How AI fits into my engineering process.

Before any code gets written, I use AI-assisted exploration to think through architecture and trade-offs — inspecting the existing codebase, surfacing edge cases, and working through multiple structural directions faster than I could alone.

Several approaches are typically considered before implementation starts. The plan that moves forward is the one I decide fits the system, not the first option generated.

Consequential decisions get written down before they get built. Architecture choices, content structure, animation strategy — each has a recorded rationale, not just a resulting implementation.

Every meaningful change follows the same sequence: inspect the current state, propose a plan, get that plan approved, implement it in small steps, validate the result, and summarize what changed. Documentation is a byproduct of that discipline, not an afterthought.

Implementation happens in small, reviewable steps, scoped to the files a change actually touches. Nothing gets built outside the plan that was already agreed on.

AI writes a large share of the code, but within boundaries I set — the folder structure, the conventions, and the constraints of each change.

First drafts are rarely final, and most of the real improvement happens after the first pass, not during it. I review what's produced, push back on choices I disagree with, and refine copy, structure, and implementation details through several rounds.

This is the most collaborative part of the process — closer to iterating with a strong engineer than accepting a tool's output as-is. What ships is usually several revisions removed from the first draft.

Every meaningful change is validated the same way: linting, type checking, and a production build, every time — not an occasional check but a non-negotiable gate.

Passing those checks is necessary, but never sufficient. Every production change still goes through manual engineering review before it's considered complete — automated checks catch errors, they don't confirm the work is right.

AI-assisted workflows accelerate execution and help explore options faster than working alone, but they don't change who's accountable for the result.

Architecture, trade-offs, maintainability, security, and every final technical decision remain the engineer's responsibility — on this project, and everywhere else this workflow applies.