The Zechner-Lopopolo Continuum: AI Engineers can't agree on whether we should still be looking at code

Alex’s recap of the best AI conference’s first European debut + a few BTS reflections from the speaker room, hallway track and podcasting from London

Hey 👋 This is Alex.

Do AI Engineers still need to read code in 2026? how about 2027? This is the most discussed question at the first AI Engineer Europe conference.

On day 1 opening keynote, Ryan Lopopolo from OpenAI took the stage and said “code is a liability” and that we should all strive to be “token billionaires” - and on the last day, Mario Zechner of Pi got standing ovations after he told folks to “slow the fuck down” and “read every fucking line of critical code” - this debate happened in every hall conversation, every dinner (speaker or otherwise) and I bet it happens in every company adopting these tools. So ask yourself, where are you on the Z/L Continuum today? Where will you be in 6 months? in 12?

I’m on the plane back home from London where I spent the last week watching, eating with, listening to hundreds of AI engineers, with backstage access to the top AI engineers in the world, and a few themes/takeaways arose.

I know that many of you expect the quick recap, what were all the things the AI Engineers were talking about, so lets start there:

AI Engineer Europe main themes TL;DR

by Alex Volkov // ThursdAI.news // April 2026

• Nobody is sleeping well - we all seem to have AI Psychosis and FOMAT
  Everyone at this conf. is coding with agents, nobody has figured out how to orchestrate them efficiently, all are suffering from FOMAT, that feeling that “just one more prompt” is what’s going to make your Agent code for hours while you’re sleeping/AFK.

• The Zechner - Lopopolo continuum

  First day keynote speaker from OpenAI Ryan Lopopolo wants you to completely stop looking at code and become “token billionaires” and second day keynote speaker Mario Zechner is pleading to “slow the fuck down” and read the code. Everyone is debating where they are on the Z/L continuum and where they would be in a years time. But it’s not a binary, is a spectrum.

• Everything is changing really fast, we need new tools
  IDEs are dying, Github is seeing unprecedented traffic levels (15x their biggest year, which was last year), so is Cloudflare, Software can think now, L33T C0d3 is dead, interviewing is impossible, an AIE is now a PM, Agent babysitter, Taste setter (choose one that applies), and it’s going to accelerate.

• But we’re still early (at least in EU)
  Many engineers and enterprises are still hesitant to use AI, in EU specifically there’s strong regulation, privacy concerns and general un-ease about AI. There is a LOT of software engineers and not all are AI Engineers yet. Nobody knows what is coming for SWEs, and how our role changes.

• AGI is nearly here - just not evenly distributed
  Everyone was talking about Claude Mythos and whether “too dangerous” to release and will be a Y2K level cyber-secutity event, or Anthropic is just doing great pre-IPO scare tactics marketing and don’t have enough GPUs to release it for everyone

• “Just talk to your Clawnker”
  Something changed in Dec25-Feb26, we all feel it, and the meteoric rise of the OpenClaw (which had a whole track on day 1) is a testament. Clawfather Peter Steinberger, maintainers Vincent Koch, Sally-Ann Omalley all told me this is the biggest thing folks don’t yet “get”. Self improving intelligent software is here, talk to it! Stop asking them for trivial advice.

• MCP is dead, long live MCP
  Many AIEs are quick to host MCP funerals, while enterprises adopt MCP and MCP Apps faster than ever, due the inherit security risks of agent skills.

• AI was supposed to make us more productive, but we all just work more
  Jevons Paradox on people’s time is a surprising one to me, we were all supposed to offload the boring, meaningless stuff like writing syntax to clankers, but everyone is just working way way more now. You’re expected to give 10x output + docs + tests + i18n +++ all on your own.

• The MHC is here to stay (Model, Harness, Context)
  This is my concept that I ran by many folks, instead of predicting what WILL change which is hard, let’s try to think what wont. Models will continue to get better, they will always need harness/tools and your personal context / memory will matter more than ever.

Day 0 - Workshop day, Great weather day, Speakers dinner

When folks pulled up to the event, there were some venue related issues with the workshops, space was limited and the queues were a bit long. The most important things about a conference were top notch: Food was great, Wifi was Great, Speakers were exceptional.

Also, there was a red fucking GIANT INFLATABLE LOBSTER right in front of the venue.

Oh and the the weather! The weather in London, OMG, it really felt like the AI Engineer conference not only have brought the brightest SF AI minds to London, they have brought the weather with them. So I think folks who didn’t get into a workshop actually won a half day in London, in great weather. So win win?

And the location! Most who stayed at the event sanctioned Park Plaza hotel, were just 7 minutes walk from the venue, crossing the River Thames over the Westminster bridge, with an incredible view of Elizabeth Tower (Big Ben) and the Houses of Parliament, Westminster Abbey, the London Eye, South Bank. It was just delightful to walk to the venue and back. 100% the location is by far the best one out of all AI Engineer locations so far. 10/10 would do this again.

The contrast was stark, between the Royal “office” of the king and royal family, parts of which were built over 900 years ago, all majestic, gothic, and regal with the conference in which the future of humanity and artificial intelligence is being discussed.

Though I will say, I was applauding the brits, as many of the speakers of AIE Europe were invited for “tea” at 10 Downing St (the “white house” of Britain and the seat of current PM), to discuss innovation, AI and the future. (Yours truly invitation must have gotten lost in the mail, but I’ll try to sneak in there next year to cover).

Speakers Dinner - Single highest density networking in AI, in all of EU, maybe the world?

Look, I’m not going to lie, when I say, if it were possible to only choose 1 part of AIE to attend; Between the workshops, the speakers green room, the hallway track, the keynotes, the presentations, and the speakers dinner, I will 100% of the time, choose the speakers dinner.

(This is also, why you should absolutely don’t hesitate and submit a CFP to the upcoming event).

The Caliber of the folks attending is sky-high, as Swyx as the curator of the talks reviews them and their talks, and only the best ones get to come on stage (and dinner) so the social ROI is off the charts. As a podcaster, there’s no one single room that’s better for me to be in.

The table I set was almost literally the two sides of the Z/L Continuum. On the one end, Mario Zechner (creator of Pi), Armin Ronacher @mitsuhiko (creator of Flask, EarendiL) and their co-worker Cristina Poncela. On the other side of the table, Ryan Lopopolo (@_lopopolo), Vaibhav and Katya from all from OpenAI.

Also the same table had keynote speaker and friend Peter Gostev (@petergostev) from Arena and my new bestie @marlene_zw who’s a Github/MSFT DevRel and a Python Software Foundation fellow and Matt Pocock @mattpocockuk who is an incredible teacher (and the other potential Conference MVP, people LOVED his /grill-me skill)

And this was just our table, other tables were as stacked as this one, and all were discussing their job, their Claws, their troubles sleeping, the dopamine rush from getting agents to do the task “just right”, the ways to handle context drift, Claude Mythos, MCPs, whether we even need to look at code anymore... while enjoying a very tasty Indian dinner.

A few anecdotes from this dinner include an introductory speech from Lia, Ben and Swyx, which was followed by a non traditional speech from the Head Chef, which was fine, but then, he called the wine guy to give a speech, to the complete surprise of the organizers and ... Not one person could understand what the hell this guy was saying, as he proceeded for 5 minutes to yap about what wine selections they have for that evening in a thick french accent. At some point this became nearly comical 😂 And while the food was great, the conversations were the most important part. Also the speaker gifts were awesome, especially the female speakers who all scored a Liavella bag (because Lia the GM of AI Engineer also has a luxury bag company)

Day 1 - Keynotes and a ThursdAI Live with 5 guests.

Beautiful opening keynotes day, and the kickoff was electric. First Keynote from Malte Ubl (@cramforce), CTO at Vercel and a great conference guy himself with JS Conf, was a great start!

Malte talked about a bunch of things, but mostly, what does an internet for agents look like. Specifically mentioning that over 60% of the traffic to Vercel comes from bots/AI Agents. Which is kind of a big deal, and I expect that to rise.

One thing I haven’t considered, until I chatted with Sunil Pai (@threepointone) from Cloudflare (on the pod, offline, coming soon!) is that when you ask your agent 1 answer, it can very VERY quickly go to over 30 web pages to bring you an answer in a fanout-multi subagent search. This puts strain not only on Vercel, but on Cloudflare and every other network provider.

And then we got the author of the L of the Z/L Continuum - Ryan Lopopolo - who opened with the brilliant line “hey, I’m Ryan, I work at OpenAI and I’m a Token Billionaire, and I want you all to be as well”. Ryan was likely the most AGI pilled speaker of the bunch, and having spoken to him the night before, I understand why. You should definitely watch that talk, where he says we need to think bigger, coding is either solved or will be solved very soon. He also says that “code is a liability” and that we should focus on tests, evals, fixing our harnesses and making sure our every time an agent does an “oopsie”, it’s the prompts fault and we need to fix the prompt.

Speakers lounge hang

I then left the audience and went to hang out in the speakers lounge, which is the second most high density room at the whole conference, not before the speakers go on stage, but definitely after.

Where I finally met a bunch of the OpenClaw crew, including the Clawfather Peter Steinberger (@steipete) who was the keynote speaker of the (first ever) OpenClaw track; Which happened on the main stage after the morning keynotes. I also met Vincent Koc (@vincent_koc), the #2 maintainer of OpenClaw after Peter, and convinced him to come on the pod, and we had a great chat!

Peter was surprisingly chill, considering the guy recently had a fiery increase in personal fame (and I’m assuming fortune), having been on every top podcast, met with everyone possible in Silicon Valley from Sam to Elon to Zuck to Jensen, has half a million followers on X, and changed the world with OpenClaw to the point that even grandma’s in China are installing it (not to mention literally every AI lab in China has an OpenClaw service)

I then went into prep and writing mode, as I had a livestream coming up, which, first time in EU timezone threw me a bit off. You can read and watch that episode here, I will say, this was by far the densest ThursdAI show I’ve ever done, we were strictly limited by the crew to 2 hours (we usually go over), and with 5 guests, me and Wolfram on the scene, Yam, Nisten, LDJ joining remote, it was quite a lot of balls in the air to juggle. I’m very proud of myself for a nearly flawless execution of this. I hope you enjoyed that ep as much as I did making it.

Anyhow, quick editing + newsletter writing, and I rushed to one of the many side-events that always happen around AI Engineer. My advice: go to the Conference website early, and apply to some of these side events ahead of time, they often undershoot the amount of people who want to attend them, and are overfilled with last second folks. Case in point, after a dinner I was able to get into (Thanks to Marlene!) we tried to get into the event at the London Aquarium, walked there, only to be told “sorry, you can’t get in” even as other people were leaving, the brits.. they do love their rules and regulations.

Day 2 - Z teaches slowing the fuck down, and speeding back up again.

The last day of the conference, and for me, Fridays are blessed — it’s after Thursday, my big thing for the week is done, and I can “relax” until next Wednesday. Except there’s a near-empty podcast booth right there, and I have access to all these amazing people I just befriended for 2 days. It would be a miss of a lifetime to not sit them down on a microphone. I thought I would get Mario on, but alas, he got too busy (I don’t blame a person who gets to chat with Gergely Orosz).

Speaking of Mario Zechner, when he went to give his (I predict, MVP talk of the conference) talk, in a billboard charity t-shirt, something weird happened. I’ve never seen a room of speaker all preparing for their own talks, and are excited for the opportunity to chat with each other, all quiet down and listen as intently. There’s a live screen in the room of the main stage, but usually people don’t pay that much attention to what’s going on on stage. This time, everyone just... quieted down and listened to Mario tell a room full for AI Psychosis having engineers, to SLOW THE FUCK DOWN. It was glorious. He also got a standing ovation when he got off stage which is also, I think a first time I’ve seen this.

Mario’s argument isn’t anti-agent. He’s not a Luddite. He uses the agents all the time. His point, and this is why it landed so strongly, is that the industry-wide quality bar is collapsing, that everyone has stopped caring about craft, that the median engineer is shipping slop they don’t understand and don’t even know they don’t understand, and that this should bother us. Mario literally said “Read every fucking line of critical code”! This is of course the Z of our continuum.

I think it’s part Mario’s Austian/Old gramps charm, and part... everyone kind of needs to hear this? AI coding is incredible, but there’s no doubt that there’s a tension where on one side, we all need to accelerate to really use these tools correctly, we feel that if we don’t rush after the next great thing, we will stay behind and NGMI, and folks from big labs are telling us to not look at code, and from the other, this introduces a lot of “small” things, and sometimes medium to big things into the codebase, not only that, it removes the engineer’s ability to even understand the codebase, which, can turn into subtle and then oh-not-so subtle bugs.

I’ve chatted with a lot of folks, and asked them, where on this continuum of “I don’t even look at code, I YOLO it to my Clankers” vs “I don’t write code but I read and code review myself and code is important” and there doesn’t seem to be a great answer. I’ve also asked people if they expect to be on the same place on Z/L Continuum in 3,6,12 months, and nobody could give a confident answer that they will still be looking at code in 12 months. Especially not with Mythos from Anthropic being announced just a few days before the conference and touted as the most incredible innovation that is so good at code it can find bugs in a 27yo codebase that were never found before.

Oh, and we were able to sneak a Weights & Biases talk in there, as Wolfram (@WolframRvnwlf) applied for a hallway track talk about Wolfbench.ai and got accepted. 👏 I wasn’t able to watch it unfortunately, but I was told it’s recorded and I will send a link in one of the newsletters once it comes out! Wolfbench is actually a very poignant new benchmark from us at W&B, created by Wolfram, where we test not only model capabilities on TerminalBench, but also Harness capabilities with each model, and whether the harnesses improve independent of the models improving. Do check it out if harness engineering interests you.

4 more pods and a surprise MC opportunity

As I was booking podcast chat with folks, I got a text from Ben the organizer, that the mains stage MC, Tejas Kumar (who is a great energy dude btw!), needed to step away, and if I can jump in as an MC?

I already had a pod slot booked and quite a few guests, but hey, a main stage is the main stage right? Also, this was the Coding Agents track, arguably the most important! So I moved the pods to before and after the MC block and jumped on. And I’m so glad that I did.

The job of the MC is to smooth the transition time between speakers, hype up, get the crowd clapping etc. But also to make speakers feel comfortable (many speakers, even the most experience ones, are getting the jitters before talks).

As I was chatting backstage with my 2nd speaker, Michael Richman, I learned so much! First, he coined the term FOMAT, which is that feeling you get when you’re about to leave for lunch, or go to sleep, and you’re about to kick off a refactor, or send a big research prompt, but are worried that your your agent / clanker will not be able to do it. That you will come back and instead of a finished feature, you’ll see a blinking question that the agent sent 2 minutes after you left and was just sitting there idling. FOMAT: is FOMO, for Agent time!

I thought it was a great talk, and it’s definitely something that everyone in the crowd (including me) have felt. Michael is also from Denver, my home town! At some point after we chatted, Michael asked me what I do, I told him I have a podcast, he asked which one? And when I said ThursdAI, his eyes lit up! He was so happy, started clapping me on the shoulder, saying “You’re ThursdAI!” haha. (That’s how I found out he’s a listener, not a watcher lol)

It’s a moment I’ll never forget. I always love meeting listeners of the show, and AI Engineer is the best place to meet them.

I jumped off the stage and rushed to back to back recordings to chat with Sally-Ann Omalley from RedHat, a wonderful conversation with a wonderful human being, I can’t wait to share with you. We chatted about OpenClaw (she’s a maintainer), safety, and all kinds of things. It was so great.

Then with Sunil Pai (@threepointone) from Cloudflare, another wonderful human, who actually reviewed my notes about him and made edits, we had an incredible flow! Sunil is just a fountain of energy, and we chatted about all kinds of things Cloudflare and AI , Agentic future, SWE future.

I can’t wait to edit and send you these chats, because you know what I noticed? When it’s not live, I actually have time to enjoy the conversation, not stress about things going wrong/long/boring/breaking, and I think it shows! I hope that I find a way to bring that fun back to the live shows as well 🙂

Ending the conference on meta and bullshit (bench)

After I was done, I was finally able to truly relax and get back to speaker lounge, to listen to Peter Gostev from Arena.ai talk about his experience with what LLM models are still failing at and his Bullshit Benchmark. I strongly recommend rewatching that, I’ve known peter for a while online, and became friends at the last AIE, and he’s joined the LMArena crew since then, and is extremely insightful about different models and their differences.

And then, Swyx took the stage and closed down the show.

Swyx, as always, is a dope speaker. He doesn’t want to take stage time away from his other folks, but I always implore him, take the stage, give a recap of where you are, as the curator, folks want to hear from you (so Swyx, if you’re reading this, Kudos, but also, please be a more active voice in your own conference, you are CEO after all! People from all over the world want you thoughts)

And oh boy did he deliver. It was a fantastic closing keynote, very meta about the conference itself, and how they have automated tons of aspects of it as a business, which agents did what. IDK if you noticed, but the AI Eng website had all the agentic AI stuff imaginable, for you and your Clanker. They have speakers.json, schedule.json, MCP, skills, there’s just everything on there.

The highlight for me was the (now infamous) red inflatable lobster at the entrance, in which Swyx saw something online, got inspired and asked Devin to go research suppliers, contacts, sites) and showed that a coding agent can do actual work, and voila, the lobster was a HUGE success.

I think I know 100% where Swyx sits on the L/Z Continuum, can you take a guess? I’m pretty sure Swyx cared to see 0 of that code himself!

AI Engineer Conf - a brief history

AI Engineer, which is a term Swyx coined nearly 3 years ago on Latent Space, has led to renaissance of glorious tech conferences that kind of died during Covid.

If you’ve been reading/listening to ThursdAI all this time, you know that I’ve attended and covered all 6, spoke at 3 of them (though everyone remembers the one time I dressed up as a judge).
The first one (AIE Summit 2023) was a monumental step up in my -back then, still brand new- career. I’ve got to network with the best people at the speaker dinner, I’ve got to interview around 100 attendees with a run-and-gun microphone and a camera crew following me around about what they think an AI Engineer even is.

Mind you, this was before (way waaay before) Claude Code, before Vibe Coding was a thing, tool calls were literally introduced that month at OpenAI, context windows were maximum 32K, and LLMs were chatbots and APIs, Vector DBs mattered a lot and Google was nowhere to be found in the AI race, and the even the most AGI pilled didn’t believe we’ll have the Z/L Continuum.

And still, the hotel Nico was buzzing with people who saw the opportunity, who could imagine where this was all going, who were already getting “AGI Pilled“, building agents (anyone remember AutoGPT? big sponsor at that first conference, talk about being too early!). “We are just in time!” Swyx told the 600 or so hand selected folks at the AIE Summit, not too early, not too late, just right. Well I have something to say to 2023 Swyx, we were ALL so early!

The tools became incredibly better, in a way that I think even the most AGI pilled of us back then could not have predicted, and so has the conference itself. AI Engineer has grown nearly 1000% since then (per Lia McBride, now GM of AI Engineer and a huge part of why this conf runs so smoothly), in both attendees and reach (Check out the YT channel, it’s nearing 350K subs and has some of the highest quality content available, but is also a sort of time machine to how we were all collectively thinking about AI Engineering just a 2.5 years ago)

As the AI Engineer conference grew, so has ThursdAI. We’ve moved from only audio to live video and boosted the production quality and guests caliber, leading to moments like a surprise live interview with Logan Kilpatrick & Jack Rae, first “on the floor” podcast studio in AIE:Code NY, back to back interviews with folks from DeepMind, OpenAI, Github and OpenClaw folks.

So when Swyx reached out for me to come cover this one, it was an immediate “Yes, can I bring Wolfram with me”?

Closing words

After the closing keynote, everyone went to Fabric, one of London’s main clubs, to celebrate. I couldn’t make it — I was fried from socializing and went to a smaller dinner instead — but I heard it was great.

If you ever had any doubts of whether an AI Engineer ticker was “worth it”, it is. If you had doubts of whether you should apply to talk at one of the upcoming events, you should. And if you ever doubted that an AI Agent can find and order a British yeoman-clad gentlemen to stand outside of a venue for two days propping a cart with a huge giant fucking inflatable red lobster ... well, I have some pictures for you to convince you otherwise.

So my last question, you guessed it, where do you sit on the Z/L continuum dear reader? Do you expect to answer with the same answer in 3-6 months? How about a year? Comment below or come tell me in person at AIE Worlds Fair in San Francisco in June. My honest truth? I personally sit on the absolute AGI pilled L place on the Z/L, but also, I also do not maintain any critical code that folks use.

P.S - if you made it all the way through, thank you so much for reading, I’ve asked a few friends to review this, huge thanks to Wolfram, Kwindla, Claude, Wolfred and Amy for proofreading, special thanks to Dex for the Z/L shorthand that made me rewrite this whole thing from scratch and an incredible debt of gratitude to everyone who made AI Engineer possible, Swyx, Ben, Lia, VB, Raouf, Phil and Tejas, Kyle and the podcast folks who came saved my bacon, and to every speaker and attendee who kept the vibes going!

Alex Volkov
AI Evangelist at Weights & Biases, Podcaster, covering AIE for the 6th time. Completely AGI Pilled.

Pictures provided by AIE and Toby Merrit Photography (website)