Everyone is asking how to keep humans in the loop of AI. The better question is how to bring an agent into the loop of human collaboration.
Every team runs a loop. We talk, we agree on something, someone says “I’ll take that”, time passes, and we meet again to find out what happened. Conversations turn into commitments, commitments into work, work into the next conversation.
It is a simple loop, and it leaks everywhere. A promise made in a call is never written down. A decision taken in one chat is unknown in another. A task sits untouched for two weeks because nobody noticed, and the person who owns it has quietly forgotten. Asana’s Anatomy of Work research, a survey of more than 10,000 knowledge workers, puts the share of working time spent on “work about work” (chasing status, searching for information, switching between tools) at around 60 percent, and 88 percent say projects have fallen behind or through the cracks. That is a vendor’s survey, but anyone who has worked in a team will recognise the shape of it.
None of this is anyone’s fault. Humans are not good at this kind of bookkeeping, and it does not scale. Every new person adds more conversations to follow, more handovers, more places for a commitment to disappear.
Turning the loop around
An AI agent, in the sense I mean here, is not a chatbot you open when you have a question. It is a persistent member of the team: it is present in the team’s conversations, it remembers what was said, and it can act on the team’s tools within limits the team sets.
Most of the public conversation about agents runs the other way. “Human in the loop” asks where a person must approve what the machine does. That is a real question, and an increasingly contested one: Amazon’s Eric Brandwine has said that a person approving agent actions over and over will “do a good job… then an okay job… and pretty quickly a poor job.” But it is a question about control. It puts a person inside the machine’s loop.
I am interested in the opposite. How do we put an agent inside our loop, the loop of human collaboration, so that it carries the part people do badly, and leaves people free for the part only they can do? (The phrase “agent in the loop” has been used before, in enterprise and research settings; I am using it here for this specific idea.)
What an agent in the loop does
An agent in the loop is always there, and always supporting. In practice that means:
- It keeps track of commitments. When someone says “I’ll send that by Friday”, it is written down, with the person’s name, and it is still there on Friday.
- It keeps track of conversations. The decision taken in one thread is available in the next, so nobody has to reconstruct it from memory.
- It follows up. Gently, once, to the person who owns the thing, not to the whole team.
- It notices what has stalled. A task that has not moved in two weeks is something the agent sees. Whether it still matters is for its owner to decide.
- It takes on the organisational overhead. The status updates, the reminders, the “where did we land on this?” that nobody signed up for and everybody pays for.
What it does not do is decide. People keep the judgment, the relationships and the direction. The agent keeps the loop closed.
Seeing the loop
There is a second benefit, and it may turn out to be the larger one. Because the agent sits inside the flow of work, it gives the team something it has never really had: a picture of how its own collaboration actually works.
Where do things get blocked? Which kinds of commitments slip, and at which handover? Which tasks sit untouched, and for how long? When the team changes how it works (a new meeting rhythm, a new way of splitting work), did it actually help?
Today most teams answer these questions with impressions. An agent in the loop makes it possible to answer them with evidence: good telemetry on collaboration itself, so the team can see where it is stuck and improve how it works together, instead of only working harder.
Visibility, not surveillance
That last point needs a boundary, because the same data could become something else. Telemetry on collaboration has to belong to the team, and it has to be about the work, not a scorecard of individuals. The person who owns a stalled task is the one who decides what it means. An agent that reports on people to someone above them is not in the team’s loop at all; it is in someone else’s.
Built this way, an agent in the loop is not a monitor. It is the colleague who remembers, follows up, and helps the team see itself clearly.
The future of collaboration
The first wave of digital tools digitised our paperwork and left the work of coordinating unchanged. An agent in the loop goes after that work directly: the commitments, the follow-ups, the memory, the seeing. Teams that have one will spend less time on work about work and more time on the work itself, and they will be able to look at their own loop and make it better.
That is the future of collaboration I am interested in: not humans kept in the loop of machines, but agents brought into the loop of people.
← All posts