Why this repo matters
Latest release 5d ago, 7 developer signals, 2 package/install signals

Gradient descent, but the parameters are agents — a parallel, asynchronous framework for self-evolving agents (skills, prompts, harnesses). Diffs are the gradients; the aggregator is the optimizer. (189 stars, 18 forks, Python, fresh release, 6 AI signals, 6 developer signals). Latest release: v0.5.1.

Repo analysis

AgentDescent is a Python framework for building self-evolving AI agents using a parallel, asynchronous approach inspired by gradient descent. It allows agents to develop their skills, prompts, and harnesses over time, with differences acting as gradients and an aggregator functioning as the optimizer. The project recently released version v0.5.1.