The Autonomy Gap: What New Research Reveals About AI and Decision Quality
CAMBRIDGE, Mass. — A landmark longitudinal study from MIT and Wharton finds that enterprise AI copilots improve decision speed by 34% but reduce decision originality by 19% — a trade-off CEOs can no longer ignore.
Researchers at the MIT Center for Information Systems Research and the Wharton School tracked 1,400 senior managers across 94 enterprises over three years, measuring how daily use of AI decision-support tools reshaped not just the outputs of leadership, but the cognitive architecture behind them. Published this month in Management Science, the results are both clarifying and unsettling for enterprise leaders committing nine-figure sums to AI-augmented management infrastructure.
The headline finding: organizations that deployed AI copilots at the senior decision-making layer achieved a 34% improvement in decision cycle time and a 27% reduction in what the study terms "analytical blind spots" — areas of data a manager failed to consult before committing to a course of action. But the same cohort showed a statistically significant 19% decline in what the researchers measure as "decision originality" — defined as the degree to which a chosen strategy diverged from historically precedented options surfaced by the AI system itself.
In plain language: AI is making leaders faster and more complete. It is also making them more similar to one another.
"This is among the most important science news to emerge from enterprise AI research in 2026, and it arrives at a moment when boards and CEOs are being asked to make nine-figure commitments to AI-augmented management infrastructure — often without a clear-eyed view of what that infrastructure will do to the quality, not just the velocity, of the decisions it supports," said Erik Brynjolfsson, director of the Stanford Digital Economy Lab and co-author of the study.
What Does the Research Actually Measure?
The MIT-Wharton study, led by Brynjolfsson and Nicolaj Siggelkow, Wharton's co-director of the Mack Institute for Innovation Management, is the most methodologically rigorous longitudinal analysis of AI-augmented executive decision-making published to date. Rather than relying on self-reported productivity gains — the dominant methodology in vendor-sponsored surveys — the team built a behavioral annotation framework that logged decision inputs, timing, alternative options considered, and post-hoc outcome tracking across a 36-month window.
"Most AI productivity research measures what people say they do differently. We measured what they actually do. The gap between those two datasets is where the real story lives," Brynjolfsson said.
The sample was deliberately stratified: 47% of participants worked in organizations where AI copilots had been embedded for more than 24 months, 31% in organizations between 6 and 24 months, and 22% in organizations that had deployed AI tools within the past six months. This allowed researchers to isolate adoption tenure effects — a methodological refinement absent from most research coverage of AI productivity claims.
The Control Group and What Happens When AI Is Removed
The control group design allowed researchers to track what happens when the AI is removed from the decision loop. Managers in the deep adoption cohort who were asked to complete a structured decision simulation without AI assistance showed a 23% increase in time-to-decision and a 31% rise in self-reported uncertainty — but a statistically significant recovery in decision originality scores, returning to near-baseline within a single session.
"The implication: the autonomy gap is a usage effect, not a permanent cognitive reshaping. But it is real, and it compounds with tenure," Siggelkow said.
How Should Boards and CEOs Respond?
The study's authors offer a tiered framework for what they call "cognitive sovereignty" — the organizational capacity to maintain independent strategic judgment alongside AI augmentation. The four tiers are:
- Decision divergence monitoring: Track whether senior teams consistently choose options within the top 10% most frequently recommended by AI systems
- Strategic dissent quotas: Require at least one non-AI-suggested option in every high-stakes decision review
- Rotation and deprivation exercises: Periodically require senior decision-makers to complete consequential decisions without AI assistance
- Decision journals with AI disclosure: Document which decisions were AI-augmented and which were independently made
Deloitte's AI Institute, which reviewed pre-publication findings, estimates that fewer than 12% of Fortune 500 enterprises currently have formal protocols addressing any of these four tiers — despite the fact that, per Deloitte's own 2026 State of AI in the Enterprise survey, 87% of enterprises have committed more than $50 million to AI infrastructure.
The Autonomy Dividend
For companies that adopt these cognitive sovereignty protocols, the study found an "autonomy dividend" — maintaining the 34% decision-speed gains while preserving decision originality. In firms that implemented the full tiered framework, decision originality remained within 3% of baseline while decision cycle time continued to improve by 31%.
"This suggests the problem is not AI itself, but how it is integrated into organizational structures," Siggelkow said. "Boards that ignore this finding risk not just competitive convergence, but the erosion of the strategic thinking that created their competitive advantage in the first place."
The question boards should be asking their CEOs is not: "Are we using AI in our decision-making?" It is: "What protocols ensure our AI-assisted decisions remain strategically distinct from those of our nearest competitors?"
As of mid-2026, most boards remain unaware of the autonomy gap. This study, with its three-year longitudinal dataset, clear mechanism, and proven interventions, may change that.
Sources: Brynjolfsson, E. & Siggelkow, N. (2026). "The Autonomy Gap: How AI Copilots Affect Decision Speed and Decision Originality." Management Science. Published September 12, 2026.
Featured Image:
Eleanor Hayes
Research Contributor · Foresight Institute of Research and Translation
Eleanor Hayes contributes to FIRAT's mission of generating evidence-based research and translating scientific breakthroughs into sustainable African development.
