🤡

A thread went viral on Reddit: a guy talks about a friend who recently got hired at Anthropic. The friend shares "insider info" from the front lines:
▫️ No one has typed code for several months.
▫️ They work not as engineers, but as product managers (PMs): they launch a bunch of AI agents in parallel and just hand out tasks to them.
▫️ If you're sitting and watching an agent generate code, you're already behind. At that moment, you should be launching a second agent.
▫️ Productivity has skyrocketed; Anthropic is currently shipping features faster than anyone in the industry.

Sounds like a perfect future, right? You sit back, sip your latte, while an army of obedient robot-Indians for zero cost builds your bloody enterprise.

But if you go to their claude-code repository, there are thousands of open issues right now. Bugs, terrible UI/UX, broken path handling. The uptime of Anthropic's own services also periodically does somersaults.

And the concept of "you're now a PM for agents" might crash against one simple thing — context switching.
Managing five juniors is painful. Managing five hallucinating machines that spew out 500 lines of untested code per second, with no understanding of the project's global architecture — that's a one-way ticket to the loony bin.

In short, the company that sells AI agents is telling (through "friends of friends") that the only right way is to use AI agents non-stop. Just a coincidence.

Just in case: I'm not against neural networks; they should be used to the fullest.