Agent reliability · human and agent collaboration

An agent's claim that it finished is worthless unless the proof could have failed.

I build tools that make agent work inspectable: shared rooms where people and agents edit the same state, runtimes that record receipts, and configurable gates that check declared completion criteria.