And half-truths are useful: as in it's helpful to understand the positives, the (shared) reasoning, and what new information you get that you did not know vs before. E.g. in the case of Uber: that it's possible for a company to run thousands of microservices (BUT: there's a lot of infra needed + new problems created this way that weren't publicly detailed all that much!) In the case of Bun: that it's possible to have a surprisingly fast and seemingly decent rewrite for a project already thouroughly tested and where the core team is small, one person can pull the trigger on major decisions etc. As with anything engineering, there will be tradeoffs: for example, could have AI been used to fix the existing codebase? Solve persistent memory issues with eg far more advanced linting etc? Could have it been worth it? This part is under-explored.

One current example is the Bun rewrite in Rust. We know one side of the story (and it was very well told by Anthropic), but will miss a lot of other parts. So before you jump to rewrite your project in Rust with AI... try to at least fill the gaps! And gather the alternative options you have ofc.