The labs are all betting on the models solving all problems.
They'll get infinitely good at avoiding prompt injection.
There will be no need to share code from anyone else.
It will just be the model, you, and your data.
This assumes the models will get very good at figuring out how to give you exactly the software you need, at the moment you need them, without the user needing to act like a PM.
Models are also a slow pace layer since they take so long to train and adapt.
Another approach: use the LLMs as dumb muscle, not the brains.
The brains would still come from real users with situated judgement.
If you could have the human judgment at a faster pace layer, and accumulate insights from real situated users that could help others, open-ended value could be unlocked.