Throwaway software

One of the pleasant promises of AI is that people with good taste but limited technical skill can build things they couldn’t before. Someone who doesn’t know how to code might still have a great idea for a website, or a dashboard, or an app.

Which makes you wonder: if AI makes it so easy to put your taste into a product, why are so many AI projects discarded? For myself, I can think of more than a handful of projects I sunk time into but never quite got where they needed to be for me to return to them after first development (RIP AI personal assistant/task managers).

That said, discarding projects is a time-honored tradition that predates AI. The question here is, given what AI is good at, what makes it hard to get to outputs that meet our expectations for value well enough for us to actually use them?

The first issue I’ll point to is that it’s easier to feel strongly about what we like than define it thoughtfully. I remember a conversation with a game designer who talked about small tricks that make a game ‘feel’ good, like having the user’s hitbox be slightly smaller than their character, and having a certain number of frames for preparation movements. Even if I had a good idea for a game, if I don’t know those deeply non-obvious features, I may end up disappointed.

But all is not lost, since we also develop taste by building, even if we don’t end up building well. I won’t say how many terrible UIs I’ve tried to make, and I won’t say my taste for UI development is especially good, but the process of trying to build UIs that are good enough has helped me make better UIs over time. Which is to say, discarded AI projects can be a sign of people experimenting and learning to use AI well.

That’s AI today. What about AI tomorrow? Elements of taste aren’t immune to AI progress, and recent models seem significantly better at UI and design than earlier versions. If AI improvement means, in practice, embedding those small, unintuitive parts of what makes something good, like easy-to-miss hitbox sizing and jump momentum best practices in game development, then anyone who creates with AI will be building from a new baseline.

Which is to say, as AI gets better, effort and incomplete or ill-defined tastes may become (even) smaller barriers to a completed project. With AI becoming ‘good enough’ across more areas of development, even narrowly effective tastes can end up with increasingly workable, decreasingly discarded, if not always very thoroughly unique, development projects.