A short list of things I don't give ground on.
Most engineering relationships fail in the same handful of ways — silent overruns, unspoken assumptions, scope drift dressed up as "iteration", a Slack channel that goes quiet for two weeks. The following principles are how I avoid those exits. None of them is clever. All of them have cost me work I was happy to lose.
Vertical slices, not horizontal passes.
One feature from start to finish before the next begins. The work has to be useful at every checkpoint, not just at the end. This is non-negotiable.
Real software beats vibe coding.
AI speeds up the work but doesn't sign the contract. I read the existing code before I generate, I name the constraints before I name the solution, and I leave a build log so the next person — including future me — can pick up where I left off.
Your data doesn't leave your network without a reason.
Cloud is fine when it's cheaper and the data is yours. For client files, HR records, tax documents, on-premise is the default — and I tell you that on day one.
Estimates are honest, or they aren't estimates.
A range, a date, and the assumption that breaks it. If the assumption breaks, you hear from me that same week. No silent slipping.
Two clients at a time.
I don't take a third. The work suffers and so does the relationship. When the calendar is full, I say so and recommend someone good.
- —Crypto, NFTs or speculative trading systems.
- —SEO content farms or affiliate-link projects.
- —"Disrupt-the-X" pitch-deck prototypes.
- —Anything I can't explain to my mother in one sentence.
- —"Just quickly add this small feature" two weeks before launch.
- —Skipping tests because the deadline is tight.
- —Declaring something done before it's run a week in production.
- —Picking a stack because the founder saw it on Twitter.