Someone who implements AI for companies told me on the podcast that they'd always said this would be the year of the agent. Then it showed up commercially faster than they expected.
What that does to an enterprise is simple and uncomfortable. You're left with an employment model that looks like the top of an iceberg: a human who manages a certain number of AI-enabled and human helpers underneath them, depending on experience. That ratio might be one to three. It might be one to forty. Organizations are not prepared for that.
This is span of control, not software theater
Forget the tech demo for a minute. In a plant, a warehouse or a project office, the question is: how many automated workflows can one manager own without losing the plot?
Exception triage that used to need three coordinators. Bid packages that used to need a junior estimator for a full day. Claims packets that used to bounce between desks. Each one that gets automated still needs a named person who can explain what happened when something goes sideways.
I land here: count the workflows, not the seats. Then ask whether the manager on top can still see quality, safety and the customer.
The tension: scale vs. visibility
On the one hand, every vendor will sell you more automation under the same headcount. On the other, your board still wants to know who is accountable when an automated step misfires on a customer order or a jobsite change.
I kind of land in the middle. Let the ratio grow where the work is repetitive and measurable. Cap it where the risk is high and the feedback loop is thin. Don't pretend one supervisor can "just manage" forty automated paths with no redesign of how they spend their day.
What the iceberg looks like in practice
- Top of the iceberg. One ops lead, project manager or plant supervisor.
- Underneath. A mix of people doing judgment work and automated workflows doing the busywork.
- The ratio. Junior roles might supervise fewer automated paths. Experienced leads might own more — if the checks are real.
- The failure mode. Nobody knows who owns a workflow until a customer calls.
An example
Say a mid-size logistics firm automates overnight exception sorting, appointment reschedules and a first-pass claims draft. One ops manager now "owns" twelve automated paths plus a five-person desk. On paper, headcount looks efficient. In practice, that manager needs a daily review of the exceptions the tools couldn't close, a weekly quality sample and a named backup when they're out. Without that, the 1:12 ratio is a mirage.
Signs you're not ready for the ratio
- Managers can't list the automated workflows under them.
- No one can say what "good" looks like for each path.
- Escalations still land on whoever answers Slack first.
- You added tools and kept the same org chart, same meeting cadence, same review habits.
What to do this week
Pick one manager in ops, estimating or claims. Have them write down every automated or AI-assisted workflow they already own. Next to each, write who reviews output before it leaves the building, and what the hard stop is (safety, money, customer). If the list is longer than they can review in a day, the ratio is already ahead of the org.
Common mistakes
- Counting seats, not workflows. Headcount went down. Oversight load went up.
- One ratio for everyone. A new supervisor and a twenty-year lead should not own the same span.
- No redesign of the manager's day. More underneath means a different calendar on top.
- Treating prep as ownership. If nobody samples the work, you don't manage it — you hope.
Who should own this
Function leaders own the ratio in their area: how many automated paths per manager, and what gets reviewed. The CEO owns the company rule: we don't grow the iceberg under someone who can't see the base.
Count the workflows. Then manage to them.
Start by measuring how many automated paths each manager already owns, and whether they can still name the check on each one. That single inventory tells you whether the year of the agent arrived before your org did.