04/12/2026 // AI Research

Vibe Jam 2026

Cursor's game jam requires 90% AI written code. That constraint makes it a useful benchmark for what agentic development can and can't do right now.

Cursor's Vibe Jam is back for 2026. The format: build a web playable game in 30 days where at least 90% of the code is AI generated. The prize pool doubled from last year to $35,000, split across gold ($20k), silver ($10k), and bronze ($5k). Pieter Levels organizes it, with Bolt.new and Glif as additional sponsors.

The jam opened April 1 and runs until May 1. There are already 121 games submitted with two weeks left. Last year had over 1,000 entries from 50+ countries. The rules are simple: one entry per person, web playable, no login required, free to play. Any engine works, Three.js recommended for 3D.

From a research perspective, this is one of the more interesting experiments happening right now. The 90% AI code requirement forces participants past the autocomplete stage into genuine agentic development. You're not getting tab completions. You're handing an agent a feature and seeing what comes back. That's a different problem than most developers have actually dealt with, and a game jam compresses it into a timeline where the architectural consequences show up fast.

I've been building games and real time systems for years. The architecture decisions in a multiplayer server or a physics simulation are the kind that punish you if you get them wrong early. What I want to know is whether AI assisted workflows change those decisions or just change who types them. So far from what I've seen, the answer is mixed. Agents are good at producing plausible code quickly. They are bad at knowing when the structure underneath needs to change.

There's a new portal webring this year that lets players teleport between submitted games with live player counts and stats. That turns the jam into something browsable rather than just a list of links, which should make it easier to study what people actually built.

I plan to dig into the submissions after the deadline and write up what I find. The interesting question is not which game looks best. It's which codebases held together under the pressure of real development, and what the developers did differently to make that happen.

  • Vibe Jam
  • AI Research
  • Game Dev