Paul Bakaus, creator of the open-source design skill Impeccable, told attendees at the AI Engineer World’s Fair that "skill engineering" can increase the practical capabilities of AI agents — while explicitly rejecting the idea that people should be removed from the creative process.
What Impeccable does and why it differs
Impeccable maps designer-friendly adjectives — for example “bolder,” “quieter,” “denser,” or “more polished” — to concrete operational instructions for coding agents. Rather than asking an agent to redesign an entire page in one pass, the system lets users apply incremental, targeted changes to specific sections.
Bakaus said the approach reflects a broader view: AI products need more than crude prompts. Agents require domain knowledge, contextual cues and clearly defined points where humans can steer outcomes. “The point is to give you a way to steer what you want to end up with,” he said during his session. “It’s never going to be a tool for one-shot design. That’s not the intent.”
Skill engineering as a discipline
Impeccable began as a modest extension of Anthropic’s frontend-design skill. As its user base expanded, Bakaus turned it into a multi-component system with workflows and routing logic, and came to regard skill engineering as a discipline in its own right.
In his workshop he discussed what he called the “dark arts” of building skills: many skills and models are not very creative and tend to converge on similar outputs, so if everyone uses the same skill for frontend work the results become homogenized. Skill engineers must also navigate differences between agent harnesses and models: Codex, Claude, Cursor and GitHub Copilot do not all expose the same subagent, permission or execution semantics, so a cross-platform skill cannot assume identical capabilities.
Bakaus has experimented with routing inside a skill to combine capabilities and direct subtasks to the most relevant instructions. He compared this to a mixture-of-experts pattern, where routing both saves tokens and increases effectiveness.
Giving agents a design vocabulary
Impeccable’s central innovation is translating design language into precise, actionable meanings. An unassisted model told to make a page “bolder” might add gradients, neon effects or glassy surfaces. Impeccable instead defines boldness with concepts like hierarchy, scale and decisive typography — changes that attract attention without necessarily breaking an existing design system.
"An adjective with nothing behind it is just a nice apostrophe," Bakaus said. The model has some innate sense of what words like “bold” or “quiet” mean, but the skill imbues them with professional, domain-specific meaning. Experts possess vocabulary non-experts lack, and Bakaus observed large differences between work produced by a designer and an engineer using the same model simply because the designer could articulate the desired outcome. “I’ve been trying to put that language — basically compress it into a skill and into a system — to be able to express yourselves better,” he said.
He also acknowledged that not every design change belongs at that level of abstraction. Direct manipulation of spacing may remain the fastest way to make a small adjustment, and open-ended prompting can help during initial exploration. The goal is not replacing every tool with an agent but finding the "exact level of control" where human judgment is most valuable.
Roles shifting up the stack
Bakaus sees boundaries between design, engineering and product management blurring. “Designers are moving into code, engineers are moving into design, and vice versa,” he said. That collision creates pressure: engineers who mainly translate Figma designs into code face increasing automation, and designers whose contribution is only making an interface look competent face similar risks.
"Designers all have to move one layer up the stack to think more about the what," he said. "I think the role of the product manager and designer is actually converging."
At the same time, designers are approaching implementation. Bakaus initially expected Impeccable to attract mostly engineers and thought professional designers might resent its emphasis on code, but he now estimates designers make up at least half of Impeccable’s users. "So rather than moving directly into code and, you know, having no help," he said, designers use Impeccable as a bridge because it communicates the way they communicate — a result he did not expect when he first built it.
Impeccable also offers a live mode that combines visual selection with a background coding agent. A user can select a section inside a development environment and request multiple alternative layouts or ask for a bolder or quieter treatment. The system operates within the project’s existing codebase and design system instead of exporting an isolated mockup from a third‑party design tool. Bakaus described this as a potential "design harness" at the intersection of chat and direct visual manipulation.
No auto mode — humans keep the final say
While parts of the industry treat full automation as the natural end state, Bakaus rejects that premise. He identifies two camps: defenders of the traditional Figma-centered workflow, and proponents of "loopmaxxing" who want agents to run with minimal human intervention. "The truth is somewhere in the middle," he said.
His preferred model is for AI to deliver the first 80% quickly — a competent layout and baseline implementation that would otherwise take substantial time — while the person retains the final 20%, where taste, context and a distinctive point of view matter. "People need purpose, and they want to play a role in whatever they create," Bakaus said. "When you work with the agent, then you feel more ownership of the product."
Users often ask for an automatic mode that would let the system pick commands on its own; Bakaus has no plans to add one. "There is no auto," he said, "and there will be no auto." When asked about visions of "software factories" or other models that appear to remove people from engineering entirely, his response was unambiguous: "I’m squarely against that."
Conclusion
Bakaus advocates a middle path: treat skill engineering as a craft that supplies agents with a precise design vocabulary, routing and composition logic to improve efficiency, and clearly defined human control points. The approach aims to speed up routine work while keeping humans in the loop for the final, taste-driven decisions, and it explicitly rejects fully automated replacement of designers and engineers.
