By embracing adversarial interoperability - instead of chasing hundreds of integration deals across industries that put LLMs in products, they focused fully on integrating product access into chat, by combination of business deals, apps/MCPs, and engineer/designer support for users, all directed towards the goal of having the LLM become the "superapp" where work is done, gradually replacing product classes in order of how easy it is.

There's lots of easy but drudge work to enable this that needs to be done at the fringes. For example, LLMs today could easily replace most people's smartphone homescreen experience, or travel/commute experience, as the data is there and LLMs have the capability, even prices are acceptable - what's missing is explicit first-party support to wire it up, keep it wired up.

One step up, what's missing is accepting this explicitly as a goal: to replace software, to make existing products (whether whole or in pieces) the tools AI uses to do work for you. All the vendors seem to carefully walk around the idea, but avoid engaging with it directly, because once they do, they'll be competing with everyone instead of milking them.