Agentic AI — S1 · 8/12: Frontline managers
Frontline managers are the forgotten variable in the AI equation — and yet they are the ones who will live it most intensely.
In regulatory affairs, they are not just people leaders: they are the experts who validate the documentation strategy, review and approve the deliverables.
Losers or winners? Both scenarios already exist.
The losing scenario is never announced. It settles in.
Agents multiply before the organisation is prepared. The manager is still expected to hold the same quality bar, on the same deadlines, on deliverables they must now approve without the means to check them — no time, no criteria, no traceability that would let them say why they signed. Nobody explained the AI strategy; they discover it through the tools landing in their scope. A scope that sometimes shrinks without explanation. And all the while, they struggle to reassure a team whose anxiety they share.
This is not a poorly supported manager. It is a control point drifting without a framework.
The winning scenario is built.
Before deployment, every workflow documents what the manager keeps, what they delegate to an agent, and what they answer for.
They are shown how these agents work, where they go wrong, and above all what they cannot do, so the manager judges instead of trusting. Leading a mixed team of humans and agents is treated as a distinct discipline worth investing in. And the clutter is cleared away — not to load them with more files, but to ring-fence what no agent will do in their place: train the juniors, settle the ambiguous cases, answer for the quality of what goes out, whatever produced it.
The manager remains a control point that holds.
No manager ends up on the losing side alone.
What tips it is a set of conditions: deep process knowledge, a seat where use cases are identified, being the reference point for transformation in their scope, and the time to exercise the critical eye that stops an error before it goes out.
And each of these conditions is decided.
What separates the two scenarios is knowing which conditions are missing — before scaling.