Generate vs Govern

AI design tools are no longer an argument. Tools like Claude Design, v0, Wonder, Lovable, and Figma Make work, ship, and come bundled into the suites your team already pays for. Prompt-to-prototype is settled.

The unsettled question is what happens after generation.

Every generated interface eventually lands in a repository. That repository has conventions: a token system, a component library, a set of patterns people agreed on, and a longer list of things they agreed not to do. None of that is enforced by the tool that generated the code. It's enforced — or not enforced — by whoever reviews the PR.

That's the gap.

Generation is a prompting problem. Governance is an infrastructure problem.

A generator can read a team's colors and typography and apply them to a new prototype. That's useful. It is not a design system. A design system is the thing that says this button never has that radius, this icon never pairs with that label, this pattern replaced the one you're about to re-invent last quarter. Those aren't style defaults. They're rules with history, identity, and enforcement.

preset AI is the governance half of the stack

• Artifact identity — every token, pattern, and preset has a stable ID that survives renames, refactors, and library migrations.
• Forbidden primitives — the rules your team already argued over, encoded as validation.
• Compliance scoring — surface a single score on every PR so reviewers know at a glance whether it conforms.
• Deprecation that sticks — search hides deprecated artifacts by default; AI agents stop suggesting them automatically.
• Upstream reconciliation — when upstream changes, downstream converges without hand-editing.

None of that is visible in a prompt. All of it shows up in the diff.

The handoff is where the two halves meet

Claude Design hands a bundle to Claude Code. v0 hands a snippet to your editor. Wonder hands React + Tailwind to your agent. At every one of those seams, someone eventually commits the output to a real codebase. That commit is where preset AI fires: validate_code, preset_compliance_score, get_forbidden_primitives. Violations become PR comments, not production bugs.

Generate with whichever tool your team likes. Then govern what it produces.

preset AI runs where the code lives. MCP for your coding agent. GitHub Action for your CI. PR surface for your reviewer. One source of truth, everywhere it matters.