Too many false positives would kill utility, but I bet many common, avoidable mistakes in terms of reproducibility, data leaking, etc. would be easy to detect.

To avoid overloading users, it would be good to have a good handle on severity levels and let users fine tune verbosity

Basically, at my company, we have methods where the point of calling them is to dispatch an event to the browser with data that gets displayed on a leaflet.js map. So naturally, I should be suppressing the rendering if I can, and doing so decreases "application time" by 14% in these requests!

To be dismissive of software (especially in favor of "AI") is to be dismissive of human problem-solving skills. And that's a really nasty, cynical place to be.

Crappy monolithic systems that don't compose well with other systems will start to die out though.

3/

So, in a lot of ways, orgs can actually miss out on the best of AI/ML if they are too gung ho on "AI jingoism" and abandon software and engineering as concepts.

Software paradigms come and go, but imo there's never been more demand for building systems to keep track of data from the real world

2/