How I Work

AI Executes the Code.I Drive the Decisions.

I build products from architecture locked in writing, not from vibe coding that vanishes the moment the chat moves on. Leadership and architectural calls stay with me; AI executes in parallel.

How I work

Documents lock first
Business, UX, and architecture become baseline before code. Material change goes through a new ADR, not “just tweak it in chat.”
Structured ADRs
Every material call records Decision, Reason, and Alternatives. Old plans are superseded, not erased.
Parallel AI, human lead
I break down the problem, design, and review. Subagents execute tasks in parallel under that control.
Truth is what is live
After release, production code is the visual and functional benchmark. The next pass starts from what people actually use, plus data.

Operating system

Five principles I will not compromise

STRATEGY

What we build, then how we build it

Product and process stay separate. “What gets made” locks first as baseline: business, UX, architecture. Code follows the locked documents, not the other way around.

SCOPE

Honest small scope, not a complete one

MVP is the thinnest surface that still tells the story (MoSCoW / Hybrid lean). Scope changes go through a new ADR, not a casual “change it in chat.”

DOCUMENTATION

Written decisions, alternatives included

Every material decision covers Decision, Reason, and Alternatives. Old plans are superseded so the trail stays readable. They are not deleted.

ARCHITECTURE

You lead, AI orchestrates

I dissect the problem, design the architecture, and review. AI subagents execute tasks in parallel. AI accelerates; accountability stays human.

TRUTH

After live, truth shifts to what ships

Production code is the visual and functional benchmark, not Figma or an old mockup. The next iteration starts from what is live, plus usage data.

Pipeline

Discover → Design → Build → Ship

The same four steps on every project. Only the scale changes. The human / AI split stays explicit.

Discover
My Role: Human Lead
Break down the problem, lock scope, and refuse what does not belong in the baseline.
AI Role: Orchestration
Synthesize market research, alternatives, and existing signals, ready for review, not blind trust.
Design
My Role: Human Lead
UX architecture, system flow, and contracts between surfaces. Product direction is not delegated.
AI Role: Orchestration
UI component boilerplate and layout variants from the locked system, not design from a blank canvas.
Build
My Role: Human Lead
API contracts, module boundaries, and PR review. Nothing reaches production without direct quality control.
AI Role: Orchestration
Parallel subagents write code and unit tests against split tasks, not a one-shot of the whole repo.
Ship
My Role: Human Lead
Final release approval and ADR updates. Calls that surface at ship time still get written down.
AI Role: Orchestration
Deployment checklist and docs synced to the live state. The trail follows the code people actually use.

Proof

ADR Vault

Not decoration. Three real records from this project: status, context, decision, and rejected alternatives. Old plans are superseded, not deleted.

ADR-001-product-process-separation.md

ADR-001: Product–Process Separation

Accepted

Context

Product knowledge and project process were easy to mix in one docs folder. Agents then wrote living status into documents that should stay stable.

Decision

Product SoT lives in product-discovery/. Process (rules, tasks, ADRs) lives in project-manager/. Skills live only in .cursor/skills/.

Alternatives considered

Merge everything into a single docs/ folder
Place skills in a non-Cursor location

Working together

A decision that is not written never happened.

If we work together, product direction does not stay in a thread. The band below opens the same conversation. Email remains the primary path.