Agentic AI hit the enterprise — most companies aren't ready

Bain, McKinsey, and MIT Sloan published agentic AI reports in the same week.

September 2025 was the month agentic AI stopped being a demo. Bain, McKinsey, and MIT Sloan/BCG all published reports in the same week: AI agents are entering the workforce, and your architecture isn't ready.

The numbers

  • 5% → 40%. Gartner: AI agents in 40% of enterprise apps by end 2026.
  • 62% experimenting. McKinsey: 62% of organisations already working with AI agents.
  • 76% of executives view agentic AI as more colleague than tool.

Three problems nobody talks about

  1. 90% fail in production. Agents built for happy paths crumble when APIs fail.
  2. Legacy architecture kills agents. Need clean APIs, event-driven systems.
  3. Governance is afterthought. Covers what AI says, not what AI does.

What to do

  1. Start with workflow agents. Bounded processes with human checkpoints.
  2. Map tool access. Restrict by default.
  3. Build an agent registry. Shadow AI agents are the next shadow IT.
  4. Invest in error handling before features.