Over the last couple of months, I’ve been experimenting with AI employees.
I wanted to understand what these tools can actually do before considering how I might implement them into client workflows, and how they compare with the growing number of AI agents entering the market.
A few weeks into one experiment, I received a notification:
My team had gone on vacation.
I had used all my credits for the month.
I chuckled at the idea of my AI employees collectively packing their bags and heading off on vacation. But running out of AI credits wasn’t particularly surprising.
Between ChatGPT, Claude, and the other AI tools I use regularly, I’ve become very familiar with some version of:
You’ve reached your limit.
And when that happens, the decision is remarkably simple.
Do you want to pay for more capacity?
Or do you want to stop?
There’s no negotiation.
I can’t explain that the project is really important. I can’t say we’re almost at the finish line. I can’t remind the AI that the work is for a good cause or appeal to its passion for the mission.
I’ve used the capacity I paid for.
If I want more, I have to make a choice.
And something interesting happens when I’m forced to make that choice.
Quite often, I wait.
Suddenly, the thing that felt incredibly urgent can sit for three days. Sometimes a week.
The deadline moves.
The project waits.
The world keeps turning.
I never thought much about this. It’s simply how AI works.
Until my AI employees went on vacation.
My first thought was:
This level of work-life balance sounds fun.
My second was a little more uncomfortable.
How did AI end up with clearer boundaries around capacity than many human employees do?
When did human capacity become negotiable?
Spend enough time in small businesses, nonprofits, social enterprises, and other mission-driven organizations and working beyond capacity can start to feel completely normal.
There’s an important deadline.
A new program is launching.
A proposal needs to go out.
A client needs something.
An event is coming up.
And the work matters.
So someone stays late.
Someone works Saturday.
Someone answers the email from the couch at 9 p.m.
Occasionally, that’s probably fine.
There are seasons when organizations need an extra push. There are deadlines that genuinely matter. Sometimes we choose to give a little more because we care deeply about what we’re building.
But temporary effort is supposed to be followed by recovery.
You work late one evening, so you start later the next morning. You put in extra hours before a launch, so the following week gets lighter.
Somewhere along the way, many workplaces seem to have lost the second half of that equation.
The extra effort stayed.
The recovery disappeared.
And slowly, exceptional effort became normal capacity.
AI makes capacity visible
This is what fascinates me about the comparison.
AI capacity is incredibly visible.
You have a certain number of credits, tokens, tasks, or hours.
You use them.
When they’re gone, they’re gone.
If you want more, there is a cost attached.
Human capacity rarely works that way.
An employee finishes everything assigned to them, so we give them something else.
They respond quickly, so people learn they’re always available.
They answer a message while they’re on vacation, so another one gets sent.
They consistently rescue projects at the last minute, so eventually rescuing projects becomes part of their unofficial job description.
Nothing flashes on the screen saying:
Capacity exhausted.
Instead, the system keeps running.
Until the person can’t.
And by then, we often call it burnout.
What if human capacity were treated as a real constraint?
Imagine if organizations responded to human capacity the way software responds to computational capacity.
Your team has 150 hours available this week.
You have 190 hours of work you want completed.
Now you have a decision to make.
What gets delayed?
What gets deprioritized?
What gets delegated?
What gets automated?
What additional capacity are you willing to pay for?
Those are leadership decisions.
Instead, organizations often quietly solve the equation by allowing those additional 40 hours to disappear into evenings, weekends, lunch breaks, cancelled vacations, and personal phones.
On paper, everything fits.
In reality, people absorb the difference.
There’s a reason the gap lands on people rather than somewhere else. When AI capacity runs out, you can buy more in about ten seconds. When human capacity runs out, you can’t. Hiring takes weeks. Training takes months. And the work was due Friday.
So the only lever available in the moment is the one you already have: the people in front of you. Which is why “we’ll figure it out” so often means “they’ll absorb it.”
And because human beings don’t come with a dashboard showing that we’ve reached 100% of our monthly credits, it’s remarkably easy to pretend there is still capacity available.
Maybe AI isn’t better at boundaries after all
Of course, AI isn’t actually setting a boundary.
Software doesn’t need a vacation. It isn’t protecting its mental health or trying to maintain a sustainable relationship with work.
Someone simply designed a limit into the system.
And I think that’s the more interesting lesson.
AI doesn’t have better boundaries than humans. Its boundaries are structural.
Humans, on the other hand, are often expected to create, communicate, and defend their own boundaries inside systems that continuously reward them for breaking those boundaries.
That’s much harder.
Especially when the work matters to you.
Especially when you’re part of a small team.
Especially when saying no means someone you care about has to pick up the work.
Especially when the organization’s mission is genuinely important.
Which makes me wonder whether we’ve been framing the problem incorrectly.
Maybe sustainable work shouldn’t depend entirely on employees becoming exceptionally good at saying no.
Maybe leaders need to become better at recognizing when capacity has been exhausted.
When capacity runs out, something has to change
My AI employees taught me something unexpectedly useful when they went on vacation.
When capacity runs out, something has to change.
You can add resources.
You can reduce the workload.
You can change the deadline.
You can change the scope.
Or you can decide that something simply isn’t important enough to do right now.
What you can’t do is pretend the constraint doesn’t exist.
AI systems make that impossible.
Human systems often don’t.
And perhaps that’s the boundary we actually need to borrow from AI:
When capacity is exhausted, stop treating additional capacity as free.
Because if the only way your organization can accomplish everything on its priority list is through people consistently giving more time than you’ve actually allocated for the work, you don’t have an output problem.
You have a design problem.
And maybe it’s time we let the humans go on vacation, too.
