Any complex work has a shape to it. The first chunk comes together quick. Scaffolding, happy path, the demo looks good. Then you hit the rest of it, the part that takes it beyond a demo. Edge cases, error states, behavior under real load. Reality has a surprising amount of detail.
AI has compressed the first phase dramatically. What took a week might take a day. But the second phase hasn’t shortened by the same fraction. And if you weren’t careful in phase one, the second phase gets longer or worse, never gets completed.
Every previous productivity tool in software, Rails, React, AWS, whatever, abstracted away work you didn’t need to do anymore. You didn’t need to understand TCP to build a web app. The abstraction held. You could forget what was underneath.
AI abstracts away the thinking process itself. And that abstraction leaks. You used to build a mental model of the system for free, as a byproduct of the slow tedious work in the first phase. AI skips the tedium. It also skips the understanding that came bundled with it.
So you reach what feels like almost-done in record time. Now you’re debugging code you didn’t write. From a collaborator who doesn’t know your system’s constraints, your deployment setup, your users. Every fix is a guess. Guesses compound. You patch one thing and break another. The project enters this zombie state, perpetually close to done.
Which leads to the counterintuitive bit. AI can make the total time go up. It’s fast in exactly the phase where understanding gets built, and that missing understanding costs you later. You end up wasting time instead of saving.
Speed determines how quickly you converge. AI front-loads the feeling of convergence. If you’re not careful, you’ll converge on mediocrity, just faster than before.
Related