McKinsey

Martin Harrison and Natasha Maniar set the frame in November 2025: coding agents produce individual gains, yet many large companies plateau at 5 to 15 percent overall improvement, because the bottleneck moved to manual code review. Jared Moon, Rory Walsh and Vito Di Leo carry the argument into delivery in Rewiring software delivery for the agentic era: CI/CD automates roughly 30 percent of the spend, while the requirements-to-code 70 percent stays manual. Chase Covington's article The agentic transformation office puts figures on the back office: 35 to 40 percent of time saved on core tasks, 70 percent or more in some cases, one central team cut from a manager and nine staff to a manager and two after ten processes were automated. Asutosh Padhi's article The AI operating system moves to governance: own the context, rent or replace the technology, hold the enterprise to one, two or three AI OS, fund the shared layer rather than isolated use cases.

Those same documents carry counterweights. The State of AI 2026 survey finds 40 percent of large companies running agents at scale and 6 percent attributing at least 5 percent of EBIT to AI. Harrison and Maniar report 70 percent of companies have not rewritten job descriptions. Covington warns that one data error can propagate across connected agents.

Elsewhere McKinsey appears as the number other authors borrow: 5,200 billion dollars of datacenter capex by 2030, 3 to 5 trillion dollars of agentic commerce revenue by 2030, 500 percent productivity in flow state. The methodology behind those projections is never given.