AI Coding

Extended and enhanced workflow with Claude Code.
Integrating AI into real development workflows — not as a gimmick, but with the same strict precision I apply to design systems.

Claude Code in practice

I work extensively with Claude Code to accelerate development, from scaffolding components to refactoring entire modules. It runs alongside my editor as a collaborator that understands the codebase, follows conventions, and executes the tedious parts so I can focus on architecture and intent.

AI as a critical reviewer

Beyond generation, I use AI as a critical reviewer: auditing code for accessibility gaps, inconsistent naming conventions, code conventions, design token usage, and optimization opportunities. The value isn't in replacing thinking — it's in getting honest, tireless feedback that improves the output.

Design System guardrails

AI output is only as good as the rails around it. I lean on Design System tokens, component contracts, and TypeScript to keep generated code on-brand and on-spec. That combination turns AI from a source of UI drift into a tool that actively reinforces consistency across teams.

Code guidelines and conventions

I write code guidelines and conventions that make a codebase stricter: less room for assumptions and better readability, for the people and the agents working in it. When the rules are written down, a reviewer, a new team member, and an AI agent all start from the same expectations.

Refactors and migrations

Large refactors — API renames, token migrations, component version bumps — are where AI pays off most. I use it to apply repetitive changes across the codebase predictably, while I review diffs, validate tests, and catch the edge cases that need a human call.

Coordinated agents

I use AI to improve productivity with coordinated agents. Each agent has one job, such as sharpening tickets, building a feature, or auditing the system, and they work in parallel on the same Jira board. The result is more work in less time, and every change still goes through my review.

Skills, conventions, and routines

On Seven Facets, my own design system, I wrote my own Claude Skills (build a component, build an icon, run an audit, turn a brand into a theme) and MDX convention files with the rules for tokens, accessibility, i18n, RTL, and theming. Since April 2026, Claude Code routines pick up planned Jira tickets overnight, open a PR, and run the audits. A nightly audit checks the whole system and files new tickets. I review every PR before it is merged.

Workflow

Prompts, project context files, and clear conventions make AI dependable. I set up repo-level instructions, automated checks, and review steps so AI-assisted work ships with the same quality bar as anything else — type-checked, linted, tested, and consistent with the Design System.

Curious how AI fits your workflow?

I can help your team adopt AI-assisted development without losing the precision your Design System depends on.
Let's talk about where it fits — and where it doesn't.

Contact me