What Claude Code Actually Changed About How I Ship Features
Not autocomplete, not hype — a real breakdown of where an AI coding agent earns its place in a solo full-stack workflow, and where it doesn't.
As the only engineer on Amber Car Rental, the bottleneck was never typing speed. It was breadth: frontend, backend, admin tooling, and the client conversation that turns a vague ask into a shippable spec, all at once. That's the actual problem Claude Code solves for me — not "writes code faster," but "lets one person credibly cover a team's worth of surface area."
Where it earns its place
- Scaffolding CRUD and admin screens from a Supabase schema, so time goes to the booking-availability edge cases instead of boilerplate forms.
- Debugging: pasting a stack trace plus the relevant Supabase Realtime handler gets me to a root cause faster than a manual bisection.
- First-pass code review on my own PRs before a client sees them — catching the obvious stuff so the conversation with the client stays about the product.
Where it doesn't
Architecture calls — Supabase over hand-rolled auth, a database-level constraint over an application check for double-bookings — are decisions I make deliberately, then implement with AI assistance. The tool is fast at execution; it isn't the one deciding what "correct" looks like for this business.
The workflow that actually works: scope the decision on paper first, then let the agent execute against a spec I already trust. Skipping the first step is where AI-assisted development gets a bad name — it just means shipping someone's untested guess faster.
A car rental agency needed to move off manual, phone-and-spreadsheet bookings to a real-time online platform — built and run solo.