Now, almost all of our competitors and us have taken a fundamentally different approach to how we use the AI tools for our work, and it's been really fascinating and fun to compare and contrast with them. We're finally different, and to me that's a great thing.
I'm sure approaches will converge into "best practices" again but for now I'm really enjoying that we don't all have to be the same.
I'm not sure if you're making excuses for those systems that are failure but I'd say no one should have an excuse for these failure - whether it's "there are no best practices" or otherwise. There were best practices previously and until someone can be sure what they're doing is good, they should be following those old approaches.
It's a bit like someone who says "everyone's doing coke now and some of them seem to be OK, that means I should be able to do it too 'til I get the hang of it".
The harsh reality is that some people/systems are adapting just fine to the new way of doing things, others are making a mess of things by rushing via vibe coding and not checking the work (either by choice or by circumstance at work). The results won't be (and shouldn't be expected to be) linear.
The 2010s brought on a very groupthinky dynamic to software dev. This led to a culture where the majority of developers just ask "what is everyone else doing." Now that we're past the blog era of development, that shared reality appears to be breaking down. Now the answer really is "whatever makes sense for/to you," but I think (anecdotally) that's freaking a lot of people and teams out.
For the first time in a long time, a "best practice" doesn't really exist.