A "class divide" is opening up inside engineering teams: veteran "craftsmen" engineers fix the AI generated commits that less experienced "vibe coders" push out, while their own performance reviews now count AI tool usage.
A venture capital partner posted on X last week that software engineers are facing an identity crisis "bordering on depression," as CTOs push AI coding tools and an engineer-subculture ethos of "tokenmaxxing," the practice of maximizing AI-token usage as a status signal. Deedy Das, a partner at Menlo Ventures, argued that the pressure is creating a "class divide" inside engineering teams between "vibe coders," who ship fast by leaning on AI without reading the output, and veteran "craftsmen" engineers stuck cleaning up the damage.
Business Insider covered Das's post as a documented industry trend rather than a one-off rant, and the mechanic Das describes is starting to show up in workplace policy, not just social media threads. Per Futurism's reporting, Meta is now factoring employee AI tool usage into performance reviews, turning the question of how much AI an engineer uses into a direct input on ratings, promotions, and continued employment. Meta has not, as of this writing, published a public policy confirming the practice; the report relies on internal accounts cited by the outlet.
The arithmetic of that policy is what makes the divide structural. Senior engineers who actually understand the systems they work on are being asked to fix the AI-generated commits that less careful or less experienced colleagues are producing at higher and higher volume, while their own scorecards increasingly count how many AI completions they ran. That is the kind of involuntary quality assurance role Das and the engineers he describes say has been grafted onto jobs that used to be measured on architecture, reliability, and review of human-written code. The downstream product, in their telling, is the kind of sloppy AI output they have started calling "workslop": machine-generated code that compiles, sometimes runs, and quietly rots the next time it has to change.
There is some independent evidence that the pattern Das describes is more than vibes. The Stack Overflow 2025 Developer Survey is the most-watched industry baseline on how developers are actually using AI tools, and the data suggests that AI is now a daily part of most engineers' workflows. That is the demand side. The supply side is darker. CodeRabbit's State of AI vs Human Code Generation Report and GitClear's AI Copilot Code Quality 2025 Research both find that AI-generated commits pull in different directions from human-written code on a number of quality measures. GitClear's data, in particular, has tracked a sustained drop in the proportion of code changes that are "refactored" within two weeks of being written, a marker the company treats as a proxy for code that the original author actually understood and intended to keep.
The repository-level consequences are becoming visible. LeadDev's reporting on GitHub and MAREF's analysis titled "GitHub Is Becoming a Giant AI Code Dump" both describe the same end state: a public source ecosystem in which a growing share of commits were never seriously read by a human. That is the world in which the craftsmen engineers are being asked to do the cleanup. They are also the people best placed to see which AI output is going to cause the next incident, which makes them the engineers the company can least afford to lose or burn out on review queues.
The pattern Das names has the structure of a management failure that is still reversible. The veteran engineers who are frustrated now are the same engineers companies say they cannot hire fast enough. If senior engineers stop signing up for jobs that grade them on AI usage while they spend their days fixing the AI output of their peers, the bottleneck for AI-accelerated software development shifts from compute and models back to the people who can tell the difference between code that works and code that just compiles. The question for engineering leadership is whether the rollout gets adjusted before the craftsmen engineers leave, retire early, or simply stop volunteering for the AI-cleanup shifts. The honest read of where this is heading is that the divide is not going to resolve on its own, and that the next round of AI-related talent departures may be senior engineers, not models.