• Home
  • Latest
  • Fortune 500
  • Finance
  • Tech
  • Leadership
  • Lifestyle
  • Rankings
  • Multimedia
AICollaboration

Are you a cyborg, a centaur, or a self-automator? Why businesses need the right kind of ‘humans in the loop’ in AI

By
François Candelon
François Candelon
,
Katherine Kellogg
Katherine Kellogg
,
Hila Lifshitz
Hila Lifshitz
, and
Steven Randazzo
Steven Randazzo
Down Arrow Button Icon
By
François Candelon
François Candelon
,
Katherine Kellogg
Katherine Kellogg
,
Hila Lifshitz
Hila Lifshitz
, and
Steven Randazzo
Steven Randazzo
Down Arrow Button Icon
January 30, 2026, 5:30 AM ET
The "human-in-the-loop" approach manifests in three radically different ways, with profoundly different implications for performance and skill development
The "human-in-the-loop" approach manifests in three radically different ways, with profoundly different implications for performance and skill developmentGetty Images

As generative AI rapidly spreads through organizations, executives face a deceptively simple question: How should humans work with AI? The common answer—”keep humans in the loop”—sounds reassuring.

But new research reveals that this answer is dangerously incomplete. What appears to be the same “human-in-the-loop” approach actually manifests in three radically different ways, with profoundly different implications for performance and skill development.

To understand how companies can truly extract value from human-AI collaboration, we conducted a field experiment with 244 consultants using GPT-4 for a complex business problem-solving task. With support from scholars at Harvard Business School, the MIT Sloan School of Management, the Wharton School, and Warwick Business School, the experiment analyzed nearly 5,000 human-AI interactions to answer a critical question: When humans collaborate with GenAI, what are they actually doing—and what should they be doing?

Three hidden patterns of human-AI collaboration

Our experiment’s most striking finding is that professionals working with GenAI naturally sorted themselves into three distinct collaboration styles—each with dramatically different outcomes:

Cyborgs (60% of participants) engaged in what we call “Fused Knowledge Co-Creation”—a continuous, iterative dialogue with AI throughout the entire workflow. They used it for each sub-task in their workflow and in different ways: They assigned personas to the AI, broke complex tasks into modules, pushed back on AI outputs, exposed contradictions, and validated results in a dynamic back-and-forth. For Cyborgs, the boundary between human and AI thinking became deliberately blurred.

Centaurs (14% of participants) practiced “Directed Knowledge Co-Creation”—using AI selectively for specific subtasks while maintaining firm control over the overall problem-solving process. They leveraged AI to enhance their capabilities, to map problem domains, gather methodological information, and refine their own human-generated content. But they kept themselves firmly in the driver’s seat, using AI as a targeted tool rather than a collaborative partner.

Self-Automators (27% of participants) engaged in “Abdicated Knowledge Co-Creation”—delegating entire workflows to AI with minimal iteration or critical engagement. They provided data and instructions to AI to conduct the sub-tasks, then accepted its outputs without modification or with only small edits. Their work was fast and polished but lacked depth—resembling outputs completed for them rather than with them.

What’s remarkable is that every participant had access to the same tools and the same task. They did not receive any different instructions about the work process with AI. Yet their emergent/instinctive choices about when to engage AI and how much authority to give it produced fundamentally different collaboration dynamics.

A framework for understanding collaboration

To make sense of these patterns, we developed a framework built around two fundamental questions that structure any collaborative problem-solving dynamic between human and machine: Who selects what needs to be done? and Who identifies how it gets done?

Cyborgs let humans drive the “what” but allow AI significant control over “how.” Centaurs retain human control and leadership over both dimensions, using AI only for targeted assistance. Self-Automators cede control of both to AI. Notably, the fourth theoretical possibility—where AI drives task selection but humans drive execution—remained empty in our study; when professionals surrender control over what to work on, they also tend to abdicate control over how to do it.

The hidden cost: What happens to expertise?

Perhaps our most consequential finding concerns what happens to professional expertise under each collaboration mode. The implications diverge dramatically:

Cyborgs developed new AI-related expertise—what we call “newskilling.” Through continuous experimentation with prompting strategies, they learned how to effectively communicate with AI, when to push back, and how to extract maximum value from the collaboration. They also maintained their domain expertise by staying actively engaged throughout the process.

Centaurs deepened their domain expertise—traditional “upskilling.” By using AI to accelerate learning about unfamiliar industries, gather methodological guidance, and refine their own thinking, they built stronger foundational capabilities. However, they did not develop significant AI-related expertise because their interactions with AI were limited and targeted.

Self-Automators developed neither—experiencing what we call “no skilling.” By delegating the entire cognitive process to AI, they missed opportunities to build either domain knowledge or AI fluency. Their productivity gains came at the cost of professional development.

This finding should give executives pause. When employees default to Self-Automator behavior—which over a quarter of our highly trained consultants did—organizations may be inadvertently hollowing out the very expertise that creates competitive advantage.

Performance implications: Who gets it right?

Our experiment evaluated outputs on two dimensions: accuracy (did they recommend the correct brand?) and persuasiveness (how compelling was the CEO memo?). The results challenge simplistic assumptions about AI collaboration:

Centaurs achieved the highest accuracy—outperforming both Cyborgs and Self-Automators on getting the right answer. By maintaining control over the analytical process and using their own judgment to evaluate AI inputs, they avoided being led astray by AI’s confident but sometimes incorrect recommendations.

Both Cyborgs and Centaurs excelled in persuasiveness—producing more compelling outputs than Self-Automators. The depth of engagement, whether through iterative refinement (Cyborgs) or human-driven analysis (Centaurs), translated into higher-quality deliverables.

Notably, Cyborgs sometimes fell victim to AI’s persuasiveness. Even when they employed best practices like validation—asking AI to check its own work—they were sometimes convinced by AI’s confident justification of incorrect answers. This highlights a critical risk: sophisticated engagement with AI doesn’t guarantee immunity from its errors.

What should companies do right now?

These findings have immediate implications for how organizations deploy GenAI:

First, abandon the myth of a single “human-in-the-loop” approach. Executives must recognize that their employees are already adopting dramatically different collaboration styles—and that these differences matter. Simply mandating “human oversight” without specifying what that means will produce wildly inconsistent results.

Second, match collaboration styles to strategic objectives. For tasks requiring maximum accuracy on high-stakes decisions, encourage Centaur behavior—selective AI use with strong human judgment. For tasks requiring rapid iteration and creative exploration, Cyborg behavior may be more appropriate. Reserve Self-Automator approaches for truly routine tasks, not the core or risky ones, and where skill development is not a concern.

Third, monitor for automation complacency. The 27% Self-Automator rate in our study—among highly skilled, motivated professionals who knew their performance was being evaluated—suggests that the temptation to over-delegate is powerful. Organizations must develop mechanisms to detect when employees are sliding toward full automation on tasks that require human engagement.

Fourth, rethink how you measure AI adoption success. Using only final outcomes—like edit rates or acceptance ratios—as proxies for engagement is insufficient. A Self-Automator who accepts AI output and a Cyborg who iterates extensively then accepts a refined version may look identical in the data. Companies need to track the quality of interaction throughout the workflow, not just the result.

Fifth, invest in developing AI fluency alongside domain expertise. Our findings suggest that the most sustainable approach combines both. Cyborg behavior builds advanced AI skills while maintaining domain knowledge; Centaur behavior builds domain skills while providing baseline AI exposure. Companies need training programs that develop both capabilities deliberately, rather than hoping employees will figure it out on their own.

The stakes: Expertise in the Age of AI

The emergence of GenAI presents organizations with a paradox. The technology promises to elevate human judgment, creativity, and speed, but it also carries a quieter risk: that in handing more thinking to machines, professionals may slowly give up the very capabilities that make them valuable. The same tools that sharpen expertise in some hands can, in others, replace it entirely, leaving organizations with impressive outputs short term but a thinning core of human judgment. This is not merely another efficiency tool, this is a revolution.The good news is that productive collaboration modes exist. Cyborgs and Centaurs demonstrate that humans can work effectively with AI while building, rather than depleting, their expertise. The challenge for executives is to create organizational conditions that encourage these productive patterns while discouraging the seductive but self-defeating path of full automation.

As AI capabilities continue to expand and improve, the organizations that thrive will be those that master not just what AI can do, but how humans should work with it. Understanding that “human-in-the-loop” is not a single approach but actually three  fundamentally three different collaboration modes—with fundamentally different consequences—is the first step toward building that mastery.

François Candelon is a partner at private equity firm Seven2 and executive fellow at D^3 Institute at Harvard. Read other Fortune columns by François Candelon.

Katherine Kellogg is the David J. McGrath Jr. Professor of Management and Innovation at the MIT Sloan School of Management.

Hila Lifshitz is professor of management at Warwick Business School, faculty associate at the Harvard Laboratory for Innovation Science, and the co-director of the AI Innovation Network.

Steven Randazzo is a PhD student at Warwick Business School, visiting researcher at the Harvard Laboratory for Innovation Science, and the Co-Director of the AI Innovation Network.

Join us at the Fortune Workplace Innovation Summit May 19–20, 2026, in Atlanta. The next era of workplace innovation is here—and the old playbook is being rewritten. At this exclusive, high-energy event, the world’s most innovative leaders will convene to explore how AI, humanity, and strategy converge to redefine, again, the future of work. Register now.
About the Authors
By François Candelon
See full bioRight Arrow Button Icon
By Katherine Kellogg
See full bioRight Arrow Button Icon
By Hila Lifshitz
See full bioRight Arrow Button Icon
By Steven Randazzo
See full bioRight Arrow Button Icon

Latest in AI

Finance
Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam
By Fortune Editors
October 20, 2025
Finance
Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam
By Fortune Editors
October 20, 2025
Finance
Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam
By Fortune Editors
October 20, 2025
Finance
Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam
By Fortune Editors
October 20, 2025
Finance
Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam
By Fortune Editors
October 20, 2025
Finance
Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam
By Fortune Editors
October 20, 2025

Most Popular

Finance
Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam
By Fortune Editors
October 20, 2025
Finance
Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam
By Fortune Editors
October 20, 2025
Finance
Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam
By Fortune Editors
October 20, 2025
Finance
Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam
By Fortune Editors
October 20, 2025
Finance
Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam
By Fortune Editors
October 20, 2025
Finance
Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam
By Fortune Editors
October 20, 2025
Rankings
  • 100 Best Companies
  • Fortune 500
  • Global 500
  • Fortune 500 Europe
  • Most Powerful Women
  • Future 50
  • World’s Most Admired Companies
  • See All Rankings
Sections
  • Finance
  • Leadership
  • Success
  • Tech
  • Asia
  • Europe
  • Environment
  • Fortune Crypto
  • Health
  • Retail
  • Lifestyle
  • Politics
  • Newsletters
  • Magazine
  • Features
  • Commentary
  • Mpw
  • CEO Initiative
  • Conferences
  • Personal Finance
  • Education
Customer Support
  • Frequently Asked Questions
  • Customer Service Portal
  • Privacy Policy
  • Terms Of Use
  • Single Issues For Purchase
  • International Print
Commercial Services
  • Advertising
  • Fortune Brand Studio
  • Fortune Analytics
  • Fortune Conferences
  • Business Development
About Us
  • About Us
  • Editorial Calendar
  • Press Center
  • Work At Fortune
  • Diversity And Inclusion
  • Terms And Conditions
  • Site Map
  • Facebook icon
  • Twitter icon
  • LinkedIn icon
  • Instagram icon
  • Pinterest icon

Most Popular

placeholder alt text
C-Suite
OpenAI’s Sam Altman says his highly disciplined daily routine has ‘fallen to crap’—and now unwinds on weekends at a ranch with no cell phone service
By Jacqueline MunisFebruary 5, 2026
2 days ago
placeholder alt text
Politics
Meet the Palm Beach billionaire who paid $2 million for a private White House visit with Trump
By Tristan BoveFebruary 3, 2026
4 days ago
placeholder alt text
Travel & Leisure
How Japan replaced France as the country young Americans obsessively romanticize—they’re longing for civility they don’t see at home
By Nick LichtenbergFebruary 5, 2026
2 days ago
placeholder alt text
Success
After decades in the music industry, Pharrell Williams admits he never stops working: ‘If you do what you love everyday, you’ll get paid for free'
By Emma BurleighFebruary 3, 2026
4 days ago
placeholder alt text
Investing
Ray Dalio warns the world is ‘on the brink’ of a capital war of weaponizing money—and gold is the best way for people to protect themselves
By Sasha RogelbergFebruary 4, 2026
2 days ago
placeholder alt text
Crypto
Bitcoin whales and ETFs are baling out of the market; UBS warns: ‘Crypto is not an asset’
By Jim EdwardsFebruary 6, 2026
17 hours ago

© 2026 Fortune Media IP Limited. All Rights Reserved. Use of this site constitutes acceptance of our Terms of Use and Privacy Policy | CA Notice at Collection and Privacy Notice | Do Not Sell/Share My Personal Information
FORTUNE is a trademark of Fortune Media IP Limited, registered in the U.S. and other countries. FORTUNE may receive compensation for some links to products and services on this website. Offers may be subject to change without notice.


Latest in AI

CEO and co-founder of Anthropic Dario Amodei speaking on stage.
AIAnthropic
Anthropic’s newest model excels at finding security vulnerabilities—but raises fresh cybersecurity risks
By Beatrice NolanFebruary 6, 2026
8 hours ago
Google data center
Big TechData centers
Big Tech’s $630 billion AI spree now rivals Sweden’s economy, unsettling investors: ‘We’ve never invested this much on anything before’
By Jake AngeloFebruary 6, 2026
10 hours ago
AISocial Network
Moltbook, the Reddit for bots, alarms the tech world as agents start their own religion and plot to overthrow humans
By Kaitlyn Huamani and The Associated PressFebruary 6, 2026
10 hours ago
AISpaceX
Musk predicts more AI capacity will be in orbit than on earth in 5 years, with SpaceX becoming a ‘hyper-hyper’ scaler
By Jason MaFebruary 6, 2026
11 hours ago
Amazon founder Jeff Bezos
SuccessBillionaires
Larry Ellison and Jeff Bezos have seen more than $66 billion swiped from their net worths since the start of this year as AI-driven slump sees tech billionaires’ wealth free-fall
By Emma BurleighFebruary 6, 2026
12 hours ago
Dario Amodei, Anthropic CEO, speaking.
AIAnthropic
Anthropic’s Claude triggered a trillion-dollar selloff. A new upgrade could make things worse
By Beatrice NolanFebruary 6, 2026
13 hours ago