Claude Code and the engineers reduced to pressing Enter

What the engineer says

The engineer, who posts on X as v0xium, said they joined the company a little over two weeks ago and found the state of engineering “terrible”: nobody knows anything.

All development at the unnamed company, v0xium claims, is done with Anthropic’s Claude Code. Employees are pushed to produce as much code as possible. Writing code is no longer the bottleneck, but the overall operation is still slow.

According to the engineer, people work 12–13 hours a day, mostly communicating with Claude and pressing Enter. The pattern is the same from L1 engineers to L7 specialists.

The post was viewed more than seven million times and received tens of thousands of likes. V0xium concluded that “it’s over.”

The response was bleak. X owner Elon Musk replied, “terrifying.” Podcast host and venture investor Chamath Palihapitiya compared the situation to turning an entire generation of technical workers around the world into retirees pressing a button on a slot machine. He said developers were experiencing the same deterioration of thinking as people who endlessly scroll social media.

The work disappears before the code does

The central complaint is not that AI writes code badly. It is that engineers are losing the parts of the job that made the work feel like engineering.

V0xium says there is no feeling of victory left, and nobody fixes the errors. Large language models do the thinking, while the people operating them lose energy and stop building products they can understand or take pride in.

That creates a division between two kinds of developers:

The source story describes AI-generated code as more likely to contain bugs and security problems when it is not fully checked. That risk becomes more serious when the same system that produces the code also removes the time, knowledge and motivation needed to review it.

Palihapitiya’s standard for development automation is higher than simply producing software faster. In his view, the system should understand the developer’s intent and improve the final product.

I think that distinction is the real fault line. An assistant that accelerates judgment can make an engineer more capable. An assistant that replaces judgment turns the engineer into a delivery mechanism for text that happens to compile.

The missing owner

The announcement is quiet about the part that determines whether this model of work can function: who owns the correction loop?

The phrase “meat proxies” was recently coined for people who reproduce chatbot answers word for word without critically understanding them. That label is crude, but it describes the failure mode in v0xium’s account: the human remains in the workflow while responsibility moves somewhere else.

Didi Das, a partner at Menlo Ventures, wrote at the beginning of this year that most software engineers were facing an identity crisis approaching depression. His description is a checklist of a profession losing control:

Das added that people eventually give up. The craft they loved dies.

The more uncomfortable possibility is that a company can mistake the disappearance of visible coding work for progress. If everyone can generate more code but nobody can explain, review or repair it, the bottleneck has not vanished. It has moved into the product, where the cost is harder to see and the person responsible is harder to find.