Latent Space
8/10 signal
Skill engineering and the case against one-shot AI design
agentictool-use
What happened
Paul Bakaus introduces "skill engineering" via his open-source project, Impeccable. Instead of asking coding agents to redesign interfaces in one shot, skill engineering provides agents with a structured vocabulary of design skills (e.g., making elements "bolder" or "denser"). This approach emphasizes steerability, domain-specific context, and modular capabilities over raw, unguided generation, while addressing model-specific differences in handling subagents and permissions.
Why it matters
Bounding agent actions to a structured "skills" vocabulary is far more reliable and steerable than relying on one-shot generation.
The take
This is a massive shift from "prompt-and-pray" to structured API-like capabilities for agents. By giving agents a bounded, semantic vocabulary of actions (skills), we get predictable, steerable, and high-quality outputs. It's a great blueprint for anyone building domain-specific agents (not just for design).
Do this
Check out the Impeccable open-source repository to understand how to design modular, semantic skill systems for your own domain-specific agents.
Don't read this site daily. Get it in your inbox.
The daily brief and Sunday deep dive — distilled, scored, and opinionated. For builders only.