Table of contentsClick link to navigate to the desired location
This content has been automatically translated from Ukrainian.
For the last two years, we have constantly heard the same thesis: AI saves time.
Claude writes code, ChatGPT prepares documents, Copilot assists programmers, Gemini analyzes information. Practically every office worker is already using at least one AI tool.
And that's true.
But a strange situation has arisen.
People have started completing individual tasks faster, yet the overall workload has hardly decreased. The workday has not become shorter. The number of tasks has not become less. Many even feel that they are working more.
In 2026, a separate name finally emerged for this phenomenon - botsitting.
And this word explains very well why the "AI revolution" often does not deliver the productivity everyone expected.
What is botsitting?
If taken literally, it means "taking care of a bot".
But it is not about needing to "turn on" or "configure" AI.
Botsitting is all the invisible volume of work that a person does after or during the work of AI:
- explaining the context;
- rewriting prompts;
- checking responses;
- looking for errors;
- correcting hallucinations;
- restarting the generation;
- combining several responses;
- adapting the result to real needs.
In fact, AI only performs part of the work.
Everything else is taken on by a person.
And this part is almost never accounted for.
The productivity paradox
The Work AI Index 2026 study conducted by Glean among over 6000 workers in the USA, UK, and Australia showed a very interesting picture.
Almost 90% of workers are already using AI.
About three-quarters say they have become more productive.
On average, AI saves about 11 hours a week.
Sounds fantastic.
But there is a nuance.
Workers spend an average of 6.4 hours each week specifically on botsitting.
This means that more than half of the saved time immediately returns in the form of monitoring the AI itself.
As a result, organizations do not achieve the effect they were counting on.
People work faster.
The company - not necessarily.
This is what is called the AI productivity paradox.
Why does this happen?
The problem is not that AI is bad.
The problem is that most companies simply "attached" AI to old processes.
Imagine a new employee.
You wouldn't give them a task with the words:
"Well, you'll figure it out."
You would explain:
- what their responsibilities are;
- where to get information;
- who checks the results;
- when to consult a manager;
- what is considered completed work.
With AI, for some reason, it's done differently.
They simply give access to the model and tell employees:
"Now use it."
As a result, each employee starts to build this process independently.
This improvisation is what botsitting is.
The reverse centaur
The analogy with a "centaur" is very apt.
In chess, there has long been a concept of Centaur Chess - when a human and a computer work together, complementing each other.
This is exactly how AI was sold to us.
A person was supposed to become stronger.
But botsitting flips this model.
Instead of AI helping a person, the person starts to work as an assistant to AI.
They:
- explain the context;
- remind of obvious things;
- check every result;
- correct mistakes;
- restart the generation;
- monitor every step.
This means that AI is no longer working for you.
You are working for AI.
The next level - botshitting
There is another problem.
After several hours of botsitting, people get tired.
And then begins what researchers have called botshitting.
This is the moment when a person stops carefully checking the result.
They see a well-formatted text, a nice table, or code that "seems to look right," and simply send it along.
Not because they are lazy.
But because they have already spent too much time on monitoring.
The most dangerous part here is not even the mistakes.
The most dangerous part is AI's overconfidence.
The model almost always responds very convincingly.
And this confidence often replaces real fact-checking.
This is a problem of processes, not people
Many companies draw the wrong conclusion.
They see a low effect from AI and say:
"Employees are not using AI enough."
After that, they buy even more licenses.
Conduct even more training.
Require AI to be used even more frequently.
But the problem is not at all that.
It is not necessary to increase the use of AI.
It is necessary to change work processes.
For every scenario, it should be clear:
- what AI does;
- what a person does;
- what information AI receives;
- who checks the result;
- when human intervention is needed;
- who is responsible for the final result.
Until this is in place, botsitting will only increase.
How to reduce botsitting
It is impossible to completely eliminate it.
And it shouldn't be.
AI still requires human oversight.
But this oversight should be part of the process, not a personal initiative of each employee.
To achieve this, it is worth:
- clearly defining what decisions AI can make independently;
- providing the model with the necessary context before starting work;
- creating a mandatory stage for checking results;
- measuring not the number of requests to AI, but the actual reduction in time, number of errors, and rework.
Otherwise, the company gets beautiful statistics on AI usage but does not achieve better results.
Botsitting is not a temporary problem or "growing pains" of modern AI.
It is a natural consequence of starting to use new technology within old processes.
As long as people are forced to manually explain context, check every response, and correct model errors, a significant portion of the gains from AI will simply disappear.
Therefore, the future belongs not to those companies that use the most AI.
But to those that first learn to build processes where humans and AI work as one system, rather than as a manager and an intern who constantly needs supervision.
Only then will AI become a true tool for increasing productivity, rather than just another source of hidden work.
Like it?React
🧵
This post doesn't have any additions from the author yet.