The Cybernetic Teammate and the Completion Problem

When AI can always generate another version, knowing when to stop becomes part of the job.

In content and marketing, where so much of the job already depends on feedback and revision, that shift is especially noticeable. Another opinion can arrive in seconds, which means the difficult part can move from producing the work to deciding when it no longer needs improvement. Since completing my master’s last year, I have been looking for ways to keep learning outside of work, particularly around changes that are already affecting how I work professionally. That recently led me to Professor Charles Ayoubi’s Scaling AI at Work: How Leaders Can Turn Experimentation Into Business Impact masterclass, where he presented research examining how generative AI is changing the way people work alongside technology.

When The Tool Joins The Team 

Ayoubi, an assistant professor at ESSEC business school, presented several studies, including The Cybernetic Teammate: A Field Experiment on Generative AI and Teamwork, co-authored with Fabrizio Dell’Acqua, Ethan Mollick and other researchers. The experiment analysed 791 professionals at Procter & Gamble working on real product innovation challenges, either individually or in two-person teams and with or without generative AI. 

Individuals working with AI matched the performance of two-person teams without it. AI also helped bridge the divide between professional specialisms. Without AI, research and development (R&D) professionals tended to produce more technically oriented ideas while commercial professionals leaned towards commercially oriented ones. With AI, their solutions became more balanced regardless of professional background. Together, those findings support the researchers framing of AI as something that can reproduce certain benefits of a human treatment, rather than simply operating as a piece of software, hence coining it a “cybernetic teammate” (Dell’Acqua et al., 2026).

That framing felt familiar because it is close to how I already use AI in my own work. When developing a piece of content, I might use it to test whether an argument makes sense or see whether I have overlooked something. Its usefulness comes from the back-and-forth. I can respond to a suggestion, add context and continue working through the idea. In that sense, the comparison with a teammate feels increasingly accurate. What interests me more is what happens when that teammate’s feedback is limitless.

The Productivity Paradox

Content rarely has one objectively correct final version. A sentence can almost always be rewritten and an argument can usually be structured another way without the original necessarily being wrong. Before generative AI, those possibilities came with more friction. Feedback depended on another person having the time to review the work, while further revision meant deciding that the additional effort was worth it. Eventually, practical limits helped force a decision that something was ready, which also meant learning to trust your own judgment.

Generative AI changes that calculation because the time cost of another opinion has become almost negligible. I can ask it to review a paragraph and receive a suggestion within seconds. After making the change, I can ask again and receive another, then another, and then… another. 

The existence of another suggestion can begin to feel like evidence that the work still needs work, when in reality it may only mean that another possible version merely exists. For someone working in content, that distinction matters because there are often many ways to communicate the same idea successfully. When those alternatives become unlimited, generating another option becomes easy while deciding whether it is actually better becomes harder.

That is the productivity paradox I have started to notice in my own work. AI can make research faster and help me work through an idea more quickly. Yet some of the time “saved” earlier in the process can return at the end, when a piece of work that would previously have been finished stays open for another round of AI-generated feedback.

That is where my own AI fatigue comes from. The problem is less the effort involved in one revision than knowing another revision is always possible.

It reminds me of playing with a cat and a laser pointer. The movement is what keeps the cat interested, but eventually you have to let them catch it. Without that point of completion, the target simply keeps moving. AI-assisted work can create the same effect. Being able to continue the feedback loop is part of what makes the tool useful, but I still have to decide when the chase is over. The longer the target keeps moving, the easier it becomes to spend energy on work that has already achieved what it needed to achieve.

Knowing When to stop 

The study looks at what AI can contribute within a defined task. Everyday work is less contained. The researchers themselves note that their one-day experiment could not fully capture the iterative rework cycles that develop inside organisations over time (Dell’Acqua et al., 2026). A one-day workshop can show what AI contributes to a task. It cannot show what happens when that feedback loop is available every day.

As AI becomes embedded in everyday knowledge work, prompting is only part of the skill involved in using it well. The harder part is evaluative judgment: recognising whether another iteration is materially improving the work or whether the work has already achieved its purpose. Otherwise, the same constant availability that makes AI useful can quietly extend the work it was supposed to shorten.

The cybernetic teammate can keep the conversation going. Knowing when the work has reached its purpose still depends on human judgment.

Sources

Ayoubi, Charles. The Cybernetic Teammate: How AI Is Transforming Collaboration, Expertise, and the Future of Work. ESSEC Business School.

Ayoubi, Charles. “Scaling AI at Work Masterclass: How Leaders Can Turn Experimentation into Business Impact,.” ESSEC Executive Education.

Dell’Acqua, Fabrizio, Charles Ayoubi, Hila Lifshitz-Assaf, Raffaella Sadun, Ethan Mollick, Lilach Mollick et al. The Cybernetic Teammate: A Field Experiment on Generative AI Reshaping Teamwork and Expertise. NBER Working Paper No. 33641, 2025.

Mollick, Ethan, Co-Intelligence, Random House UK, 2024.