Who's Minding the Farm?
How companies are trading expertise for efficiency—and why that's a dangerous bet.
Who's Minding the Farm?
The AI efficiency race is creating a human capital risk almost nobody is governing
By Mary Hartwell
Hartwell & Co., AI & Data Governance Advisory
I've sat in enough of these meetings now that I can basically predict them before they start. Someone puts up a slide. Efficiency targets, automation opportunities, the usual. How fast can we adopt AI. Where can we cut. And then, always near the end, somebody asks it quieter than the rest, how many people do we still need.
I'm not going to pretend that's not a legitimate question. It is. AI is going to change how work gets done, and pretending otherwise is its own kind of denial. But there's a question I've started asking back, and I'll be honest, it usually kills the momentum in the room, which is exactly why I keep asking it.
Who's minding the farm.
Because right now, a lot of companies are doing two things at once that don't fit together, and almost nobody's stopped to notice. They're chasing AI efficiency as hard as they can, and at the same time they're clearing out the people who've been there the longest. And buried under both of those moves is an assumption nobody's said out loud, which is that somehow technology is going to absorb everything those people carried around in their heads. The judgment. History. The relationships nobody wrote down anywhere.
You're not laying off a job. You're laying off memory.
When somebody with twenty, thirty years walks out for the last time, sure, the org chart loses a box. That part's easy to see. What's harder to see, what doesn't show up until months later, sometimes years, is everything that box was quietly holding up.
That person knew why some process exists even though it looks completely dumb on paper. They knew which customer has a weird exception baked into their account and why nobody's touched it in six years. They knew the decision made back in whatever year that looks irrational now unless you were in the room for it. They knew which vendor to call first, and which one you never call unless you have no other option. They knew which dataset everybody quietly doesn't trust, even though there's no flag on it anywhere, no note, nothing official - just a feeling people have learned to trust. And they knew the workaround. The one that never made it into any document because, honestly, who had time to write it down.
We talk a lot in data circles about lineage - where did this come from, what happened to it, who's vouching for it. I think we need the same thing for what people know. Where did this understanding come from. Who's checking it. What happens the day the person holding it in their head retires, or gets laid off, or just leaves because they're tired.
Most of what a company knows was never in a system. It's in a person. Still is, in most places. That's the whole problem.
AI can process knowledge. It doesn't have wisdom. Those aren't the same thing and I'm tired of people acting like they are.
AI's genuinely good at some things. It'll read ten thousand pages before you've finished your coffee. It finds patterns. It gets a new hire up to speed faster than we ever could. That's real, I'm not knocking it.
But having access to information isn't the same as having expertise, and somewhere along the way a lot of smart people started treating those as interchangeable. And a fluent answer isn't the same as actual thought leadership, which is a phrase I hate, honestly, because it gets thrown around by people who've never had to be one.
Thought leadership isn't a clean answer. It's knowing the obvious answer is wrong before you can even say why. It's pushing back on the question itself instead of just answering it. It's realizing something that's technically correct is going to fall apart the second it touches a particular client, a particular culture. It's remembering something from ten years ago and knowing, somehow, that it matters right now. It's sitting in a room, looking at a beautiful analysis everyone else is nodding along to, and saying, I don't know how to prove this yet, but something here is wrong.
AI can help that person get there faster. It can poke holes, surface a connection somebody missed, save a week of digging. What it can't do is replace the person who's already lived through the version of this that went badly. And if your business sells expertise - that's not a small distinction. That's the whole business.
Nobody's asking the client what they think about any of this
Here's the part that bugs me most, honestly, and it's the part nobody brings up.
The client.
When somebody hires a firm, consulting, managed services, whatever, they're not just buying a deliverable. They're buying somebody who's seen this exact kind of trouble before and knows what to do about it. That's the actual product half the time. Not the deck. The person sitting across from the client building a relationship, getting to know the person they are trying to help.
So, picture that same firm quietly thinning out its senior bench. Nothing on paper changes. Same contract, same invoice, same SLA. But the people doing the actual work are different now. Younger. Faster on the tools. Learning on AI more than the people before them did, maybe leaning on it for things a senior person used to just... know.
And that raises a question a lot of firms would rather not answer: do the clients they are servicing know who's behind their account anymore? Because this stopped being a staffing decision a while ago. It's a trust question now. If someone paid for expertise and what's delivering it has changed underneath them, they deserve to know that. Full stop.
This is bigger than a model problem
This is where I break from the more technical crowd on AI governance. You can't hand this whole thing to IT, or Legal, or whoever owns risk, and call it handled. AI is changing how the business runs, not just what the systems do. So, the governance must cover the business. Not just the model.
- Who's doing this work now, and who used to.
- What did those people know that never made it into a document anywhere.
- Who's checking the AI when it's wrong?
- Clients know when AI's doing real work on their account. And when it screws up, and it will eventually screw up, who owns that?
- Who is accountable for those mistakes?
- What happens when the client gets a huge fine for a compliance issue?
Map what you'd lose before you touch the org chart
If somebody brought me in before a big AI-driven restructuring, I wouldn't start with which roles to cut. I'd start with what this place cannot afford to lose. Then I'd go find it.
Not job titles. Capabilities. Who knows the customers, not what's in the CRM, the real version. Who remembers why the process is built this way and not some other way. Who knows exactly where the data quietly falls apart, the stuff nobody's flagged. Who's the person everyone still calls when the manual runs out of answers.
Those people are infrastructure. Just because it's not a server rack doesn't make it less real.
Repurpose before you remove. It's not complicated, it's just inconvenient.
AI transformation doesn't have to mean layoffs, even though everyone's treating it that way right now. Some roles are genuinely going away, fine, that's not new, that's always happened. But there's something sitting right in front of most companies that they're just walking past: take your most experienced people and make them the ones who build whatever comes next, instead of the first ones out the door.
- Make them the ones checking the AI's work before it reaches a client.
- Let them train the people who'll eventually replace them, on an actual timeline, not a rushed one.
- Put them on the governance council.
- Have them break the model's answers against real situations they've lived through.
- Have them teach the place, and the AI , how the business really works underneath whatever the documentation says.
That's what change management is supposed to be. Not a memo. An actual handoff, from a person who knows something to a system that doesn't yet.
Cutting first and figuring it out later isn't a strategy
Understand what people are doing before you touch anything. Capture what they know. Get it into something that outlasts them. Then bring in the AI. Then redesign around whatever's working. Optimize the workforce last, not first.
Most companies run it backwards. Cut, deploy, patch the holes later. It looks great on a slide for a quarter or two. It gets expensive in ways that don't show up until it's already a problem.
Efficient isn't the same as capable, and I don't know why that needs saying but apparently it does
AI can make a company dramatically more efficient. That's not in question. But efficient and capable are not the same word, and cutting cost is not the same thing as creating value, no matter how good the first slide looks.
If you clear out your experienced people and expect the AI to just fill in behind them, that's not a transformation. That's efficiency built on nothing, and things built on nothing tend to stay upright for exactly as long as nobody leans on them.
The numbers look great at first. Headcount down, cost down, productivity up on the dashboard. Then the stuff that's harder to put a number on starts creeping in. Decisions get a little worse, a little at a time. Exceptions take longer because nobody remembers the last time this exact thing happened. New hires don't know what they don't know, and there's nobody left with the mileage to tell them. The AI gives an answer with total confidence and nobody in the room has the standing left to say no, that's wrong. Clients start feeling like an account number instead of a relationship. Nobody pushes back on the obvious idea because nobody remembers why the obvious idea didn't work the last three times.
None of that shows up in next quarter's report. It shows up eventually. It always does. It just takes long enough that whoever made the cut is usually gone by the time the bill comes due.
Your knowledge is part of your data strategy, whether you've written that down or not
Every experienced data person I know has spent years drilling one line into people: Garbage in, garbage out, and now we are saying AI is only as good as the data behind it. Same rule applies to what your people know. Feed a model incomplete documentation and half a history and it doesn't become an expert. It just gets very confident with whatever scraps you handed it.
So, stop asking is our data ready for AI, and start asking is our organization ready for AI, because that includes the data, sure, but it also includes the people, the process, who owns what, and whether anyone's managing this instead of just announcing it.
It was never humans or AI
I don't think the winners here are the companies that pick humans over AI. I don't think it's the ones that just swap humans out for AI either. It's whoever gets honest about which is which, what the humans should be doing, what the AI should be doing, and where putting them together gets you somewhere neither one gets to alone. That takes actual leadership. Not a slide that says "AI-first."
Nobody's saying saves every job. Businesses change, they always have, that's fine. What's worth saving is what the place knows, even while everything around it changes.
So, if you're cutting experienced people while ramping up AI at the same time, here's what I'd want you to sit with for a minute, past the productivity forecast. What are you losing that you haven't written down anywhere. Who's left to check the AI when it's wrong. Who still remembers why things are the way they are. And if you sell expertise for a living, does your client actually know who's behind the curtain these days, or do they just assume it's still the person they signed with.
AI can help run the farm. It can tell you everything the manual says about how the farm's supposed to work. It might even help you run it better than you ever have.
But before you send everybody home because technology looks like it's got this handled, you'd better make sure somebody wrote down what the people running that farm learned over thirty years of doing it. Otherwise, one day you're going to look up at a beautifully efficient operation and realize nobody left remembers why any of it works in the first place. And there won't be a single person standing there who can tell you when the machine's got it wrong.
Mary Hartwell
Hartwell & Co.
AI & Data Governance Advisory
Good questions. Stronger systems. Better outcomes.
