My work
AI coding agents made writing code fast. They didn’t make software engineering easy.
Most agent-led projects fail at the exact same places human-only projects do: vague scope, fragile architecture, and shallow verification. Without strong technical direction, faster code generation just produces legacy debt at machine speed.
I bring twenty years of web engineering, frontend architecture, and team leadership to the decisions that determine whether software holds up—turning complex product goals into clear architectures and shipping them with engineers and AI agents.
How I work
- Architecture before prompts: Define system boundaries, data flow, dependencies, and verifiable acceptance criteria before a human or agent writes code.
- Repeatable agent workflows: Design durable instructions, scoped tasks, and automated review loops so coding agents ship maintainable changes instead of noisy diffs.
- Verification and ownership: Challenge unnecessary complexity, enforce automated tests and behavioural checks in CI, and own what goes to production.
Proof in the browser
This site doubles as my live workbench—architected on the open web platform with strict security headers, edge SSR, and automated test gates:
- On-device WebGPU AI: Local LLM Chat, AI Background Removal, and Numbers Extractor run quantized models directly in your browser via Web Workers—zero server inference, zero uploads.
- Zero-roundtrip browser tools: Markdown & Mermaid to PDF, Document to Markdown, and Browser Image Compressor push heavy file processing to the client and edge.
- Interactive systems & studies: Press
/anywhere to test the semantic command terminal, or explore the full Lab for canvas studies and arcade rebuilds.
Where I help most
I’m most effective when there is substantial technical work to structure and see through: architecting a new web platform from scratch, untangling and modernising an established product, or helping an engineering team turn AI agents into a reliable delivery multiplier.