The Emerging Agentic Enterprise (MIT SMR–BCG, November 2025)
TL;DR
The ninth annual global study on AI and business strategy by MIT Sloan Management Review and Boston Consulting Group, by Sam Ransbotham, David Kiron, Shervin Khodabandeh, Sesh Iyer and Amartya Das. It is about agentic AI, not about maturity or value.
- Adoption, by type of AI. Traditional AI adoption “has climbed to 72%” (2025). Generative AI reached 70% in three years. Agentic AI is at 35%, with “another 44%” planning to deploy it soon (p. 7, Figure 1).
- The tool-coworker duality. 76% of respondents view agentic AI “as more like a coworker than a tool” (p. 5). The report builds four tensions on this: scalability vs adaptability, experience vs expediency, supervision vs autonomy, retrofit vs reengineer (pp. 8–9).
- Expected autonomy rises. The share agreeing that AI systems have decision-making authority goes from 10% now to 35% expected in three years (+250%). Working independently from humans goes from 14% to 39% (p. 13, Figure 3).
- Organisational change expected by extensive adopters. 66% expect operating-model changes (42% of non-adopters), 58% governance changes, 45% fewer middle-management layers, 43% plan to hire more generalists, 29% expect fewer entry-level roles (pp. 16–19, Appendix Figure 1).
- No maturity share. The 2024 segmentation (15% “Augmented Learners”) is not repeated. The report gives no share of organisations achieving significant value or financial benefit.
Key data
| Measure | Value | Page |
|---|---|---|
| Respondents | 2,102, more than 21 industries, 116 countries; survey fielded spring 2025; 11 executive interviews | p. 5 |
| Organisations using traditional AI (“piloting or deploying AI solutions”) | 46% (2017) → 50% (2023) → 70% (2024) → 72% (2025) | p. 7, Fig. 1 |
| Generative AI adoption | 70% “in just three years” | p. 7 |
| Agentic AI adoption | 35%, plus 44% planning to deploy soon | p. 7 |
| See agentic AI as more like a coworker than a tool | 76% | p. 5 |
| Extensive agentic adopters: AI increases ability to stand out | 73% | p. 8 |
| AI systems have decision-making authority: now → in 3 years | 10% → 35% | p. 13 |
| Extensive adopters expecting operating-model change (vs no plans) | 66% (vs 42%) | p. 16 |
| Extensive adopters expecting governance-structure change in 3 years (vs no plans) | 58% (vs 37%) | p. 22 |
| Extensive adopters expecting fewer middle-management layers (vs no plans) | 45% (vs 30%) | pp. 17, 23 |
| Extensive adopters: AI positively affects job satisfaction | 95% | p. 19 |
| ”I hope AI will do some of my tasks” / “I fear …”, all adoption levels | 78–85% / 21–32% | p. 30 |
Method and definitions
- Sample. A global survey fielded in spring 2025, “records from 2,102 respondents representing more than 21 industries and 116 countries”, plus interviews with 11 executives (p. 5). The report calls it a “global executive survey” (p. 5). It does not describe the sampling frame or response rate.
- “Adoption” in Figure 1 is the share of organisations “piloting or deploying AI solutions” (p. 7). Pilots count as adoption.
- Segmentation. Respondents are grouped by agentic-AI adoption: no plans, plans to adopt, piloting, some adoption, extensive adoption (appendix figures, pp. 22–33). The report does not give the share of respondents in each group, and it does not define “extensive”.
- Sponsor and authors. “In collaboration with and sponsored by Boston Consulting Group” (p. 2). Khodabandeh, Iyer and Das are BCG partners or principals.
Scope and reliability
- Self-report, by individual respondents about their organisation and their own jobs. Many figures are expectations (“in three years”), not observed outcomes.
- No value measure. Unlike the 2024 edition (15% Augmented Learners; financial-benefit questions), this edition reports no share of organisations getting significant value. It cannot be set beside the maturity bars of the thread as a newer measurement of the same thing.
- Series continuity. The adoption series is consistent with the 2024 report (50% in 2023, 70% in 2024). The 2025 sample (2,102 respondents, 116 countries) is smaller than 2024’s (3,467, 136 countries).
- Small print. Reference 3 dates a WSJ story “May 12, 2005” (p. 35); the context places it in 2025.
Dynamic-capabilities reading
digital-sensing/digital-scouting: an annual survey tracking how fast traditional, generative and agentic AI spread through organisations. Its sidebar reads the speed of agentic adoption through diffusion-of-innovations theory (p. 8).digital-seizing/balancing-digital-portfolios: the “experience vs expediency” tension is an investment-timing and portfolio question. The report argues that standard net-present-value and depreciation models fail for systems that improve with use (pp. 11–12).digital-transforming/redesigning-internal-structures: the second half prescribes structural redesign: governance and decision rights, fewer management layers, dual career paths, and finance, HR and IT working jointly (pp. 15–21).
How it connects
- 2026-05-07-ransbotham-augmented-learners: the eighth edition (2024). Same survey series and lead authors. Its 15% Augmented Learners figure is the Ransbotham/Kiron bar on enterprise-ai-adoption. This edition does not re-measure it.
- 2026-05-07-kiron-schrage-compound-benefits: Kiron and Schrage’s column on compounding gains. This report makes a related point about agents as assets that “appreciate” with experience (p. 12).
- ai-maturity-measurement-comparison: as of November 2026, the newest MIT SMR–BCG report reports adoption, not a value or maturity share.
- technology-adoption-theories: the diffusion-of-innovations sidebar (p. 8).
- ai-agents, agent-oversight-and-delegation, automation-vs-augmentation: the tool-coworker duality and the supervision vs autonomy tension.
What was actually ingested
The full 40-page PDF from BCG’s asset server. bcg.com answers curl with HTTP 403, so the file was retrieved with the WebFetch tool and converted locally with pdftotext -layout. Printed page n is PDF page n + 4. Charts are flattened in the text extract. The BCG press release of 18 November 2025 (WebFetch) confirms the date and the 2,102 / 116 sample, and mentions neither Augmented Learners nor a value share.
Linked entities and concepts
Entities: Sam Ransbotham, David Kiron, MIT Sloan Management Review, Boston Consulting Group.
Concepts and threads: enterprise-ai-adoption, ai-maturity-measurement-comparison, ai-agents, agent-oversight-and-delegation, automation-vs-augmentation, technology-adoption-theories.
Dangling (single-source mention, deferred): Sesh Iyer (BCG, BCG X), Amartya Das (BCG Henderson Institute), interviewees Margery Connor (Chevron), Naomi Lariviere (ADP), Prem Natarajan (Capital One), Steve Preston (Goodwill Industries), Walter Sun (SAP), Kuo Zhang (Alibaba.com).