AI speeds up the work.I own what ships.
My everyday workflow for turning a clear problem into working software. Faster execution, careful decisions, and a human accountable for the result.
Part of the process.Every day.
I use AI throughout the product lifecycle. The goal is to spend more time on decisions that matter, and less on the work that repeats.
- 01Automation
Less repetition. More progress.
Component scaffolding, documentation, migrations, and release notes. I delegate repeatable work with a clear brief and review the result.
- 02Research
Start with the right context.
Read the codebase, explore the problem, and compare approaches. Gather the evidence before choosing a direction.
- 03Analysis
Work through the tradeoffs.
Break down requirements, map the data flow, and surface edge cases. Challenge the plan while changing it is still cheap.
- 04Development
Build across the stack.
React interfaces, typed APIs, database changes, and background jobs. Small, reviewable changes grounded in the existing architecture.
- 05Prototyping
Make the idea usable.
Turn a brief into something people can try. Test the important interaction in a working prototype before committing to the full product.
- 06Testing
Show that it works.
Type checks, targeted tests, accessibility, and browser verification. Evidence against the brief, with the gaps made explicit.
- 07Agents
Give the tools shared context.
Connect the issue, design, code, and documentation through agents and MCP. Keep each task bounded, with a clear handoff.
- 08Products
Own what goes to production.
AI helps with execution. I make the architecture decisions, review the changes, and stay responsible for the product after it ships.
From a briefto a working product.
A product I run and tools I’m building. Explore the problem, the architecture, and the decisions behind each.
One place for a driving school to manage admissions, scheduling, payments, and communication.
- 01
Intake
Enquiries, admissions, onboarding
→ - 02
Operations
Scheduling, payments, branches
→ - 03
Delivery
Durable workflows and queues
→ - 04
Customers
Isolated tenant data
→
Deduplication belongs in the delivery layer, so a retry cannot send the same message twice.
Founded and built end to end. Live with paying customers. An agent loop helps reproduce, fix, and verify tickets; I review the result.
View case study· Pick a project, or type help below.Bridge walkthrough readyTHREE BUILDS. THE SYSTEMS BEHIND THEM. SELECT A PROJECT TO EXPLORE ITS FLOW AND DECISIONS.
The tools,and their place in the process.
A familiar toolkit for thinking, designing, building, and delivering. Each tool has a job; judgment connects them.