Tag
The author argues that a good tool description for AI agents should answer five key questions — when to use it, what each input means, what success looks like, all possible errors, and permission rules — since most tools only address one of the five, forcing the agent to guess the rest.
Built riviera.so, a tool that lets developers edit the tool and parameter descriptions that AI agents read via OpenAPI or MCP servers, without redeploying the backend. It supports draft/publish versioning, failure investigation, and turning agent mistakes into tool-selection tests.
The author shares four mechanical failures experienced while running AI agents on a cron job in production for a month, emphasizing that all issues stemmed from setup problems rather than model deficiencies.