AI: The slacker's dream come true (www.theregister.com)

A provocative opinion piece argues that the corporate rollout of generative AI—exemplified by Microsoft Copilot—has not reliably raised productivity and instead is reshaping work by making it easier to produce plausible‑looking but low‑substance outputs that clog communication and processes. Citing a Stanford survey of U.S. companies and a UK trial of M365 Copilot, the piece highlights rising anxiety, confusion and annoyance: AI is increasingly used to generate internal emails and documents that are hard to act on, cost organizations “dollars per worker per month” to manage, and lower trust in other people’s work. The author frames this not merely as a technological failing but as a new enabler of “work avoidance,” with LLMs perfectly suited to crafting the kind of evasive, confirm‑seeking messages that slow decision cycles. For AI/ML practitioners and managers the takeaway is technical and organizational. At the model level, large language models excel at surface plausibility—recasting facts into coherent narratives—while often lacking verification, grounding and actionability, which amplifies downstream friction. At the deployment level, vendor pushes (e.g., Copilot adoption even around IT controls) can accelerate misuse without governance, prompting ethical and productivity trade‑offs. The piece suggests a need for tighter evaluation focused on actionable outputs, stronger guardrails and workplace policies that align AI capabilities with real work flows rather than inadvertently optimizing for plausible shirking.