What “AI-native” looks like when it isn’t a demo
The actual loop: research in conversation, briefs as source of truth, staged sign-offs, code last.
Most of what gets called "AI-powered design" is a party trick bolted onto an unchanged process: generate some options, cherry-pick one, carry on exactly as before. The tell is that if you removed the AI, the process would be identical, just slower at one step. That's AI-flavoured, not AI-native.
AI-native means the process itself is built around the collaboration. Remove the AI and the process doesn't slow down, it stops existing. After running every project this way for over a year, including the site you're reading, here is what that actually looks like day to day.
Stage one: research happens in conversation
Market analysis, competitor teardowns, user constraints, technical feasibility: all of it happens as a long, argumentative conversation before anything is drawn. The key word is argumentative. The value isn't asking for answers; it's having a tireless counterparty that will stress-test a position at 1am, surface the competitor you hadn't considered, and force you to write down WHY you believe something.
For this portfolio, that meant studying reference sites live (not screenshots, the actual running sites) and pulling apart what made each one work before a single decision was made about mine.
Stage two: findings become briefs, and briefs become law
This is the stage most people skip, and it's the one that makes everything else work. Research gets written into structured briefs: the positioning, the constraints, the decisions already made and the reasons they were made. From then on, the brief is the single source of truth every later stage answers to.
The discipline this enforces is the point: if a design decision can't trace back to a finding in the brief, it's taste masquerading as strategy, and it gets challenged. Taste arguments end at the brief instead of dragging through endless revision rounds. The brief is effectively the API between the stages, and between me and the AI.
Stage three: design against the brief, with gates
Design proceeds system-first (tokens, type, components before screens) against the brief, with an explicit sign-off gate at every stage. Nothing advances on momentum. The gates matter because speed is the failure mode of AI-assisted work: it is now trivially easy to produce forty screens of confident-looking wrongness. The gates are where human judgment does its actual job.
Stage four: built, not handed off
The design goes to production code with no handoff document, because the same collaboration that designed it builds it. The handoff gap (that lossy translation layer where padding drifts and interactions get "interpreted") simply doesn't exist. Milo, the robot who assembled himself as you scrolled in, went from sketch to rigged, expressive production code without a single redline.
That's also the honest limitation: this loop currently fits a designer who codes or works solo-plus-AI. Scaling it to a large org is an open problem, and anyone telling you otherwise is selling something.
Why this is worth writing down
The counterintuitive part is that AI-native process is MORE disciplined than the traditional one, not less. Every stage produces a written artifact. Every decision has a traceable reason. The AI never gets tired of being asked why. If your AI process has fewer documents and fewer checkpoints than your old one, you don't have an AI process. You have a slot machine.