CODE

This is the way. Don't share apps, share ideas with AI Agents. Karpathy just introduced a new primitive: "the idea file". Instead of sharing a repo with code to clone, you share a markdown doc with the idea. What to build, not how to build it. Your agent reads it and figures out the rest. I pointed my OpenClaw agent Monica to it. Within one session, they read the idea file, compared it against our existing setup (6 agents already coordinating through markdown files on a Mac Mini), identified what we're already doing and what's missing, and started building the parts we don't have. Turns out we already had the ingestion layer without knowing it. Our intel agent scans sources twice a day and writes structured signals to a daily file. The raw data gets saved, but nobody ever looks at it again. What we were missing: compilation. 60 days of daily signals sitting in files, but no agent turning them into structured knowledge. My agents only see today's intel. They can't say "this is the third local OCR tool this quarter" because that context isn't compiled anywhere. Monica caught that gap on her own from the idea file. You don't need someone else's code. You need their thinking. Your agents handle the rest, and they'll customize it to what you actually need. We are moving from cloning repos to sharing ideas with Agents. Shubham Saboo (@Saboo_Shubham_) x.com/i/article/202179384677… — https://nitter.net/Saboo_Shubham_/status/2022014147450614038#m

→ View original post on X — @saboo_shubham_, 2026-04-04 17:57 UTC