Guillaume Laforge @glaforge.dev ยท 1d #AI coding agents waste a lot of time and tokens repeating the exact same trial-and-error mistakes across sessions. To fix this, I created theย ๐ข๐ฝ๐ฒ๐ป ๐ฅ๐ฒ๐ฎ๐๐ผ๐ป๐ถ๐ป๐ด ๐๐ผ๐ฟ๐บ๐ฎ๐ (ORF) (a file-based specification that lets AI agents learn from past runs)
27 likes 5 replies
Replies
That repeated mistake loop is exactly the annoying part of coding agentsโฆ writing the lesson to a file so the next run can reuse it makes a lot of sense.
Nice work. Is this specific to Antigravity or can be used with other ai-assisted coding tools? Was wondering, don't we need to mention the experience folder and its structure in agents.md or custom agent files? so that it refers to it when required? Also, how will this work with alongside ADRs?
Session amnesia is the real tax on agentic coding right now. Curious how ORF handles conflicting lessons across projects though, since one repo's fix can be another repo's anti-pattern.
Very interesting approach! I have a very large ongoing FAILS.md for Claude (in global user settings) and now Iโm temped to let an agent parse it into experiences so that I donโt start from zero ๐ค
๐ Article:ย glaforge.dev/posts/2026/...โจ ๐ป ORF optimizes agent trajectories: when an agent resolves a tricky bug or trap in one run, it records the lesson as a structured playbook.