How I ship software faster with AI — and where I don't trust it
AI has multiplied what one engineer can deliver. A look at how I use it day to day, and the guardrails that keep client work production-grade.
Brandan Pratt · September 30, 2026 · 2 min read
A few years ago, a project like a full client portal with auth, dashboards and integrations meant a small team and a few months. Today I regularly deliver that scope as one engineer, faster — because AI is part of every stage of how I work.
That's not a claim that AI writes the software. It's a claim that AI removes the slow parts, so more of my time goes into the parts that matter: understanding the business, designing the system and making it reliable.
Where AI speeds me up
Exploring unfamiliar territory. New API, new framework, new domain? I can get from zero to a working spike in an afternoon instead of days of docs-reading.
Scaffolding and boilerplate. Forms, CRUD screens, migrations, typed API clients, test fixtures. Necessary, repetitive and perfect for AI assistance with tools like Cursor and Claude.
Prototypes clients can touch. Instead of describing a solution in a document, I can show a clickable version early. Feedback gets sharper and expensive misunderstandings disappear.
Documentation. Clear handover docs used to be the first thing cut. Now they're cheap to produce and keep current.
Where I don't trust it
Security and access control. Authentication, permissions and anything touching personal data get written and reviewed by hand, against a checklist.
Data integrity. Migrations, financial calculations and reporting logic are tested against real cases. A confident-looking query that's subtly wrong is worse than no query.
Architecture decisions. AI is a useful sparring partner, but it doesn't carry the context of your business, your budget or your team's ability to maintain the result. That judgment is the job.
Anything I can't explain. If I can't explain why a piece of code works, it doesn't ship.
What this means for clients
- Shorter timelines for the same scope.
- Lower cost for prototypes and MVPs.
- More of the budget on the hard problems, not the boilerplate.
- Production standards unchanged — the same discipline I apply to government digital services.
AI multiplies an engineer's output. It doesn't replace the engineering. If you want that combination on your project, let's talk.