> The companies that will win long-term are the ones building open protocols and letting users bring their own model.
These seem contradictory. It sounds like you're saying that the long term winners are the ones who do the easy part. The future I see is open source harnesses talking to commodity models.
There is barely any magic in the harness, the magic is in the model.
Try it: write your own harness with (bash, read, write, edit) ... it's trivial to get a 99% version of (pick your favorite harness) -- minus the bells and whistles.
The "magic of the harness" comes from the fun auxiliary orchestration stuff - hard engineering for sure! - but seriously, the model is the key item.
It turns out that "most of the bell and whistles" could amount to instructing models how to use tools like tmux
Perhaps there's enough overlap among the low hanging fruit that you can initially sell a harness that makes both genomics researchers and urban planners happy... but pretty quickly you're going to need to be the right kind of specialist to build an effective harness for it.
The interesting thing is how this differs from traditional infrastructure vendor lock-in. With Datadog or PagerDuty, the lock-in is mostly data gravity and workflow integration. With AI coding tools, the lock-in is muscle memory and workflow habits. The switching cost is not migrating data -- it is retraining your fingers.
But the moat is thinner than they think. Models are converging in capability faster than anyone predicted. The real value in these tools is not the model, it is the harness: context management, tool integration, session continuity. And that is the part that is easiest to replicate.
The companies that will win long-term are the ones building open protocols and letting users bring their own model. Same lesson the monitoring industry learned the hard way -- proprietary agents and custom query languages felt like defensible moats until OpenTelemetry showed up and made the collection layer interchangeable.