If you've ever undertaken the task of documenting entire workflows, then you know that you quickly put up the white flag at the word "entire".
When you actually talk to people about what they do there are often many, many nuances, micro-events, micro-decisions and micro-actions in their work. This is why it can take days/weeks/months to completely train a new person for a job.
This level of detail is barely documented - anywhere. There is a huge amount of information buried in workflows that AI has barely had access to for training. A lot of this is more in the realm of world models, rather than LLMs.
So imagine trying to use AI to improve these workflows it knows so little about. Then imagine AI trying to reinvent them across an organization.
We find these use cases where AI provides great value - totally true - but these barely scratch the surface of what goes on.
This strikes me as a very solid methodology for improving the results of all AI coding tools. I hope Anthropic, etc take this up.
Rather than converging on optimal code (Occam's Razor for both maintainability and performance) they are just spewing code all over the scene. I've noticed that myself, of course, but this technique helps to magnify and highlight the problem areas.
It makes you wonder how much training material was/is available for code optimization relative to training material for just coding to meet functional requirements. And therefore, what's the relative weight of optimizing code baked into the LLMs.