Why companies burn out their best people and call it efficiency.
In 2026, we’ve watched AI reshape how companies think about work. We’ve also watched companies drastically cut their workforces, while others hire "top talent" only to discover that what looked like talent was sometimes just a great salesperson in an interview.
And I think both of those things point to the same problem.
Companies are very good at valuing what they buy and surprisingly bad at valuing what they already have.
Over the last couple of years, I’ve watched that play out in budgets, hiring decisions, software, systems, and eventually in the quiet exit of people nobody realized were holding the whole thing together.
The first time I saw it clearly, I was sitting in a series of interviews with consultants who were promising major growth in 90 days. Some were talking about doubling or even tripling revenue within the first few months. What caught my attention wasn’t really what they were promising.
Words are cheap, results aren’t.
What interested me was how they arrived at those conclusions. I don’t really approach problems the way a traditional marketer or manager might. I tend to approach them more like a developer or systems architect. I want to understand the system before I start changing it.
- What are the actual problems?
- What does the data tell us?
- What has already been tried?
- Why did it work or fail?
- How do the pieces connect?
And maybe most importantly, how does the person sitting across from me actually think? Are they looking at the whole board, or are they just moving pieces?
So I wasn’t particularly interested in hearing the plan. I wasn’t interested in the perfect solution they had somehow developed after looking at our website for five minutes or talking to us for an hour. I wanted to understand how they thought.
And that’s where things started getting interesting.
So I asked things like:
- What would you do differently based on what you’ve seen so far?
- How would you use our actual data?
- What would you change first?
- How would you measure whether it worked?
The answers were usually pretty generic: change the logo, post more on TikTok, copy competitors, spend more on marketing, hire another agency.
And when I started asking why, the answers got a lot less specific.
- What data supports that decision?
- How much would you spend?
- What are you expecting that spend to produce?
- How would you know if it worked?
Sometimes the answer was basically, "It’s in my head. You just need to trust me." These weren’t inexpensive recommendations either. Some of these people were asking for six figures.
At one point we hired someone based largely on a recommendation. Someone the ownership side had heard about and believed could help. About a week and a half later, he was gone. In that time, he had produced a six-page report explaining what we were doing wrong and why we needed a complete rebrand. New colors, new logo, new website, essentially everything.
He had also spent roughly $1,800 on products for "market research."
The report was AI generated. Not assisted, generated.
And the problem wasn’t that AI was involved.
The problem was that none of it came from our actual system.
There was no analysis of our sales. No look at our margins, customers, channels, market position, or previous experiments. No real attempt to understand what we had already tried or why we had made the decisions we had made.
It was a generic strategy applied to a specific company.
And that’s an important distinction. Because none of the recommendations were necessarily bad. Rebranding can work, SEO can work, more content can work, market research can work, outside expertise can absolutely work.
But a strategy can sound intelligent and still be completely wrong for the system it is being applied to.
I also don’t think the AI did anything wrong. It did exactly what it was asked to do. The person using it just didn’t give it anything meaningful to work with, and then sold the output back as expertise. That’s one of the things I think AI is exposing in 2026.
We are getting very good at producing answers. We’re not necessarily getting better at asking the right questions. Gallup’s 2026 report looked at US workers whose companies have already implemented AI. 65% say it has had a positive impact on their own productivity. Only 12% strongly agree it has transformed how work gets done in their organization.
What the company thought it was buying
And that brings me to the bigger problem.
There was already someone inside the company who had years of context. They understood the data, the customers, and the market. They knew which vendors would actually answer the phone, which processes had exceptions that were never documented, and what had already been tried and why it didn’t work. They had accumulated all of the little pieces of information that don’t show up in a job description but can completely change the outcome of a decision.
That person was expected to explain the systems, transfer the knowledge, answer the questions, and sometimes continue doing the actual work while someone else came in to "manage" it. Often for several times the compensation.
Nobody has a line item for this
And this is where I think companies have a measurement problem.
Companies can put a price on almost everything they own. Equipment depreciates, inventory has carrying costs, IP gets valued, and goodwill gets a number when a company is acquired. But there is no line item for accumulated capability. There is no place in the accounting system that says:
- This person knows why the numbers look the way they do.
- This person knows which customers are actually valuable.
- This person knows which processes are broken.
- This person knows what we’ve already tried.
- This person knows why we stopped doing it.
- This person has spent years building relationships that would take years to replace.
What does show up?
Payroll.
So the person holding all of that context looks like a cost, while the consultant arrives with a proposal, an invoice, and a deliverable that looks like an investment. I don’t think that’s usually malicious. I think it’s measurement. And what we measure tends to become what we value.
AI didn’t create this problem, it sped it up
AI didn’t create this problem, it just sped it up.
AI makes output cheap. Reports are cheap, code is cheaper, marketing content is cheap, analysis is cheaper, ideas are cheap.
But judgment is still expensive, context is expensive, experience is expensive, knowing what not to do is expensive.
And AI doesn’t automatically have any of that. It can write the email, but it doesn’t necessarily know which customer you should send it to. It can analyze the data, but it doesn’t necessarily know why the data looks the way it does. It can build the system, but it doesn’t automatically understand the history of the business, the relationships involved, or what happens if that system is wrong.
And it doesn’t inherit the trust someone spent years building.
That’s why I don’t think AI makes experienced people less valuable. In a lot of cases, it does the opposite. If you already understand the system, AI gives you leverage. You can take the knowledge you already have and move faster, build systems that used to require entire teams, analyze more information, test more ideas, enter markets faster, and turn something that used to take three months into something you can do in three days.
Capability is not capacity
And this is where another problem starts.
The company sees the increased capability and thinks: "If they can do more, give them more." So they do, then more, then more. And eventually that person becomes the backbone of the company without anyone really stopping to ask what that is costing them.
Capability is not capacity.
AI can multiply what someone can accomplish. It cannot multiply their hours. And when the only response to increased capability is increased workload, eventually that person reaches a limit.
That’s burnout. Not a lack of commitment, not a bad attitude, not suddenly becoming less loyal.
A system that has been drawing down an asset without accounting for the cost. Gallup’s 2026 report puts manager engagement at 22%, down from 31% in 2022. Managers also report more stress, more anger, more sadness, and more loneliness than the people they lead. The people carrying the most are the ones coming apart first.
And the warning signs usually show up before the person leaves. They stop volunteering for things outside their scope, stop working the extra hours, start documenting everything, stop trying to fix every problem, and become less emotionally invested. And leadership sometimes looks at that and thinks the person is checking out.
Maybe.
Or maybe they already did the math.
The question executives should be asking
If your most capable person walked out tomorrow, could you actually replace what they bring to the company? Not their title, not their job description. Everything.
The relationships, the knowledge, the systems, the customers, the markets, the judgment, the history, and the mistakes they’ve already made so someone else doesn’t have to make them again.
And if you could replace all of that, what would it actually cost?
Because the replacement cost of a great employee isn’t their salary. It’s the recruiting, the hiring, the training, the ramp time, the mistakes, the delayed revenue, the lost relationships, the rebuilding, and the months or years it takes for someone new to understand the company at the same level.
Sometimes you can’t even hire someone with the same capabilities for what you were paying the person who left. And that’s the average cost of replacing a role. It doesn’t describe replacing the person who was quietly holding four of them together.
And this is where I think the loyalty conversation gets interesting. I don’t necessarily think people are becoming less loyal. I think they’re becoming more aware. They understand what they’re capable of.
They understand what AI allows them to build. They understand what their experience is worth. And they have more options.
What the company should build instead
So if you have someone in your company who consistently solves problems other people can’t solve, opens markets, creates revenue, builds systems, and keeps the operation moving, don’t just keep giving them more because they’ve proven they can handle it. Build around them. Give them authority, resources, people, and room to grow. And most importantly, don’t confuse their ability to carry the weight with an obligation to carry it forever.
You don’t own someone’s loyalty because you gave them an opportunity. You earn it by continuing to give them reasons to stay. Because if one person has become essential to your company, that’s not just a people problem. It’s a systems problem.
And if the solution is to keep that person running at 100% forever, then you haven’t built a system.
You’ve built a dependency.
That person can burn incredibly bright, light up the entire room, and become one of the most valuable people in the company. But they’re still a candlestick. They still have a limit.
And once they reach it, the fire doesn’t ask permission to go out.
If you have that person right now, build around them while they’re still burning.
Frequently asked questions
What is key person dependency?
Key person dependency is when a company’s operations rely on one individual’s accumulated knowledge, relationships and judgment to the point that the work does not continue properly without them. It is usually discussed as a people problem. It behaves more like an accounting problem, because that capability sits on no balance sheet, so nothing in the company’s reporting shows it building up or being drawn down.
Why don’t companies notice key person dependency until it is too late?
Because a company can see what it buys far more easily than what it already has. A consultant’s value arrives as a contract, an invoice, a proposal and a deliverable, so it reads as an investment. The person who spent years learning the business, the customers and the market has none of that paperwork, so the only number attached to them is payroll, and payroll reads as a cost.
Does AI reduce key person dependency?
Usually it increases it. AI multiplies what a capable person can produce, so the common response is to hand that person more work. Capability is not capacity. AI can multiply output, it cannot multiply hours, and it does not inherit the context or the trust that made the output correct in the first place.
What are the warning signs that a key employee is about to leave?
The signals show up long before the resignation does. They stop volunteering for things outside their scope, stop working the extra hours, start documenting everything, stop trying to fix every problem, and become less emotionally invested. Leadership often reads that as someone checking out. It is more often someone who has already done the math.
How do you fix key person dependency?
Treat it as measurement rather than sentiment. Write down what isn’t written down, and treat an undocumented process like any other single point of failure. Price the transition before you need it. Then give the person authority, resources, people and room to grow, so that capability becomes organizational capacity instead of a dependency.
