Agents
I build agents as small, observable steps with clear limits on what each tool may do, so every run can be traced, capped and stopped.
What I do with Agents
- Tool calls with explicit permissions
- Step and cost limits per run
- A log line for every step
Projects that use it
No project write-up on this site names Agents yet.
Need Agents done properly?
Send me your prototype’s link. I read every message the same day, then we talk for twenty minutes about what breaks first.