They mention the LLMs working for days but it seems crafted with marketing speak towards clueless C-suites. It doesn't say if it's actually as effective.

I can't tell if that's the claim being made or this is part of their company's ongoing strategy to stay as a well known term in the news cycles, itself a form of marketing.

I'm not looking forward to engineering with an AI that will need to be carefully reviewed and corrected anytime there is something slightly novel or not boilerplate. Either make it smart enough to do the entire work, or it can just remain as a useful tool like in autocomplete and the occasional prompt.

I'm still not convinced that the code it writes will actually speed things up by any significant amount, unless you are producing a garbage product. Any feature that actually provides value is well thought out and carefully built down to every statement of code. Building code line by line is not that time consuming and saves a lot of time in the long run. In other words we will need AI to carefully think wholistically and understand why it wrote what it wrote in the context of adding value to the product.