I didn't read the whole thing, I will get back to it tomorrow and read because it is interesting but it is also alarmist, I stopped reading where it says that AI tests as a human would... And that is not EXACTLY true...

I have used this feature myself. The AI can indeed open the website you are building and navigate but it is not doing the job like a human, it uses playwright, an automation tool. Having worked with playwright I myself saw how sloppy the AI tests stuff because it just can not do stuff like a real tester would. I can pinch and zoom and try to click and type where it is not supposed to and do unexpected stuff. That is precisely why I need to manually review everything the AI does.

That connects to what I wrote though, that if everyone can pump code using AI then the differentiator changes, suddenly it is not who can pump the most features and bug fixes that has the best product because everyone can do that, but the best product will be who can test better, like a human would. This has happened with code, AI got better at coding and now reviewing and testing is what matters, when AI geta good at reviewing and testing then the meta of the game will shift... My theory is that humans will always be needed because if some technique is a commodity then you need someone to do something different to stand out as a company and as a product.

And about code reviews, yes, it is great, I love seeing copilot review Jules PRs and tagging Jules to address them (when jules doesn't ignore the tags or pretend "they don't see the comment they were tagged at"

I am sorry for the hypocrisy of not (yet) reading the full article but then writing an article back but basically AI is limited. Gemini may have 1 Million tokens of context, but the more context it uses the worse result it has. The more code AI writes the more small details it missings and the more reviews are needed and the more reviews AI does the more testing is need and the more AI tests... Well honestly I don't know what comes next after reviews but at that point AI has churned through so many tokens and so much context window that it can't possibly do the next step well, whatever the next step is, and that is where humans will go.

edit: sorry for the long text and the typos, it is late at night and I am going to sleep, I am on my phone and all but this is an extremely interesting subject