AI tools churn out 'workslop', but 'the buck' should stop with bosses (www.theguardian.com)

A string of industry studies — KPMG (only 8.5% “always” trust AI search results), Gartner (over half of consumers distrust AI searches), McKinsey (80% of firms saw no meaningful bottom-line impact; 42% abandoned projects), MIT (95% of large-company pilots failed) — sets the stage for a Harvard Business Review finding that more than 40% of US full-time employees have been handed AI-generated “workslop”: outputs that look competent but lack the substance to advance work and end up destroying productivity. The story reframes the problem away from “bad models” or hype: AI can produce plausible but shallow results, and unchecked adoption amplifies errors and wastes time. The piece’s core argument is that responsibility lies with employers, not just vendors. Effective deployment requires investment in training (including prompt engineering), standardized assistants and app governance, a designated owner for AI tools, explicit policies on acceptable use, integration into workflows, and measurable KPIs for impact. For the AI/ML community this underscores real-world gaps: robustness and evaluation beyond benchmark scores, human-in-the-loop systems, MLOps for monitoring and quality control, domain-specific fine-tuning, and tooling that supports explainability and verifiable outputs. In short, models aren’t magic — organizational processes, measurement, and skilled users are essential to turn AI from one-off “workslop” into reliable productivity gains.