As the tooling matures I think we'll see better support for mixing models — local and cloud, picking the right one for the task. Run the cheap stuff locally, use the expensive cloud models only when you actually need them. That would go a long way toward managing costs.

There's also the dependency risk people aren't talking about enough. These providers can change pricing whenever they want. A tool you've built your entire workflow around can become inaccessible overnight just because the economics shifted. It's the vendor lock-in problem all over again but with less predictability.

But I don't think the models are going to get there on their own. AI will generate a working mess all day long if you let it. The pressure to write good code has to come from the developer actually reviewing what comes out and pushing back. The incentive is there but it only matters if someone acts on it.