Side projects usually die in the gap between a promising idea and the first version you can actually react to. AI is useful when it compresses that gap.
A tighter loop
Start with a narrow outcome, generate the boring scaffolding, and get something running. Once the artifact is real, judgment becomes much easier: keep this, cut that, rethink the premise.
Judgment is still the job
Fast output increases the need for taste, verification, and clear intent. A plausible implementation is not automatically a good one. The maker still owns the choices.
Use AI to buy more iterations, then spend those iterations deliberately.
Working practice
- Define the smallest testable version.
- Ask for structure before detail.
- Run and inspect the actual result.
- Keep a short log of decisions and surprises.