The Strongest Tree on the Street Until It Wasn’t
June 24, 2026
Top 10 Common Signs You Need a Digital Transformation Strategy
July 28, 2026
“ I need to shrink my team. How do we go from 100 employees to around 20?”
I heard similar comments in two different conversations last week. Both executives were asking the wrong question.
I explained to both leaders that this strategy is backward. Before you can know which roles to restructure or eliminate, you have to know which tasks AI can own. If you skip that step, you’re making a lot of assumptions.
That conversation is happening in boardrooms everywhere right now. Gartner surveyed 350 global executives in 2026 and found that among organizations deploying autonomous AI, workforce reductions are common. Those reductions, however, have not translated into a return on investment.
Cutting headcount isn’t a strategy. It’s cost reduction. You might be the CFO’s favorite person for suggesting it, but it will not get you far with other executives who are looking for more than just cost-cutting.
Why the Headcount Conversation Is Fatally Flawed
When executives hear that AI can automate data extraction, generate reports, reconcile datasets, and write code, they jump ahead to a seemingly obvious question:
‘How many people do I really need?‘
The question’s reasonable. The timing’s wrong.
The right sequence is to stand up the technology, map what it actually handles, find the gaps, and then make decisions. Headcount comes at the end of that process, not the beginning.
Here’s the problem with starting there: the people you’re cutting aren’t doing just one thing. They’re doing five things, and you only noticed the one AI can handle. The analyst running data extractions is also the person who flags when the numbers look off, knows why the Q3 anomaly happened, and maintains the relationship with the business unit that relies on accurate reporting.
Cut the analyst, and the extraction is automated, but everything else walks out the door with them. So, when the AI produces something inaccurate, and there’s nobody there to catch it, who does the business team call?
You’re going to need at least some of these people, but you don’t yet know how many or which ones. And you can’t rebuild institutional knowledge on demand. Once those people are gone, they’re gone.
You didn’t get leaner. You got fragile.
There’s another problem with making that decision before you’ve transitioned to AI: you don’t know who you’re cutting. Your team has already started organizing itself around AI, whether you know it or not. Some of them are already supporting what you’re trying to build toward. You just can’t see it yet.
The Two Camps on Every Team
Walk into your organization right now, and you’ll find two groups already in place.
The first are the Quiet Adopters. They’re coding with AI today. They’ve figured out how to generate SQL queries on demand, summarize complex outputs for business teams, and build and test data workflows in a fraction of the time.
They’re faster than they were six months ago. Some of them are using AI tools that the company hasn’t approved or doesn’t know about. That’s shadow AI. Others are using sanctioned tools and not saying how much faster they’ve gotten. Both stay quiet for the same reason: if leadership knows a task that used to take a day now takes an hour, the next question is obvious.
The second group is the Steady Hands. They haven’t changed their workflows. Not because they can’t, but because they don’t see a reason to yet. They’ve been doing their work well for years. The company depends on what they know, and nobody’s given them direction otherwise.
It doesn’t mean they can’t adapt. It means nobody’s shown them why they should.
The first group is embracing the fact that AI is changing what’s possible. The second group is holding on to experience and institutional knowledge that still matter. The headcount conversation doesn’t account for either.
Without a plan for either group, both become problems. The early adopters apply AI where it fits their workflow, not necessarily where it’s safest or most valuable. The experienced people either get cut before their knowledge is documented or dig in harder as the pressure to change increases.
A company with no strategy for either camp isn’t managing AI adoption.
The adoption is managing itself. And we’ve seen this before.
The COBOL Problem
Some of the highest-paid consultants right now write code in a language the industry declared dead thirty years ago. COBOL still runs a big portion of financial systems, insurance infrastructure, and government functions.
According to a Reuters report, 95% of ATM card swipes and 80% of in-person transactions are handled by Cobol. (Side Note: I can tell your age if you do know what Cobol is, or if you don’t know what an ATM is…)The companies that depend on COBOL can’t get off it. The people who know how to maintain it are becoming harder to find. As they retire, the knowledge retires with them.
This is what happens when a technology transition happens without a transition plan. The company ends up hostage to the past while trying to move toward the future, paying a premium to maintain COBOL legacy systems rather than plan their replacement.
AI transitions are setting up the same situation. Companies moving fast on headcount decisions without mapping which workflows are being transitioned, and which aren’t, will spend the next decade managing a similar mess.
An analyst builds a workflow around a custom prompt template. It becomes load-bearing; three other processes depend on it. It’s undocumented. The analyst leaves. And now nobody knows what the prompt does or why it works. The first time it breaks, the team discovers they’re as dependent on that prompt as any bank is on COBOL. Different decade, same debt.
The right questions aren’t being asked. Which parts of the transition require active management? Where is human oversight still necessary after AI is doing the work? What needs to be documented before the people who understand the work move on?
Some of those questions have the same answer.
What AI Can't Replace
Before you can know who stays, you have to figure out what can’t be replaced. That list is likely longer than many expect.
- Someone still has to tell the AI what to do.
That means translating business problems into specifications that the model can work from. Get the requirements wrong, and the AI executes the wrong task with high confidence and high efficiency. That’s not a productivity gain. - Someone still has to know when the AI is wrong.
In financial reporting, compliance documentation, or customer-facing decisions, “the model seemed confident” isn’t a sufficient standard. - Someone has to be responsible for what the AI does.
When it makes a bad call, a name goes on it. When the model drifts, someone has to catch it. When the regulation changes, someone has to update what the AI is working from.
The people who define requirements, validate outputs, and take accountability are where the majority of AI value lies.
Get the Sequence Right
The headcount conversation can’t come first. When you cut the people who tell the AI what to do, catch its mistakes, and answer for its decisions, you’ll not only hobble your program, but spend years rebuilding what those people knew.
The real challenge isn’t replacing people. It’s figuring out who’s ready, who can grow, and what roles remain irreplaceable.
The Quiet Adopters are already showing you the answer. A task that used to take a day now takes an hour. You don’t cut the person doing it. You give them five more to handle.
That’s how you deliver ROI: not through reduction, but through capacity.
This is the conversation I have with mid-market companies before they act, not after. If your company is having this conversation and this analysis hasn’t happened yet, reach out to talk through where to start.
Key Takeaways
- Gartner’s May 2026 research is direct: AI-driven workforce reductions don’t translate to ROI. Cutting headcount is cost reduction. The headcount conversation is happening before anyone has done the analysis that would make it a strategy.
- AI can’t tell itself what to do, catch its own mistakes, or answer for its decisions. Those functions have to be owned by people. Know who owns them before you decide who stays.
- Your Quiet Adopters are already showing you the path to ROI. A task that used to take a day now takes an hour. The opportunity is in giving them five more to handle. AI delivers returns by expanding what your team can do, not by shrinking it.
Frequently Asked Questions
Companies cut headcount before mapping which tasks AI can actually own. The people they're cutting aren't just doing the one thing AI is replacing — they're doing five things. When they're gone, so is the capacity to make AI work. Gartner's May 2026 research is direct: AI-driven workforce reductions don't translate to ROI.
Someone still has to tell the AI what to do. Someone still has to know when it's wrong. And someone has to be responsible when it causes a problem. Those functions can't be automated. They have to be owned by people, and you need to know who owns them before you decide who stays.
That pressure is real, but there's a faster path to margin improvement than cutting headcount. Your Quiet Adopters are already identifying process improvements. Celebrate them. Give them more to handle, and you've expanded capacity without adding headcount. That shows up on the P&L faster than a severance cycle does.
