Your LLM is Both Partner and Stakeholder, and That Changes How You Design

Bryan Elkus of the consultancy Atomic Object argues that designers building with an AI model should refine one piece at a time and check it against the product.

Bryan Elkus leads the software design practice at Atomic Object, a software consultancy in Ann Arbor. He describes how working with an AI model changes each stage of a design project, from early research to the shipped product. In his account, the AI model is both a partner that generates ideas and a stakeholder whose output the designer must check and correct.

• In early research he talks a problem through with an AI model by voice, starting from a short brief and copying what came out back into the project's shared notes.
• To shape screens, he moves rough concepts written by the AI model into Paper, a design canvas. He rearranges them there and gives the result back as the next instruction.
• Once the team builds in the live codebase, the prototype becomes the product, and he finds that a separate Figma file drifts from what ships.
• He warns that the AI model fills each gap in a request with a plausible default. He changes one piece of the interface at a time and checks it against the current product before merging it.

A design team building in the live codebase with an AI model can work in small, separate changes and check each against what it knows about its users. This is an opinion essay from the author's practice, without measured results.

Derived by Working Surface from the article; more in the Playbook. Source line: Before anything goes back in, check it against the product as it stands, and against the user you set out to serve.

29 October 2026: as AI agents raise the volume of work to check, teams move the check closer to the work: fixed rules for what agents may fix, design rules written where agents read them, reviewers trained on how AI-written code goes wrong, and leaders who do the work and are reviewed in public.