Product operating model · Human governed
How I Work with AI
Direction stays human. Repetition becomes system.
I use AI as a governed operating system for product work, not just a chat interface. My workflow spans research, planning, design exploration, asset creation, development, troubleshooting, unit tests, and accessibility review. I also codify reusable skills, guidelines, and guardrails so the system stays consistent as projects evolve.
- Context
Define the problem, evidence, constraints, and standard of quality.
- Plan
Separate decisions from execution and establish how the work will be checked.
- Execute
Use agents and tools where they create leverage across the product lifecycle.
- Review
Inspect the actual artifact, test behavior, and challenge the load-bearing assumptions.
- Codify
Turn useful decisions into skills, rules, templates, tests, and documentation.
Across the product lifecycle
AI participates in the work. It does not own the outcome.
Research and planning
I use AI to explore directions, synthesize source material, expose assumptions, and shape an evidence-backed plan before implementation begins.
Design exploration
I generate and critique visual directions, test composition ideas, and accelerate early concepts without outsourcing the final judgment.
Asset creation
I create supporting imagery, copy, variants, and production assets when rapid iteration materially improves the work.
Development and troubleshooting
I direct agents through implementation, compare approaches, trace failures to their mechanism, and reduce friction across build cycles.
Unit tests and QA
I use AI to generate test ideas, probe edge cases, reproduce failures, and verify that the complete experience behaves as intended.
Accessibility review
I review semantics, keyboard behavior, contrast, responsive reading order, and assistive labels as part of delivery—not as a final audit.
Governance and reusable skills
I turn repeated decisions into skills, guidelines, acceptance criteria, and guardrails so quality survives across agents, projects, and time.
The control boundary
Acceleration is delegated. Accountability is not.
Agents can
- Propose directions
- Gather and synthesize evidence
- Generate variants and implementation
- Run checks and surface discrepancies
I retain
- Problem framing and product intent
- Design judgment and tradeoffs
- Verification against the actual artifact
- The final decision about what ships
Proof artifacts
The workflow leaves evidence behind.
Reusable artifacts make the operating model inspectable, teachable, and easier to improve. These are working templates, not presentation props.
Product brief
Frames the problem, evidence, constraints, risks, and definition of done before agents begin producing.
product-brief.md↓Markdown · Review protocolHuman review loop
Separates proposal, inspection, verification, and ratification so speed never silently becomes authority.
review-loop.md↓Markdown · GuardrailsAgent working agreement
Defines scope, source-of-truth rules, escalation points, and what requires human approval.
agent-guardrails.md↓Markdown · QA checklistAccessibility review
A compact pre-delivery pass for structure, keyboard access, labels, contrast, motion, and responsive order.
accessibility-checklist.md↓