Across 40+ manufacturers, AI agents haven't cut a single headcount. What they did instead: let teams finally catch up on the work their ERP couldn't absorb.
What we've actually seen across 40+ manufacturers implementing AI
AI agents are not reducing headcount in supply chain.
Across 40+ manufacturers, the same people are still there. They're just finally catching up on work the ERP buried them under.
That's not the story most people expect, and it isn't the one on the slide that got the budget approved. But it's the pattern I keep running into, and I think it points at something more useful than the headcount question ever did.
Most of these operations were firefighting long before anyone mentioned AI.
The reason isn't mismanagement. It's that supply chains don't sit still. Tariffs move. Suppliers get swapped mid-contract. A part that was single-sourced last quarter is dual-sourced this one. Requirements shift faster than anyone can re-configure for.
The ERP, meanwhile, was set up for the business that existed on go-live day.
It's a snapshot. And the snapshot doesn't move when the business does.
That gap — between the business as configured and the business as it actually runs today — is where the firefighting lives. It doesn't announce itself. It just quietly becomes the job.
Here's what that looks like in practice.
Every gap the ERP can't keep up with becomes a spreadsheet. Every exception it can't represent becomes an email thread. Every rule that changed after go-live gets encoded in someone's head, or in a tab nobody else can read.
None of this is anyone's fault. It's what a competent team does when the system of record can't keep pace with the business: they build a faster one alongside it.
But it accumulates. And what accumulates is debt.
Work debt — the backlog of unstructured, un-logged work the ERP couldn't absorb, carried by people reconciling by hand.
Compliance debt — the quieter one. The audit trail for real decisions ends up in inboxes instead of in the system. The decision was made. It just wasn't made anywhere you can point to later.
Both of them are invisible on the balance sheet. Neither shows up in a dashboard, because the dashboard is reading the ERP — and the ERP is exactly the thing that fell behind.
This is where my expectation and the reality diverged.
The agents didn't replace the people carrying that load. They made those people more efficient at carrying it.
More work done in less time — that's the plain version. But the part that matters to anyone who's been burned by a clever workaround is what comes with it: the work becomes repeatable and compliant instead of one-off heroics.
A spreadsheet that only one planner can drive is fast, but it's fast once. It isn't repeatable, it isn't auditable, and it walks out the door when that person does.
That distinction is the whole thing. Efficiency that doesn't survive contact with an audit — or with the person who built it taking a new job — isn't efficiency. It's a loan.
For the first time, the teams could catch up. Not because there were fewer of them, and not because the process got magically simpler, but because the same people could get through more of it, in a form that held.
This is the part I'd push back on hardest when I hear it framed as a governance failure.
The spreadsheets and the email threads weren't people being sloppy. They were where the communication of the business had migrated — because that was the only place fast enough to keep up.
When your ERP can't represent this month's exception, the exception doesn't stop happening. It moves to email. The decision still gets made; it just gets made somewhere unstructured, uncontextualized, and unlogged.
Which is why "clean up the spreadsheets" initiatives fail. The spreadsheets aren't the disease.
The gap between a frozen system of record and a moving business — that's the disease. The spreadsheets are the immune response.
When an agent takes over that communication, something changes structurally: the work gets logged and structured instead of buried. Every exception becomes an event with context attached, and that context can be written back where it belongs.
That's why catching up actually holds, instead of the backlog quietly rebuilding six months later.
Before AI even comes up, these are the questions that tell you how much debt an operation is carrying:
None of these questions are about AI. They're about how far the business has drifted from the system that's supposed to describe it. The answers tend to be more uncomfortable — and more actionable — than any readiness assessment.
I understand why headcount became the metric. It's legible, it's on a spreadsheet a CFO already reads, and it makes the business case easy to write.
But across 40+ deployments, it isn't what changed. And I'd argue it was never the right thing to watch.
Whether the team can finally get ahead of the work — whether the business is running on a system of record that reflects reality, or on a shadow layer of spreadsheets and inboxes that happens to be faster — that's the one that moved.
That's also the one that compounds. A team that's caught up can respond to the next tariff, the next supplier failure, the next demand swing. A team that's behind is just choosing which fire to let burn.
Headcount was never the number to watch.
Working through the same thing in your operation? I'm curious how it's landed for you — when you brought AI in, did it change your headcount, or did it change how much your existing team could get through?
Nordoon co-founders
Subscribe to our newsletter, and we will send AI automation insights like this straight to your inbox on a regular basis.
