▌ GitHub radar
Agent Apprenticeship: a school where AI agents learn on the job
A new tool that lets AI agents practice on real tasks, save what they learned as reusable lessons, and pull in experience that other people's agents already figured out. Think of it as on-the-job training plus a shared notebook for agents.
Agent Apprenticeship is a command-line tool and shared ecosystem built around one idea: let AI agents learn from real work the way a junior employee learns on the job. You run your agent through a library of hundreds of curated practice tasks, and it turns each run into a reusable lesson and a full execution trace — a record of what it tried, what worked, and what to do differently. It plugs into the popular agent setups (Claude Code, Cursor, Codex and others) and offers different mentor modes, from fully model-assisted to expert-led. The twist that made it spread is the shared layer: lessons and experience packs can be searched and traded, so agents benefit from what other people's agents already figured out.
Why a vibe-coder should care
Most people's AI agents are amnesiacs — they nail a task today and forget the lesson by tomorrow. This is a serious attempt to fix that, giving your agent a memory of hard-won lessons plus a way to borrow other agents' experience instead of relearning everything alone. If you lean on coding agents day to day, it points at a future where your tools quietly get better the more they're used.
More finds