Field Note · Practice
Against the Fake Demo
Why this site ships negative results and unfinished prototypes.
Key Takeaway
Prioritize publishing negative results and unfinished edges in agent research to accelerate community learning, exposing the real blockers over cherry-picked highlight reels.
It is easy to record a cherry-picked run and call it a capability. A demo shows the best case; it is an existence proof under ideal conditions, and it tells you almost nothing about the typical one. I would rather publish the negative result, the system that works but has not been run as an experiment yet, the prototype that points at an idea it does not yet fully implement.
The honest version is less impressive and far more useful — to me, because it tells me where the real blocker is, and to anyone deciding whether an approach is worth their time. A failure log ages better than a highlight reel. If the work is real, the unfinished edges are part of what makes it worth showing.