesome so I ran out to play with my kids in Clapham Common.

Talks Thursday April 9th 2026

I recorded a Podcast from a LONDON BUS otw to the event:

I’m glad I volunteered. Shout out to the organiser Natalie for being accommodating with my dress code. Was great to meet other volunteers like @oalbacha & @sergiopesch.

Vercel

Finally get to see @cramforce IRL. Loved his talk (and Vercel aesthetics), and his final slide about:

1. Future A - the labs win
2. Future B - we win

… left me a little conflicted tbh. May be that was the point?

OpenAI

Listening to Ryan Lopopolo felt like I was in the Anthropic workshop again the day before. Harness, get agents to do things, etc.

OpenClaw

The legend Peter seemed a bit exasperated and I feel like there was nothing new in this presentation. Maybe AI will finally cut through the CVE noise and surface what actually matters.

Scraping the Web

Whilst helping out with crowd control in Rafael Levi’s Your Agent’s Biggest Lie: “I Searched the Web”, I learned about https://brightdata.com/ai/mcp-server which after testing, is able to scrape https://www.screwfix.com/ aka bypass Cloudflare’s captchas IIUC! WIN!

Reverse engineering a Viking VOIP phone protocol with Claude Code

Again I was manning the door, but this interested me greatly since I have a couple of VOIP phones I need to maintain! The https://elevenlabs.io/ demo where you needed to answer some questions with an AI with Michael Caine’s voice (used with permissions) was VERY impressive. One of the questions was about a Deepmind game, I answered “Baduk”, it said wrong answer, the right answer “Go”, I protested and the quiz corrected the score. Amazing.

Why Your AI UX Is Broken (and It’s Not the Model’s Fault)

@christensencode showed how to keep sessions alive with @ablyrealtime’s AI Transport, which is impressive. Atm I use a messy combo of Whatsapp with Openclaw and https://cmd-ctrl.ai/ to keep in contact with my sessions, though this technology seemed awesome for building something that scales to the public over the Web.

Was great meeting with my ex-boss & Ably CTO Paddy. An incredible guy.

The Production AI Playbook: Deploying Agents at Enterprise Scale

This was long but really well structured presentation, which kept you baited to the end (like a raffle) to get the artefacts: https://drive.google.com/drive/folders/1acC5n1iE_63onW0S6960cYsAdHOMxVQE

Lots of good advice for consultants like myself to deliver successful AI deployments.

I went to join my family at 4PM since I was exhausted by then…

Talks Friday April 10th 2026

Gemma, DeepMind’s Family of Open Models

I learned from @osanseviero that Google publishes capable open weight models. The only use case I could think of using it for is Automatic Number Plate Recognition (ANPR), but an attendee said that probably wouldn’t work.

The Future of MCP

David Soria Parra came out fighting with “it fucking works” MCP and I loved it. David comes across as an “Alpha” technologist, super driven and good to see. MCP is very much alive!

AgentCraft: Putting the Orc in Agent Orchestration

This was one of the most creative / insane talks I’ve ever witnessed. Who knew that those countless hours playing Warcraft (nowadays Starcraft 2) as a kid has equipped me to orchestrate Agents? Amazing Ido Salomon.

Building pi in a World of Slop

This was my favourite talk and what I wanted to hear: “Lets slow the fuck down!”

People of pi. You can view my 18 minutes @aiDotEngineer Europe talk here.

May it spark joy. And laughter.https://t.co/rSDnZAnowQ

— Mario Zechner (@badlogicgames) April 10, 2026

The Friction Is Your Judgment

This felt like a follow on from Mario’s talk. I was loving it.

Loved "Building pi in a World of Slop" @badlogicgames 👏... yes, we need to slow the fuck down. Love this embracing friction programming @aiDotEngineer ... pic.twitter.com/qo5veVOGvO

— Kai Hendry (@kaihendry) April 10, 2026

Every API Is a Tool for Agents

Finally I met @mattzcarey IRL. My take away: Cloudflare are sorting out their confusing CLI. May it compete with the aws cli!

Expo

Spent time chatting with people for a change! I was impressed by https://arize.com/ AI observability demoed to me by the 10x Engineer Laurie Voss. I learned the term AX.

Let’s Talk About FOMAT – Fear of Missing Agent Time

I met bit.ly veteran Michael earlier and I supported him by attending his talk. I am now a https://cmd-ctrl.ai/ user and I love it!

The venue

https://qeiicentre.london/ was in a fantastic location, but it was a little weird!

Closing thoughts on ai engineer conf:
- weather so great
- did not like the venue one bit
- can’t say any talks really stood out for me
- but had very interesting hallway conversations. Especially with @_lopopolo who works opposite of me: he just lets it rip and not worry. 1/n

— Lucas Meijer (@lucasmeijer) April 11, 2026

The wifi held up really well, but some domains were blocked — the TLD .dev, bizarrely.

I noticed the toilets all were missing the hand driers, I presume because of COVID?

One of the overflow rooms had a leak. I ended up using the stairs a lot of the time since they were replacing the elevators during the conference!

The navigation of the QE2 venue was quite confusing. I understood where things were on day three.

AI Engineer conference

https://www.ai.engineer/ by @swyx & pals was a triumph. Best AI conference by far, tastefully localised and I am feeling much better for it.

My take away

I came to this conference with a lot of FOMO and imposters syndrome. After the presentations and conversations, I feel confident about my experience with AI.

I’m going to endeavour to get back to basics, use simpler tools like https://shittycodingagent.ai/ and remind myself to slow down and have fun with this.