Excited to share a new mechanism design challenge from me and @danrobinson Can you build the best AMM fee strategy that thrives in a dynamic trading environment? Give it a shot and submit at the link below 1/

This challenge is designed so that the user can have a quick iterative experimental loop with AI coding tools. The AI coding tools on their own do not score very well, but with a few key insights from the user they can unlock large score jumps 2/

After attempting this puzzle for a few days I think it tracks intuition for real world AMM design pretty well. A few of the ideas for improving real AMMs seem to work well in the competition Rough score thresholds 450+: Not totally slop 480+: Solid 510+: Strong 540+: ??? 3/

This problem was what sparked my appreciation for codex. I was totally unable to make progress on this problem with Opus. If you find something different I would love to hear about it

The next generation of breakthroughs in difficult applied domains will likely come from setups like this. A well designed simulation loop and target metric that an agent and a human can iterate on together. This structure makes the process of discovery and validation much faster