McKinsey's 30% automation projection rests on fundamental misreadings of current agent capabilities. Their model assumes seamless chaining of legal contract review, financial analysis, and code generation while ignoring real-world enterprise IT complexity. Salesforce's Agentforce deployments required average 23 hours of manual rule configuration and exception handling per 'successful' customer service automation case. More critically, McKinsey equates 'partial automation' with 'job elimination'—a lawyer using AI-assisted drafting with 40% productivity gain doesn't disappear; the legal services market likely expands. History is instructive: spreadsheets in the 1980s 'destroyed' bookkeepers while creating the entire financial analysis profession.
McKinsey's projection is actually conservative rather than overstated. Their 30% figure covers only clearly definable structured knowledge work, while AI agent penetration is rapidly spreading into semi-structured domains. Microsoft 365 Copilot achieved 67% enterprise adoption by Q4 2025, with 'deep research' functionality replacing average 4.2 weekly hours of junior analyst information gathering. More disruptive are vertical agents: Harvey reaches 91% associate-level accuracy in legal due diligence with 20× processing speed. Headcounts may not collapse, but skill premium structures are violently restructuring—junior role demand is deteriorating faster than McKinsey models assume.
The core problem is McKinsey conflating 'technical feasibility' with 'organizational adoption speed' timelines. Their technical assessment is largely accurate: current GPT-5.4-class models with tool use can indeed handle 30% of knowledge work subtasks. But organizational transformation involves process restructuring, compliance review, cultural adaptation, and retraining investments—friction coefficients severely underestimated. MIT Sloan's 2025 tracking study of 500 firms found AI tool technical deployment averaged 8 months while organizational integration required 31 months. More subtle are power dynamics: middle managers systematically resist AI that makes their decision processes transparent and auditable. McKinsey's 2028 timeline needs extension to at least 2032 for organizational reality.