Cheap Ambition

AI has made the projects I’m considering about ten times more ambitious. Things that once seemed to require a large team now look possible for one dedicated person working for a week. That’s an extraordinary change. But I’m suspicious of how quickly my ambitions have adjusted to absorb it.

The obvious conclusion is that choosing what to build matters more now that building is easier. But that leaves something unexplained. If building were the main obstacle, removing it should make our decisions easier. We could finally do the things we’d been waiting to do. Instead, the list expands.

Part of the explanation is that we weren’t keeping a fixed list of worthwhile projects. We were keeping a list of worthwhile projects we thought we could build. AI has removed a filter. Ideas that would once have been dismissed before breakfast now deserve serious consideration.

That’s mostly good. The old filter excluded plenty of useful things. But it also spared us from having to reject ideas for more difficult reasons. Needing twenty engineers ends a line of thought quite efficiently. Being able to build something, but unsure whether anyone would care, leaves you with work to do.

A new capability can also start to feel like an obligation. Once you can build something elaborate, a small project seems like an insufficient use of your tools. You begin judging ideas partly by how much of your new power they exercise.

Suppose you want to help people figure out what to cook for dinner. You could build a page where someone enters the ingredients they have and gets three suggestions. But now you could also build something that learns their tastes, plans a week of meals, tracks what’s in the fridge, and orders the missing groceries. The second project is more exciting to describe. It gives you more interesting technical problems. It looks more like the sort of thing that ought to be possible now.

Whether someone will use it at six o’clock on a Tuesday is still an open question.

The larger project might be the right one. Perhaps suggesting dinner is useless if people never have the ingredients, and planning the shopping solves the actual problem. But that would be a reason arising from the problem. The excitement of finally being able to automate the whole thing arises from the tools. It’s easy to mistake one for the other because both produce enthusiasm.

This is the danger of cheap ambition: you can increase the scope of a project without increasing your understanding of why it should exist.

Implementation supplies a nearly endless sequence of answerable questions. How should this work? Could it be faster? How would it connect to the grocery store? You can spend a productive day answering them and finish with something visibly better. The question of whether you chose a worthwhile problem is less cooperative. It may require showing someone an awkward prototype and discovering that the part you find fascinating barely interests them.

AI makes the answerable questions cheaper. That gives us a choice about where to spend the savings. We can investigate the uncertain parts sooner, or add enough scope to stay busy with the comfortable parts.

Of course, building is itself a way to investigate. People often can’t tell you whether they want something until they’ve tried it. And some ideas sound pointless in a description but become obvious once they exist. Cheaper experiments are one of the strongest reasons to be excited about AI.

But an experiment needs some chance of changing your mind. Imagine spending a week making the cooking app more elaborate. What could happen at the end that would persuade you to abandon it, simplify it, or approach the problem differently? If every possible response leads to adding another feature, the week probably wasn’t much of an experiment.

This suggests a different way to use the increase in capability. Before enlarging a project, ask what was previously too expensive to find out. Perhaps you can now try both dinner suggestions and a weekly meal plan, and discover which one people actually return to. Perhaps you can put a working idea in someone’s hands before becoming attached to its architecture. The same tools that make a large commitment possible can make it unnecessary.

There are still good reasons to build big things. Some problems require them, and AI may let individuals attempt work that previously needed an organization. But the size of a project should follow from what the problem requires. Greater capability alone doesn’t tell you that.

When something that once required a team becomes a week’s work, there’s no need to respond by finding a project that requires a team again. That week can be spent finding out whether the smaller thing deserves to become larger.