Every time I attend a conference like AI Engineer Code Summit, I wish I had a space to try what I learned. Today’s hackathon is meant to be that space for each of you. It’s an opportunity to experiment and play with everything you’ve absorbed. Whether you attended in person or watched the livestreams, we all spent the last three days hearing from the industry’s best AI practitioners and engineers. Now you get to put those ideas to work. I’m jealous. If I weren’t helping run the hackathon, I would be where you are. Use the time to build something awesome.
AI Engineer Code Summit had three core themes this year: building with System One models, fighting slop in software factories, and rewriting critical apps. TypeSafe’s Jev catapulted System One (decision models) into the limelight. If you’re feeling out of the loop, Jev launched only two months ago, on September 15th, and with how fast our industry moves, that may already feel like years ago. Unlike typical chat or reasoning LLMs, these models skip chain-of-thought generation to return fast, calibrated, strictly typed decisions. Other labs have quickly followed TypeSafe’s success with decision models of their own. If you haven’t used these yet, they’re my personal recommendation for your hack. There are so many cool, unexplored use cases.
Beyond System One models, the conference aimed to reduce slop in software factories. Dark and light software factories are popular terms to describe how humans interact with these automated pipelines. In a dark factory, the machine is designed to run itself: you minimally review intermediate code because you front-load inputs, like specs and repo rules, and automate output verification. Light factories keep humans more regularly in the loop. Lauren “poteto” Tan from SpaceXSI had a fantastic keynote on Wednesday about setting your codebase up for success. Lauren famously ships thousands of PRs a month with her pstack system. Even with limited time at the hackathon, setting up your repo, agent skills, and codebase upfront pays dividends over jumping straight into development.
Initially, I feared rewriting critical apps was the least relevant theme for today’s hackathon, but thinking about it more, I disagree. We have superpowers right now. We can build performant apps from the get-go. Last week, Discord announced a native rewrite of their application in Rust. Before the current AI cycle, I would never have reached for Rust to build a hackathon project. Today, agents are quite capable of writing code in any language. Lower-level, efficient languages are no longer a bottleneck to shipping fast, making today a great opportunity to experiment with an unfamiliar language or framework.
I hope you use today to experiment and apply new knowledge from AI Engineer Code Summit. Hearing from our industry’s best developers makes a trip to any AI Engineer conference worthwhile, though I’ve always struggled with the follow-up. I leave a conference with 100 great ideas and no time to make them. Take one of yours and make something today.
Top comments (0)