It kind of feels like we've been getting away with something lately.
Building with AI right now is almost too easy.
You plug into an API, send a few prompts, and suddenly you've got something that feels production-ready.
But… are we actually paying what this costs?
Because it doesn't really feel like it.
A lot of what we're using today is heavily subsidized.
Infra is expensive. Models are expensive. And yet, the barrier to entry is still pretty low.
Which is great, until it isn't.
At some point, those costs have to show up somewhere.
And when they do, a lot of current products might start to feel… fragile.
Like, what happens if:
- your token costs double?
- rate limits get tighter?
- the “best model” is no longer affordable at scale?
Not in a hypothetical way, in a real, your app depends on this kind of way.
I think this is something we're not talking about enough as developers.
We're moving fast (which is great), but sometimes it feels like we're building on top of pricing that won't exist in a year.
And that changes how you should think about things like:
- caching
- model selection
- how often you actually need to hit the API
- whether your product still works if costs go up
It's not just an optimization problem anymore, it's part of the design.
We put together a quick short about this if you want the 30-second version.
Curious how you're thinking about this.
Are you building with future costs in mind, or just trying to move fast and figure it out later? 🤔
Top comments (0)