AI agents are joining teams faster than teams are deciding what to do with them. The interesting question isn't what agents can do. It's who decides, who checks, and who is accountable when your newest colleague is software. Teams that answer this deliberately will fly. The rest will find out the hard way.
Your newest colleague never sleeps, never complains about the coffee, and will confidently invent a company policy that doesn't exist if you ask nicely.
Welcome to the team, agent.
Agents are no longer a lab curiosity. They triage tickets, draft analyses, reconcile invoices, write code and chase follow-ups. Some of them do it inside your teams already, often without anyone having formally decided they should.
“By the end of 2026, many organizations will have more AI agents than employees.”Amit Joshi, Professor of AI, Analytics, and Marketing Strategy, IMD — AI trends for 2026
Whether or not that happens in your organisation by December, the direction is clear. Which raises a question that is, at heart, about leadership and teamwork, not technology: how do you lead a team where some of the members are software?
The autonomy dial
The first decision is how much an agent is allowed to do on its own. Most teams never make this decision explicitly. They drift from "it suggests things" to "it sends things" without noticing the moment they crossed a line.
What changes for the people
Someone has to own each agent
Every agent needs a human who is accountable for what it does, the way a product owner is accountable for a product. Not the vendor, not IT, not "the team". When the agent gets something wrong, and it will, there must be a name, not a shrug.
The team coach's job gets more interesting
Agile and team coaches have spent years helping humans collaborate. Now they need to help humans collaborate with agents: which tasks to hand over, how to review agent output without rubber-stamping it, and how to stop the team from quietly deskilling itself. That's why we're investing in AI-enabled team coaches. It's the same craft, aimed at a new kind of team.
Middle managers don't disappear. Their job changes
“Middle management compression becomes real; expect a 10–20% reduction in roles.”Michael Wade, IMD — AI trends for 2026
That prediction will make some readers nervous, and it should make every leader think. When agents take over reporting, forecasting and follow-ups, what's left for managers is the part that was always the real job: judgement. Weighing things that can't be measured. Giving hard feedback well. Deciding when the algorithm is wrong.
IMD makes the same point from the other side: many managerial decisions are judgement problems, not optimisation problems. Design your systems so managers can still exercise that judgement, including overriding the agent, or you'll end up with people whose only job is clicking "approve".
Write it down: the team working agreement
Good teams already have working agreements. Teams with agents need a few extra lines. Here's a starting point you can steal:
Our working agreement with our AI agents
- Each agent has a named human owner: ______________.
- We've agreed what each agent may decide alone, and what needs a human yes.
- Anything touching customers, money or people's careers is reviewed by a human.
- We know how we'll notice when an agent is wrong, before our customers do.
- We keep doing some of the work ourselves, so we don't forget how.
- Anyone in the team can pause an agent, and nobody gets in trouble for doing so.
- We review this agreement every PI, quarter or sprint review, whichever comes first.
Signed: the humans. The agents were not consulted.
“If our agent made a serious mistake tomorrow, who would find out first, who would fix it, and who would explain it to the customer?” If the room goes quiet, you've found your next priority.
The bottom line
Agents will make good teams better and confused teams faster at being confused. The difference isn't the technology. It's whether someone sat down and decided how the team works now, who owns what, and where the dial is set.
That's not an IT task. It's leadership. And, conveniently, it's what we help teams do.
