Skip to content

Skill Engineering: Targeted Control Instead of One-Shot Design with AI Agents

In a nutshell: Skill Engineering defines design commands like “bolder” or “quieter” for agents through concrete operational parameters instead of vague instructions to increase quality and consistency.

Paul Bakaus has developed Impeccable, a system that gives AI agents a precise design language instead of letting them blindly redesign an entire website. His approach demonstrates that agents need not creativity without limits, but rather domain expertise vocabulary and human control.

Bakaus, creator of the open-source system Impeccable, argues against the widespread notion of one-shot design with AI. Rather than having an agent reshape an entire interface at once, users should guide it to make individual areas “bolder,” “quieter,” or “denser.” Behind this lies a fundamental conviction: AI products do not need less human intervention, but rather more structure in how it is applied.

Impeccable extends Anthropic’s original frontend design capabilities into a more complex system with multiple components. Bakaus describes the resulting field of “Skill Engineering” as a standalone discipline with “dark arts.” A central problem: most skills converge in the same direction. If all developers use the same design skills, the result looks identical everywhere. Moreover, different agent platforms such as Claude, Cursor, or GitHub Copilot handle sub-agents and permissions differently; a skill cannot simply treat them all the same way.

The core idea of Impeccable is to translate designer concepts into operational precision. An uncontrolled model might interpret “bolder” through gradients or glass effects; Impeccable instead defines it through hierarchy, scaling, and decisive typography — changes that direct attention without breaking the existing design system. Bakaus puts it this way: “An adjective without meaning behind it is just a platitude. You have to tell the agent what you mean.”

Bakaus has observed that designers and engineers using the same model produce completely different results because designers have the right vocabulary. The skill compresses this expert language into a system so that even non-specialists can express precisely what they want. However, Bakaus acknowledges that not every aspect of design can be squeezed into this abstraction — directly manipulating spacing can still be the fastest way for minor adjustments.


Source: www.latent.space · Published July 2, 2026
Lumi AI News — AI-assisted curation pursuant to Article 50 EU AI Act. Paraphrasing and classification via Lumi News Pipeline v1.7.2.

Share on: