A software engineer posting as v0xium said on 20 September 2026 that Anthropic's Claude Code had turned his new job at an unnamed large company into a 'soul-sucking' routine. He claimed colleagues were working 12 to 13 hours a day while relying on the AI tool to generate much of their software-development work and racing to ship more.
His complaint was not simply that Claude Code was writing more of the code.
V0xium said engineers were being given too little time to read, understand and review what the system produced before moving to the next task.
Engineer Says Claude Code Handles Much of His Workflow
In his X post, v0xium said Claude Code was producing specifications, code, tests, product-requirement documents, tickets, ticket resolutions and reports. He claimed people were working 12 to 13 hours a day 'just to press enter' and later wrote that 'nobody is thinking anymore.'
V0xium told Business Insider by email that the deeper problem was management pressure to ship more features. He said the push encouraged AI use across tasks ranging from product-requirement documents and system designs to user interfaces and backend code.
In his account, faster code generation had not created more time for engineers to understand their systems. Instead, it had raised expectations for how much work teams should deliver.
The employer has not been identified.
Business Insider reported that v0xium asked to remain anonymous because he feared trouble with his employer. His claims therefore remain one engineer's account of an undisclosed workplace, not an independently verified description of the company or software engineering generally.
AI Coding Can Save Time, but Review Work Piles Up
Survey data supports a narrower part of v0xium's argument: many teams report productivity gains from AI coding tools, while review and verification remain significant bottlenecks.
Black Duck and UserEvidence surveyed 831 software engineering and DevOps professionals in March 2026. The study found that 92% of teams reported improved productivity and release velocity from AI coding assistants, with an average reported saving of eight hours per week.
At the same time, 90% reported problems with AI-generated code. Manual review was the most common bottleneck at 52%, followed by security testing at 51%, code rework at 48% and repeated prompt iteration at 41%.
The survey also found a sharp difference by seniority. Seventy-four per cent of C-suite respondents reported a major productivity and velocity improvement, compared with 38% of technical contributors. The sample itself skewed senior, with 52% of respondents in C-suite or senior-leadership roles and 13% classed as technical contributors.
Anthropic Has Flagged 'Approval Fatigue'
Anthropic has identified a related problem in Claude Code's permission system.
On 25 March 2026, the company said Claude Code users approved 93% of permission prompts. Anthropic said repeated requests can create 'approval fatigue,' with users paying less attention to individual decisions over time.
The company introduced auto mode, which uses model-based classifiers to automate some approval decisions. Anthropic also cautioned that the system is not a replacement for careful human review on high-stakes infrastructure.
That does not verify v0xium's claims about his workplace. It does show that supervising an AI coding agent can introduce its own form of friction.
Developers Still Do Not Fully Trust AI Code
Sonar's 2026 State of Code Developer Survey, based on 1,149 professional developers surveyed in October 2025, found that respondents estimated AI accounted for 42% of committed code.
Yet 96% said they did not fully trust AI-generated code to be functionally correct. Only 48% said they always verified AI-assisted code before committing it, while 38% said reviewing AI-generated code required more effort than reviewing code written by a human colleague.
A randomised controlled trial from METR offered a different caution. Sixteen experienced open-source developers completed 246 tasks in mature projects they had worked on for an average of five years. With access to early-2025 AI tools, they took 19% longer on average than when AI use was not allowed.
METR warned against generalising that result. In February 2026, the group said developers were likely getting larger speed-ups from newer tools, but selection effects in its follow-up research made the size of any improvement too uncertain to estimate confidently.
The Bigger Question Is What Companies Do With the Speed
Anthropic's own June 2026 analysis of about 400,000 Claude Code sessions found that users made roughly 70% of planning decisions in a typical session, while Claude made about 80% of execution decisions. Sessions involving greater domain expertise were also more likely to succeed.
That distinction goes to the centre of v0xium's complaint. His argument is not that Claude Code should do less. It is that companies risk turning faster generation into higher output targets without leaving engineers enough time to understand, test and improve what the system produces.
The wider evidence does not show that Claude Code universally slows developers or strips meaning from their work. It does show that faster generation can move more pressure downstream into review, testing and verification.
For teams adopting AI coding tools, the more useful productivity question may be how quickly reliable software reaches users, not simply how quickly code is produced.
Frequently Asked Questions
- What is Claude Code?Claude Code is an AI tool used to assist in software development by generating code and other related documents.
- What are the reported benefits of using AI coding tools?AI coding tools reportedly improve productivity and release velocity, saving teams an average of eight hours per week.
- What challenges do AI coding tools present?Challenges include manual review bottlenecks, security testing, code rework, and repeated prompt iteration.
- What is 'approval fatigue' in the context of AI coding tools?Approval fatigue occurs when users become less attentive to individual decisions due to repeated permission prompts.
- Do developers fully trust AI-generated code?No, 96% of developers do not fully trust AI-generated code to be functionally correct.
