The Visual Refinement Gap: Why AI-Generated Code Needs a New Kind of Tool
AI gets you 80% of the way there. The last 20% — visual polish — is where you actually live. Here's why existing tools don't solve it, and what visual refactoring means.
Blog
Thoughts on visual editing, design engineering, and building with AI.
AI gets you 80% of the way there. The last 20% — visual polish — is where you actually live. Here's why existing tools don't solve it, and what visual refactoring means.
The story of how years of switching between Figma, VS Code, and browser DevTools led to building a visual editor that treats your codebase as the design tool it already is.
The idea that a product could be built in a week used to be a cliché. It's now a real number. Teams that treat prototyping as the central discipline — not a preamble to it — are winning.
For a decade, specialization paid. The designer-developer was a curiosity. That balance is inverting. AI is making generalists valuable again — and it's making the hybrid role the highest-leverage one on many product teams.
Generative tools have collapsed the cost of producing plausible design. They have not collapsed the cost of producing good design. The difference is taste, and it remains defensible.
When the source of truth for a design system lives in a design tool, it drifts. When it lives in the codebase, it compounds. This is a quiet but consequential shift.
There's a perverse truth about AI-generated work: nearly-correct output is harder to fix than blank-slate output. Here is why, and what to do about it.
Great software isn't made by one big idea. It's made by a few hundred tiny decisions nobody mentions in case studies — and each of them is individually unglamorous.
Modern product work fragments across Figma, IDE, browser, Slack, Linear, and more. The tax this imposes is worse than most teams realize — and the fix is not better tools, it is fewer.
For twenty years the pipeline ran design first, code second. AI-assisted development has flipped the arrows. The handoff as we know it does not survive that flip.
AI tools have shifted the work. Most of a developer's day is now archaeology, not authorship. Here is how to get good at it.
The rendered application is the highest-fidelity design artifact you will ever have. Everything else is a guess at what it might look like.