How to lead a team that works with AI agents: a 6-step guide for commercial managers

Before anything else, be clear about which parts of the work you're happy for an agent to draft and which parts need a person. In my experience the split falls quite naturally. Agents are good at gathering and first drafts, such as pulling together account research, summarising call notes or drafting a first version of an email or proposal. People need to own the judgement calls and the relationship, which means deciding which accounts matter, developing the thinking around what a customer really needs, and having the conversation.

Write it down, even as a simple list, as you would with your 'human' job descriptions, so the two line up alongside each other. When the split is left unspoken, some people hand over far too much and others won't let the tool near anything, and you end up with an accountability gap.

You'll end up with: a short list of tasks agents can draft, and clarity on which decisions remain with people.

This is the step most teams skip, and it's the one I'd put the most time into as part of the Embedded Adoption stage. If the standard for good only exists in your head, or in the heads of your two best people, everyone else is guessing. Pull together three or four real examples of strong work, perhaps a proposal that won or an outreach message that got a reply, and talk through what made them good. Then do the same with a couple that fell flat. More heads are always better than one here, and this needs to be done in a 'learning and building on what's good' mode, so it's psychologically safe for the team to work on it together. There are still blind spots when it comes to deciding what good looks like.

From that conversation you'll get a handful of plain-English tests the whole team can use. Does this reflect what we know about this specific customer? Is the thinking underneath it sharp, or does it only look polished on the surface? Could I explain why it's good? And could I train my own AI tool to achieve this standard?

You'll end up with: a shared set of examples, and three to five questions that define good for your team.

When you review AI-assisted work, the most useful thing to look at is what happened between the first draft and the finished piece. What did the output tell them, what did they add, and what did they change? That's where you can see whether someone is building on the tool or passing its work straight through. I wrote about this gap in more detail in why teams stop using the AI tools they rolled out.

I've found it helps to ask each person to talk you through one piece of work a week in this way. It takes ten minutes and tells you far more about their judgement than reviewing twenty emails for tone.

You'll end up with: a weekly ten-minute walkthrough with each person, focused on their judgement.

Decide where a person checks AI-assisted work before it reaches a customer, and who that person is. For a new starter it might be every proposal for the first month. For someone experienced it might be a spot check of a few pieces a week. What matters is that everyone knows where the check sits and what it's looking for, which is the standard you agreed in step 2.

Keep it light, because a review step that adds two days to every proposal will be worked around, regardless of how you position it.

You'll end up with: a clear, agreed point where work is checked, and by whom.

If agents are saving your team time, somebody needs to decide where that time goes. BCG's 2026 AI at Work research found that 66% of employees still get limited or no guidance on what to do with the time they save (BCG, 2026). Left alone, it tends to disappear into more activity, whether that's more emails, more meetings or more of the same.

Help them identify the work that needs a person: more time preparing for important conversations, more time with existing customers, and more time thinking about the accounts that matter most. Encourage consultation around what this could be, so that adopting tools becomes performance-enabling and an exciting opportunity for career growth rather than a threat.

You'll end up with: an agreed plan for where saved time goes, reviewed once a month.

If your 1-to-1s still run through activity numbers and pipeline, people will carry on working the way they always have. The questions you ask tell your team what you value, so try adding a couple that focus on judgement and learning.

You'll end up with: two or three new questions in every 1-to-1.