AI will unbundle the app.

Why are apps the size they are today?

Why do we have the "chunkiness" of the app, in terms of the number of use cases that are typically bundled into one app?

Part of it is because software is expensive to write, cheap to run.

High fixed cost, low marginal cost.

Over time prices tend to reduce to the marginal cost of production.

Over sufficient time, with sufficient competition, software thus reduces down to ~free, but is only created if there's a large enough market to justify the fixed cost of creating it.

Caveat: high switch costs effectively reduce competition and slow this inevitable margin compression.

But LLMs make software much cheaper to create.

For small, non-generalized software, the cost approaches zero.

When the fixed cost of writing software goes to 0, the minimum viable market for a bit of software also approaches 0.

That implies that we'll see an explosion of software that is so crappy and bespoke to a particular use case that before no one would have bothered to build it.

Marvelously powerful and situated and niche.

When software has a high fixed cost, the use cases have to congeal into a bundle that is big enough to have a market to make the bundle viable… and that also contains within it a business strong enough to at least cover the fixed cost.

But when it's zero fixed cost, those bundles no longer need to coalesce, and they also don't need to have a business model inside: the fixed cost is zero, there's nothing to pay for!

Software today does not actually have zero marginal cost: it has a distribution cost.

The equilibrium point of "chunkiness of app" that we see today is based on the fixed cost of software production and the marginal cost of distribution.

AI reduces the fixed cost of production.

An alternative set of laws of physics could reduce the marginal cost of distribution.

Together it could lead to wildly different outcomes.