AI Engineer

The 2026 World's Fair is completely sold out 🫡 ✅ The largest AI industry expo on earth ✅ Sold out on Leadership track for CTOs & VP AI's ✅ Sold out on Workshops tomorrow ✅ Sold out on ALL late bird tickets 🙌 65 side events still FREE all over SF (see website) What we will never sell out: Our commitment to publishing all the best AI engineering content for free online on YouTube. We have now opened limited overflow tickets for our expo and engineering tix — no seating guaranteed, sessions are first come first served. If you ARE one of our attendees, DO come down to Moscone for New Engineer Orientation tonight from 5p-9p to meet new friends and skip the morning crush for tomorrow. We expect EXTREMELY heavy last minute registration and need your help to load balance across days. Please give your speakers and sponsors all the love for all the effort they are putting into making this the greatest show we have ever done!

secondaryslogan.jpg

"Every enterprise has realized they have to make their UX agentic, or the current one will die." We sat down with @KaranVaidya6, Co-Founder of @Composio, at our @aiDotEngineer booth. - Why 2026 is the year agents actually work - What actually counts as an "agentic harness" - Distilling 10M+ daily tool calls into reusable skills - How self-improvement becomes the next milestone

AI models are only as good as their data feed. From @HumanXCo and @devworld_conf to @superai_conf, @aiDotEngineer, and @RaiseSummit, we kept hearing the same thing from builders: real-time, verifiable, fresh web data is the missing layer for reliable LLMs. We've spent years turning the open web into that grounding layer, structured for every model, at scale 🛠️ Follow us to see where we're headed next 👋

@AvimanyuRoy3 @headinthebox A monad is just a monoid in the category of endofunctors, what's the problem?

@headinthebox @aiDotEngineer alas I haven't dived into Monads! But yes, it is insane the amount of control agents have been given and the utter lack of control over them! The harness design is so poor!

Forty-odd slides that weave four story lines together with subtle and not so subtle jokes featuring Dario, Daniella, Sam, the Pope, Bernie, Claude, .... and code samples in Lean and Dafny. I guarantee that my AIE World's Fair talk [0] will be more fun than all those after parties you have been waitlisted for!

the @aiDotEngineer talk from @geoffreylitt is very good and lines up with everything we’ve been building towards for agentic coding - - accelerate human understanding - make understanding a team sport - what is to be done and/or what was done - embedded html in docs as a key level-up - embrace the natural method - learn what happened in your code the way you learn languages, by immersing yourself in the details youtu.be/WkBPX-oDMnA?is…

10 agent terminals running at once. That was @steipete setup at OpenAI in January. "I was the scheduler, the router, and the memory. Unlike tokens or compute, I can't simply add more attention." He thought he was orchestrating. He was polling. Every engineer flexing about parallel agents is describing the same bottleneck they moved onto themselves. The model got faster. You didn't. 6 minutes from OpenAI's @aiDotEngineer World's Fair keynote, from the guy they call "the claw father." Watch it, then read the article on AI in brownfield below.

Truly wild that prompt engineering is beating fine-tuning, RL, distillation when it comes to model adaption. Also wild that twice as many AI engineers are using RAG than fine-tuning. The top 3 techniques for adapting model behavior actually don’t touch the original model weights at all. And there’s a huge drop off when you look at techniques that do touch model weights. A few reasons for this: - There’s a skill gap → adapting model weights is much harder than other techniques - High quality synthetic data generation is very hard at scale - There are few good out-of-the-box solutions for fine-tuning Ofc the sample of participants (AI engineers at @aiDotEngineer World Fair) plays a role too, as they likely have less experience with hands-on ML techniques, but it’s still an important subset of people using AI models. Really interesting report from @AmplifyPartners.

🆕Don't Build Agents You Can't Answer For youtu.be/n97BCfyFIvw @addyosmani's closing keynote on: - roles/titles < system ownership - on @mitchellh: "Taste is the ability to make high-quality qualitative judgments where no objective metric exists yet." - "if generation scales faster than comprehension, the scarce resource becomes judgment backed by evidence." His new rule: "Explain it or don't ship it."

Best part of @aiDotEngineer was not the talks. It was the people at the booth who stuck around to talk agents. @CoreWeave ARIA, our AI research and iteration agent, came up in almost every one of those chats. Thank you SF. @altryne and @WolframRvnwlf had a blast!

@wandb @CoreWeave @altryne @WolframRvnwlf nice video!!!!

@ThePrimeagen @RhysSullivan i humbly suggest watching the talk I gave at @aiDotEngineer youtube.com/live/htM02KMNZ… getting loops to work well is ~ a control theory problem

Dominic Kundel (from @OpenAI) gives the inside scoop on where GPT-5.6-Sol gets magical: computer use. Background browser tabs, app control, multi-agent fanout, and Codex verifying its own work all change when latency drops. Join us in the Token Billionaires Lounge, presented by @cerebras and @aiDotEngineer. In conversation with @dkundel // @MilksandMatcha

@WolframRvnwlf @TheCodingSoup @wandb @CoreWeave @altryne ai.engineer/nyc

@TheCodingSoup @wandb @aiDotEngineer @CoreWeave @altryne My throat hurt and my feet ached, but I am ready to do it again anytime. We need more AI Engineer conferences! Bring on the next one, wherever it is.

@solomonstre @simonw @headinthebox live now: the closing keynote from the one and only @headinthebox youtube.com/watch?v=-CnA2l…

🆕 In Code They Act, In Proof We Trust — Erik Meijer last year, @solomonstre defined agents as "an LLM that's wrecking its environment in a loop", and @simonw coined the Lethal Trifecta for agents, that remains unsolved. our closing keynote @headinthebox introduces the main motivations behind Automind and the Universalis interpreter - agents that carry their own verifiable proof of safety! link to talk below

@headinthebox @AvimanyuRoy3 oh wow ok linking and publishing now. didnt see that you had started talking about it!!

@AvimanyuRoy3 x.com/aiDotEngineer/…

@abshkbh @gdb 🆕From fork() to Fleet: Designing an Agent Sandbox Cloud youtube.com/watch?v=OqM67Q…

🆕From fork() to Fleet: Designing an Agent Sandbox Cloud youtube.com/watch?v=OqM67Q… @abshkbh spoke at AIEWF Online last year about Arrakis, his open source MicroVM-based secure sandboxes. @gdb hired him immediately for his expertise and passion - and now, after a year at OpenAI and recent release of @ChatGPTApp Work, he makes his onstage debut covering the three pillars of agent cloud engineering - Runtime, Persistence, and Orchestration - and ends with a surprising conclusion on why storage and filesystems are an integral part of agent clouds!

@willccbb 😅 we will have to give a five timer medal soon

@aiDotEngineer thanks for having me back!! four-time now 😉

Congrats to PI on the unicorn round and $100M ARR! we were proud to have @willccbb introduce verifiers at the first AIE NYC a year ago and now... it is v1! Will joins a rare list of three-time AIE speakers, and his talk on the full PI stack is linked below!

Key Components of the @PrimeIntellect Stack: youtube.com/watch?v=V-EDrh… Environment & Verification (0:50): Prime Intellect has overhauled their Verifiers library to create a modular, task-based system. It decomposes environments into task sets (data/rules), harnesses (agent logic), and runtimes (execution environments), allowing for greater flexibility when evaluating models. Asynchronous RL Framework (PrimeRL) (29:20): The core training framework is built to be fully asynchronous. By decoupling training from inference, they can handle long-horizon agentic rollouts without stalling the entire system, making it more efficient for complex coding tasks. Modern Post-Training Algorithms (38:00): The stack is designed to support a wide array of training recipes, including Supervised Fine-Tuning (SFT), On-Policy Distillation, and Self-Distillation. The system abstracts the loss functions and algorithms, allowing users to swap methods without changing their infrastructure. The Lab Platform (42:35): Prime Intellect offers a hosted platform that handles the underlying compute, providing multi-tenant LoRA training today and full fine-tuning capabilities soon, while keeping the developer experience consistent from local prototyping to cloud-scale runs.