GenAIClosing soon
Ship an agent that books a real workflow
Three tools, one messy task, and a scorer that fails you the moment your agent invents a tool call.
- Time needed
- 1 week
- Tools
- 3 mock APIs
The brief
The agent has to complete a multi-step request using only the tools provided, recover from a deliberately flaky endpoint, and stop rather than guess when a required field is missing. You are scored by an automated harness plus a mentor read of your trace logs.
What you submit
- Your agent implementation with tool definitions
- Trace logs for the provided task set
- A note on how you handled the flaky endpoint and missing fields
How to participate
- Register free: create your account and tell us a little about yourself.
- Clone the harness repo with the three mock tools.
- Build your agent and run the local scoring script.
- Submit the repo; mentors review your traces and publish feedback.
How it’s judged
- Task completion rate
- No invented tool calls
- Graceful failure behaviour
What you walk away with
Feedback for every entry · Top 3 win up to 75% off a flagship seat
Resources
Everything you need, in one zipThe dataset, the full brief and the guides for this challenge.Register free to unlock