Co-existence is a team sport.
Most advice about working with AI is written for one person at a keyboard. A company is not one person.
Written by NickStrategy & Adoption Lead, Zipline
Almost everything written about AI at work is addressed to one person at a keyboard. Write better prompts. Protect your career. Become ten times more productive. It is useful advice, and it describes a workplace that does not exist. In a real company nobody's work ends at their own desk. A quote depends on a price list someone else maintains. A reply to a customer depends on what the warehouse knows. Work is mostly handovers.
That is why I think "how do I work with AI?" is a question of the wrong size. What decides whether AI pays off in a company is how a team works with it. I call the answer co-existence, and I mean something specific by it: people and AI sharing the work in a way that leaves the team more capable than it was, and no less in charge.
An old idea with a new urgency
None of this started with chatbots. In 1960 the psychologist J.C.R. Licklider described what he called man-computer symbiosis: people set the goals and make the judgements, machines do the routine work that prepares the way. Two years later Douglas Engelbart set out a programme for augmenting human intellect, and gave us the mouse along the way.
The sharpest version came from Shoshana Zuboff in 1988. Studying paper mills and back offices as computers arrived, she saw that the same technology could be used in two ways. It could automate, taking the work out of human hands. Or it could informate, making the work visible so that the people doing it understood more than before. The technology did not choose between the two. Managers did.
In 1995 the cognitive scientist Edwin Hutchins showed where thinking in an organisation actually happens. He studied how the crew of a Navy ship fixes its position and found that the calculation lives in no single head. It is spread across charts, instruments and a team of people who each do one part and catch each other's mistakes.
What the evidence says now
The best recent test of this is a field experiment at Procter & Gamble, published in 2025 by Fabrizio Dell'Acqua and colleagues. They gave 776 professionals real product problems to solve, with and without AI, alone and in pairs. One person with AI did about as well as two people without it. That is the result that made the headlines.
Two other results matter more to me. People with AI stopped thinking only inside their own function: technical staff proposed more commercial ideas, and commercial staff more technical ones. And the most highly rated ideas came most often from teams using AI together.
Ethan Mollick, one of the authors, has since written the most readable account of the individual side of this in his book Co-Existence, including four advantages a person should cultivate. My argument starts where that list stops. In a company, personal advantages are only as strong as the organisation around them. An expert who is never asked to check the output is no safeguard. Good judgement without the authority to say no changes nothing.
Three tests of co-existence
When we design an AI coworker for a client, we hold it against three tests. I use the same three when someone asks me to judge a system they already have.
- Ownership. Did the person who does the work choose to hand this part over, and can they take it back? People know which of their tasks are dull, which are delicate, and which only look dull. A tool that arrives from above skips that knowledge.
- Judgement. Is there a named person who decides, with the expertise to tell good output from plausible output? And do they still do enough of the work themselves to keep that expertise? Matt Beane's research on skill shows how quickly it fades once juniors stop doing real work next to experts.
- Learning. Does the team understand its own work better because of the system? This is Zuboff's distinction turned into a design rule. A system that makes the work visible teaches. A black box that only emits answers makes the team dependent on it.
A system that passes all three is usually slower to design and faster to adopt. A system that fails one of them tends to end up in the drawer with the other pilots.
What this looks like on a Tuesday
Take our email coworker. It reads every incoming message and drafts the reply in the company's own tone. A person still decides what gets sent. That one design choice carries all three tests. The people who answer the email decide what it may draft. A named colleague owns every answer that leaves the building. And correcting drafts turns unwritten rules into written ones, so the team ends up knowing more about how it treats customers than it did before.
We also measure adoption next to hours saved. A tool that saves time on paper and that nobody opens has saved nothing.
What I am not sure about
I do not know where the line between human work and machine work will sit two years from now. It moves every few months, and tasks that clearly needed a person a year ago no longer do. Anyone who tells you where it will settle is guessing.
What I am sure about is the unit. Co-existence is not something one employee achieves with a chatbot. It is a working arrangement between a team and its tools. Like any working arrangement, somebody has to design it, and somebody has to keep it honest.
Sources and further reading.
The research referred to in this note, for anyone who wants to read it in full.
- J.C.R. Licklider, "Man-Computer Symbiosis", IRE Transactions on Human Factors in Electronics (1960).
- Douglas Engelbart, "Augmenting Human Intellect: A Conceptual Framework" (1962).
- Shoshana Zuboff, In the Age of the Smart Machine (1988). The distinction between automating and informating is hers.
- Edwin Hutchins, Cognition in the Wild (MIT Press, 1995).
- Fabrizio Dell'Acqua and co-authors, "The Cybernetic Teammate", NBER Working Paper 33641 (2025). Source
- Ethan Mollick, Co-Existence: The Next Phase of AI (2026). Source
- Matt Beane, The Skill Code (2024).
More field notes
AI changes fast. People don't.
Every company has an absorption rate: the speed at which its people can take in a new way of working. Trust sets it, not technology.
4 min readEveryone wants AI. Few ask what for.
AI is brilliant at some tasks and unreliable at others that look just as easy. A plan names the task.
4 min readWhen building is nearly free, the wrong thing gets expensive.
Cheap software sends three bills that never appear in the quote. Your team pays all of them.
4 min read