Paul Bakaus proposes a middle ground between adjusting individual pixels and asking an agent to design an entire page autonomouslyFront-end code generation uses AI to create or modify the user-interface code of websites and applications from instructions or examples.. His central claim is that useful design requires knowledge of the audience, intended emotional effect, references and constraintsContext engineering designs the information, instructions, memory, and tool state an AI receives so it can perform a task reliably., followed by iteration rather than a single prompt. He presents this as a way to keep human judgment active as design and engineering workflows converge.
Paul Bakaus uses Impeccable as a case study in encoding design knowledge into reusable agent skillsAn AI agent skill is a reusable package of instructions, resources, and tool guidance for performing a bounded kind of work.. Commands such as bolder, quieter, distill, harden and polish should carry project-specific meaning. For example, making a section bolder should adjust hierarchy, scale and decisive typography while respecting the design system, rather than introduce arbitrary gradients or colors.
Paul Bakaus maps the design workflow from initial shaping through crafting, revision, hardening and cleanup. He argues that agents need different levels of direction at different stages, while people still handle important choices and final refinement. A described comparison and shader example illustrate his approach, but this transcript review does not independently assess the visual quality of those demonstrations.
Paul Bakaus rejects a fully automatic mode because choosing the direction is the purpose of the interface. He argues that tools can amplify a person’s craft and help sharpen taste, while taste remains contextual and cultural. The practical lesson is to define what a command means for the project and use it to guide a continuing design process.
Watch on YouTube
