MIT Sloan Management Review and BCG · Report
The Emerging Agentic Enterprise: How Leaders Must Navigate a New Age of AI
Agentic AI is “a tool that learns and a worker that is owned”, and that breaks how companies manage both.
- Published
- Reading time
- About 35 min
- Authors
- Sam Ransbotham, David Kiron, Shervin Khodabandeh, Sesh Iyer, Amartya Das
Full report is paywalled; the summary page is free.
What is it about?
The ninth annual MIT SMR and BCG AI study, based on 2,102 respondents in 116 countries, shows agentic AI spreading faster than earlier waves. Traditional AI took eight years to reach 72% adoption and generative AI three years to reach 70%; agentic AI reached 35% in two years, with another 44% planning to deploy.
Its core argument is managerial. Agents need HR-style supervision and asset-style investment management at the same time, which creates four tensions: scalability versus adaptability, experience versus expediency, supervision versus autonomy, and retrofit versus reengineer. The authors recommend an “HR for agents” function and cross-functional decision rights rather than leaving agents to IT.
Key numbers
- 76% of respondents see agentic AI as more like a coworker than a tool.
- Among extensive adopters, 45% expect fewer middle-management layers within three years, versus 30% of companies with no adoption plans.
- 66% of extensive adopters expect operating-model changes; 58% expect governance changes.
Why read it
- The four tensions make a solid skeleton for an AI strategy document.
- The “HR for agents” lifecycle (recruit, onboard, evaluate, retrain, retire) is a concrete governance model you can adapt.
- Interviews with Chevron, SAP, ADP, Goodwill, Capital One and Alibaba.com ground the framework in real operations.
Who should read it
- CIOs writing an agent strategy
- CHROs and COOs who will share ownership of agents
- Executives deciding whether to retrofit or redesign processes
About the author: Sam Ransbotham
Professor, Boston College; lead author of the MIT SMR–BCG AI series
Ransbotham trained as a chemical engineer at Georgia Tech, where his first paper, in 1988, simulated penicillin production. He then founded a software company whose clients included UN agencies such as the IAEA, FAO and WHO.
He earned his PhD 18 years after his MBA, with a dissertation on how technology innovation spreads, and did NSF-funded security research before turning to AI. He co-hosts the “Me, Myself, and AI” podcast with BCG's Shervin Khodabandeh and once wrote a chapter titled “Don't Let Artificial Intelligence Supercharge Bad Processes”.
Read it carefully: common misquotes
- “Adoption” includes piloting. It does not mean 35% run agents in production.
- The workforce figures are expectations, not observed cuts.
- The study is co-produced with BCG.
How it shapes the CIO's AI Playbook
- Step 4.2: Give every agent a human owner and a lifecycle · Govern agents like a workforce
- Step 4.3: Share decision rights across business, HR and IT · Govern agents like a workforce
Notes by Pavel Veselov, last reviewed . Summaries are my own; quotes and figures belong to the original authors.
All 11 sources in the CIO's AI Library