Method
Supervised AI. A human makes the calls.
The bottleneck is no longer writing code. It is specifying clearly, testing and deciding. That is the core of product work.
Phases
Four phases. Short cycles. You approve every step.
01Frame
Interview, goals, audience, pages, content and constraints.
Deliverable
- Brief
You approve
- Scope and priorities
02Prototype
I turn the brief into a clickable prototype.
Deliverable
- Prototype
You approve
- Flows and look
03Iterate
Short rounds of feedback and adjustments.
Deliverable
- Versions and a changelog
You approve
- Each version
04Build and ship
Production code, tests, deployment and handover.
Deliverable
- Live site, code and docs
You approve
- Final acceptance
AI
Where AI helps, and where it does not
AI is good at
- Generating pages and components from a clear spec
- Writing the tests that prove nothing is broken
- Suggesting fixes, which I review before merging
A human stays in charge of
- Architecture choices
- Payments, personal data and security
- Anything without a clear success criterion
Measure
Decision metrics, not vanity metrics.
Success criteria are agreed in the Frame phase. After launch, product analytics show whether the product meets them, and what to change next.
Vanity metrics
Look good in a report, change no decision.
- Page views
- Sign-ups
- Downloads
- Time on site
Decision metrics
Tied to the goal in the brief. Each one can trigger a change.
- Activation: users who reach the first real value
- Retention: users who come back
- Task completion on the key journeys
- Conversion to the goal set in the brief
Instrumented with PostHog: events, funnels and retention.
Four guardrails
- 01
A written, approved spec before any generated code.
- 02
Success criteria agreed with you before building.
- 03
Human review and tests before anything goes live. The AI never touches production or secrets.
- 04
A human on every irreversible decision: payments, data, domain name. Scope capped per phase.
Ready for phase one? It starts with a short brief.