Engineers at large enterprise blame AI tool Claude Code for slowing shipments

Engineers at a large, unnamed corporation have raised alarms that the AI‑powered tool Claude Code is slowing shipments, despite management's claim that code pushes aren't a bottleneck. Half a month after the author joined, staff report 12‑ to 13‑hour days just to press enter, a concrete sign that the workflow is breaking down.

Key points

• Claude Code produces specs, code, tests, PRDs and tickets automatically.
• Team members describe the output as opaque and unreviewable.
• Management repeatedly says code‑push speed is not the limiting factor.
• Employees are logging 12‑13 hour days to complete minimal tasks.
• The frustration spans the entire engineering ladder, from L1 to L7.

What happened

The internal memo, collected by Simon Willison on September 20, 2026, describes a six‑week period in which a newly hired employee discovered that every artifact—specifications, test suites, PRDs, tickets and reports—was generated by Claude Code. The employee notes that “Nobody knows anything here,” highlighting a systemic lack of human‑authored documentation. Higher‑level managers reiterated that pushing code was not a bottleneck, yet engineers reported working “12 to 13 hours a day just to press enter.”

The situation escalated as the same AI‑generated outputs were handed off across teams, forcing engineers at every level to spend additional time deciphering, correcting, or re‑writing content that the tool produced. The result was a paradox: more code was being shipped, but the effort required to make each change grew, leading to missed deadlines and growing burnout.

The backstory

AI‑driven code generation has been touted as a productivity lever for enterprises since the early 2020s. Vendors promised that large language models could draft functional code, write tests, and even generate full product requirement documents, reducing the need for manual drafting. Early adopters reported mixed results—some saw modest speed gains, while others encountered quality and maintainability issues. The unnamed company appears to have doubled down on the technology, delegating the entire documentation pipeline to Claude Code without a parallel human review process.

Why the AI tool backfired

When an AI system produces every artifact without human oversight, the resulting output can become a black box. Engineers cannot validate assumptions, trace logic or assess test coverage without original context. The lack of transparency forces developers to spend time reverse‑engineering the AI's output, eroding the very efficiency the tool was meant to deliver. In this case, the promised speed of code pushes is offset by the hidden cost of deciphering AI‑generated work.

Who stands to gain or lose

Leadership gains a narrative of rapid deployment, which can please investors and board members seeking headline metrics. However, the engineering workforce bears the operational cost, experiencing longer workdays and reduced morale. Vendors of Claude Code benefit from continued licensing fees, but risk reputational damage if similar complaints proliferate across the industry. Competitors offering hybrid AI‑human workflows may capture market share by emphasizing auditability and developer control.

What builders and businesses should do

• Institute mandatory human review of AI‑generated specs, tests and tickets before release.
• Measure actual cycle‑time savings versus added effort to decode AI output.
• Align management expectations with data on code‑push bottlenecks.
• Provide training so engineers can effectively prompt and validate AI tools.

The bottom line

Watch for quarterly reports on AI‑assisted development efficiency; a widening gap between shipped code and developer hours will signal deeper adoption challenges.