AutoAgent

We’ve all been there. Write a system prompt, run the agent, squint at failure traces, tweak the prompt, add a tool, repeat. It’s the kind of grunt work that makes you question your life choices.

AutoAgent automates this entire process. Point it at your agent and a benchmark, walk away, come back to find it’s tested hundreds of configurations and optimised everything from prompts to tool selection. No more manual iteration loops.

The library treats agent optimisation as a search problem. It systematically explores the space of possible configurations, learns from failures, and converges on better-performing setups. Think of it as a meta-agent that builds other agents.

This is exactly the kind of tooling the agent development ecosystem needs. We spend too much time on mechanical optimisation tasks that could be automated. AutoAgent lets us focus on the interesting problems whilst the library handles the tedious bits.