What to build in a world where implementation cost goes to zero

The recent progress of LLMs, especially in coding, is striking. The very notion of “the effort of implementation” is fading, and even non-engineers can now turn an idea into something real, instantly.

Here’s an interesting angle: treating apps as disposable.

Most people, when they build something, hope it becomes a permanent asset. But given how fast features evolve today, that’s inefficient. Much of what an individual builds because “this is handy” gets absorbed into a big platform’s standard features a few months later — because what my intuition needs tends to line up with what the market wants.

If so, there’s no need to cling to piling up permanent code assets. What matters is speed: how fast you can fill the gap of this very moment.

What made this shift possible is that AI gained the ability to turn ambiguity into something deterministic. In traditional development, vague requirements were a fatal cost. But an LLM can derive structure from vague instructions and draw the missing information out of a human. The human’s role narrows to “providing information” and to “deciding” — that is, throwing away everything else — from the options it lays out.

Right now I’m building an agent for fuzzy domains, like arranging a get-together. There’s a good chance this becomes a standard feature someday. Even so, if you feel the inconvenience right now, there’s meaning in building it now.

From an age of making “assets” to an age of enjoying “the pleasure of giving a thought immediate form.” That lightness — not forcing permanence, using things up functionally — may be the optimum for intellectual production in the age of AI.