What happened
Anthropic's essay "When AI builds itself," published in early June 2026 by The Anthropic Institute, closes with a two-sentence reflection from an unnamed employee that Business Insider singled out as capturing the state of AI confusion at work. The employee wrote: "On days where everything works well, I can't help but think nothing I do matters, everything is automated and better and faster than I ever will be. But then there are days where everything breaks and I don't understand why and I realize I have no idea what I've been up to anymore." Anthropic notes the quotes throughout the piece are drawn from internal discussions, used with permission, and reflect individual views as of May 2026 rather than official company positions.
What the essay reports
The quote sits inside a broader argument about recursive self-improvement. Anthropic writes that, as of May 2026, more than 80% of the code merged into its codebase was authored by its frontier LLM Claude, up from low single digits before Claude Code launched in February 2025, and that a typical engineer in the second quarter of 2026 was merging roughly 8x as much code per day as in 2024. The same essay states that "large performance gaps persist when it comes to Claude exercising judgement in choosing goals in both engineering and research." A separate employee is quoted saying it had been "~5 months since I last wrote any code myself," per Anthropic.
Why it resonated
Business Insider framed the two-sentence quote as crystallizing how knowledge workers feel as capable AI takes over more execution: useful and unsettling at once. The pairing of strong automation with brittle, hard-to-diagnose failures is exactly the tension the employee describes, and it tracks the essay's own claim that execution is increasingly automated while judgment and goal selection remain harder for current models.
Editorial analysis - what it means for practitioners
This is an industry-pattern observation, not a statement of any company's plans. Teams deploying advanced LLMs commonly see the same split the essay describes, where models handle routine synthesis, scaffolding, and well-specified tasks well but underperform on open-ended goal selection, error diagnosis, and multi-step reasoning. In practice that shifts the human role toward review, oversight, and debugging of opaque failures, and raises operational demand for observability, human-in-the-loop escalation, and test coverage for nonroutine scenarios rather than raw throughput.
What to watch
- •Whether independent benchmarks corroborate lab-reported productivity multipliers like the 8x code figure.
- •Whether review and verification, rather than code generation, become the binding constraint on AI-assisted teams.
- •How labs and adopters staff for oversight roles as more execution is automated.
Key Points
- 1Anthropic's essay "When AI builds itself" closes with an employee quote that Business Insider says captures workers' confusion as AI automates jobs.
- 2The same essay reports Claude now writes most of Anthropic's code but still shows large performance gaps in choosing goals and judgment.
- 3For practitioners, it signals AI increasingly handles execution while human judgment, oversight, and debugging of opaque failures remain the durable bottleneck.
Scoring Rationale
The story's primary source is a substantive Anthropic Institute essay on recursive self-improvement that discloses previously unreported internal data, over 80% of merged code now authored by Claude and roughly 8x more code per engineer since 2024, plus a candid admission that "large performance gaps persist" in model judgment. That lifts it above pure anecdote, though this event frames the softer workplace-confusion angle rather than the essay's harder data or its proposal for verifiable slowdown options. It is solidly relevant to practitioners deciding how much execution to delegate to AI.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
