I don’t have a CS degree. I have eleven years of fixing the next thing in front of me until the next thing got bigger. That path teaches you exactly one lesson, repeatedly, until it sticks: ship the small real thing, not the big imaginary one.
AI has not retired that lesson. It has put it under a microscope.
The imaginary big thing is now cheaper to start
That’s the trap. The cost of beginning an over-ambitious project has collapsed. You can scaffold the dream architecture in an afternoon now. What hasn’t gotten cheaper is finishing it, maintaining it, or being correct. So the gap between “started” and “shipped” didn’t close — it moved, and it got easier to fall into without noticing.
What I actually do
A few habits that have held up across a live SaaS, two published games, and a pile of open-source tools:
- Define done before you define how. If you can’t write the one sentence that says what success is, AI will happily help you build the wrong thing in great detail.
- Keep the human on the load-bearing parts. Generate the scaffolding, the tests, the glue. Write the part that would be expensive to get wrong, by hand, slowly, on purpose.
- Read everything it gives you. Not skim — read. If you wouldn’t sign off on it in a code review from a junior, don’t sign off on it because a model wrote it confidently.
- Optimize for the maintainer, who is future you. The most expensive code is the code nobody understands six months later, and AI is very good at producing exactly that if you let it.
Self-taught is a feature here
The thing I worried was a gap — no formal training — turned out to be good preparation for this moment. Self-taught means you’ve never been able to assume something works because someone credentialed said so. You learned to verify. That instinct is the entire job now.
The plot was never the code. It was the judgment about which code to write. Keep that, and the tools are extraordinary. Lose it, and they’re just a faster way to be lost.