🛠 Build

From a scored, researched, briefed idea, the builder writes the PRD — product name, core loop, features with acceptance criteria, data model, and the landing page as part of the product, not an afterthought. Then it one-shots the whole app: hosted at its own URL with signup, auth, hash-routed views, full CRUD against a real data API, tracking beacons wired into every step of the funnel, and payments-ready product gates.

After it's live, you never 'develop' again — you talk. "Move the paywall to right after first end-use." "Tighten the mobile hero." Chat edits apply surgically to the running app. Improvement sprints run bigger passes; agent-mode adds a merciless review pass that finds dead buttons and broken flows and fixes them before you ever see them.

When an app has been refined for weeks — conversion-tuned hero, mobile QA'd flows, monetization choreography that actually converts — that work shouldn't die with one product. Template builds seed a new app from a proven app's actual code: every file cloned, every slug reference swapped, products carried over.

Then the transform stage rewrites only the domain content — the words, the offers, the examples — while the machinery, design system, and funnel choreography stay byte-for-byte intact. Weeks of tuning, inherited in minutes.

Flip staging on for an app and every edit — chat, sprint, auto-repair, even external mirror pushes — lands in a staged copy while production keeps serving untouched. Read per-file diffs, preview privately (there's a shareable phone-preview link that needs no login), then promote exactly the changed files, or discard the lot.

Separately, every push to production — promotes, finished builds, sprints, mirror imports, manual saves — freezes a numbered version: the complete file set, browsable as a live page with a version ribbon, restorable in one click. A restore auto-saves current prod first, so even the undo has an undo.

Every build ends with smoke suites — hard checks committed alongside the app so they run on every future change too. Above that sits the walkthrough tester: an AI that drives the real app in a real frame like a real user, reading each screen and deciding its own next move toward a goal — it signs up when it hits a login gate, dismisses overlays, and adapts instead of replaying a script.

SEO audits score the live pages and push their fixes straight into the builder. And the A/B layer runs landing-page experiments (sticky 50/50 assignment, signups attributed to the variant actually seen) and in-app experiments on signed-in users — with archived variants kept re-runnable.