AI Model Adoption, Switching Costs and Growth

Fable 5, like Opus 4.7 and 4.8 before, is disappointing. Or rather, you can’t swap it into existing projects that have been using a different model already for, say, months, without feeling like something’s not working. It’s like changing the scaffolding of a project mid-project. Often results in something being torn down. Or like changing a programming language in production software mid-stream (pre-AI). Hard to justify.

At first it felt like swapping in a new model could be more akin to upgrading OS. But it’s not. The model chosen and used in a project locks in an interaction pattern. Changes in models, and even version iterations, seems to disrupt that pattern. That means for the models to gain adoption the work involving AI has to increase. Is there more greenfield work still to be had? Or will the LLM producers be on a path of price increases only? Another path would be waiting for the ecosystem to more fully expand, bringing more novel applications, killer apps, to the market.  On the positive, it means the ROI on older models that already had many workstreams starting to be built with them, should grow. But it feels like the hyperscalers are looking for hockey sticks and not dividends.  

APPLICATION
The lock-in described here is a form of option foreclosure, but not a clean one. Choosing a model and building work around it narrows the option set incrementally, through accumulated interaction patterns rather than a single commitment. This has implications for how organizations think about AI adoption and speaks to the kind of department or infrastructure thinking that needs to be involved in governing, pattern setting, for AI adoption. By the time an organization evaluates whether to switch models, the switching cost may already be structural. The greenfield question then becomes central to AI economics. If existing work compounds around its original model, growth requires new work, not upgrades. That is a constraint the hyperscalers may not yet have priced into their growth narratives.