Mark Zuckerberg wants AI to do your job.
Not someday. Not in the distant future. Right now. In 2026.
And to make that happen, he is doing something that has shocked the tech world this week — he is secretly recording every single thing his own employees do on their computers.
Every key they press. Every mouse click. Every time they open a dropdown menu. Even random screenshots of their screens — captured without warning.
And here is the most extraordinary part.
The workers whose every move is being recorded are the same workers Meta is planning to fire.
This is the story of how Meta went from giving workers special gloves to collect their hand movements — to now recording their keyboard and screen activity — all in the race to build AI that can replace human workers entirely.
Let Us Start From The Beginning — What Is Meta Actually Doing?
On April 21, 2026, Reuters reported something extraordinary.
Meta — the company behind Facebook, Instagram, and WhatsApp — is installing new tracking software on its US-based employees' computers.
The software is called the Model Capability Initiative — or MCI for short.
Here is what MCI does:
- Records every keystroke — every key you press on the keyboard
- Records every mouse click — every time you click on anything
- Records mouse movements — where your cursor moves on screen
- Takes random screenshots of your screen for context
- Runs silently in the background on a list of approved work apps and websites
The memo announcing this was sent to Meta staff through a channel belonging to Meta Superintelligence Labs — the same team building Muse Spark, Meta's new flagship AI model.
Workers were told something remarkable in that memo.
They were told that they can "help improve AI by just doing their daily work."
In other words — just show up and do your job. We will record everything you do. And use it to build AI that will eventually do your job instead of you.
Why Is Meta Doing This?
To understand why, you need to understand what kind of AI Meta is trying to build.
Meta is not just trying to build a chatbot that answers questions. They are trying to build AI agents — AI that can sit at a computer and do real work. By itself. Without a human.
Think about what that means.
Right now, if you want to do something on a computer — like fill in a spreadsheet, navigate through software menus, copy and paste between applications, or use keyboard shortcuts — a human has to do it.
AI models today are actually quite bad at this. They can write text. They can answer questions. But they struggle with the real, messy, practical work of actually using a computer the way a human does.
Why? Because nobody ever properly trained them on it.
That is what Meta is trying to fix.
By recording how real humans navigate dropdown menus, use keyboard shortcuts, switch between apps, and organise their work — Meta is creating a training dataset of real human computer behaviour.
Feed that data into an AI model and — in theory — the AI learns to use a computer the way a human does.
No more needing a human to do it.
This Is Not The First Time Meta Did This
Here is what most people do not know.
This keystroke tracking is actually the second chapter of Meta's data collection from workers.
The first chapter involved something far more physical — haptic gloves.
A few years ago, Meta's Reality Labs division was doing research into haptic gloves — special electronic gloves that can track every movement of your fingers, every bend of every joint, and every gesture your hands make.
The research goal was originally for virtual reality — so that your hands in VR would move exactly like your real hands.
But there was a hidden bonus.
Those gloves generated an enormous amount of data about how human hands move when doing tasks. Every finger movement. Every grip. Every gesture.
That data became incredibly valuable for training AI — specifically for training robots and AI models to understand how human hands work.
Meta was not alone in this. Companies like HaptX were already building haptic gloves specifically described as tools for generating training data for AI models. The haptic glove industry was quietly becoming a data collection industry.
Meta's research into haptic gloves gave them a blueprint — if you want to teach AI to do something humans do physically, you collect data from real humans doing it.
First it was hands and fingers with gloves.
Now it is keyboards and screens with software.
The method changed. The goal stayed exactly the same — collect human behaviour data to train AI to replace human behaviour.
The Timeline — How We Got Here
Here is the full story in simple steps:
2021 → Meta announces haptic glove research
Goal: teach AI how human hands move
Method: sensors tracking finger movements
2023 → Meta launches Llama — open source AI
Still behind OpenAI and Google in AI race
2024 → Meta pays $14 billion for 49% of Scale AI
Brings in Alexandr Wang to lead AI
Early 2026 → Meta launches Muse Spark
First serious rival to ChatGPT
April 2026 → Meta installs keystroke tracking on
employee computers
Goal: teach AI how humans use computers
Method: record every click, key, screenshot
Each step is the same story — Meta needs more data. And it keeps finding new ways to get that data from the humans around it.
What Exactly Gets Recorded?
According to the internal memo seen by Reuters, here is what the MCI software captures:
Keystrokes — Every single key you press. If you type a word, delete it, retype it — all of that is recorded. This teaches AI how humans actually type and correct themselves in real work situations.
Mouse clicks — Every time you click on anything. A button, a link, a menu item. This teaches AI how humans navigate software.
Mouse movements — Where your cursor goes on screen, how it moves. This teaches AI the natural flow of how humans look around a screen and decide what to click next.
Screenshots — Random captures of whatever is on the screen at a given moment. This gives context — the AI can see what the human was looking at when they made certain clicks or keystrokes.
The software runs on a specific list of work-related apps and websites — not everything on the computer. But that list includes the core tools Meta employees use daily.
The Part Nobody Is Talking About
Here is the detail that makes this story genuinely uncomfortable.
Meta is planning to lay off 10% of its workforce starting May 20, 2026. Reports suggest they may cut as much as 20% of staff total later this year.
So the timeline looks like this:
- Install software that records what employees do all day
- Use that data to train AI to do what employees do
- Fire the employees
Bloomberg put it bluntly in their headline: "Meta Is Making Workers Train Their AI Replacements."
The workers have not been given a choice to opt out. The memo informed them about the software — it did not ask for their permission.
Meta's position is that employees were told about the tracking. But being told about something and being able to say no to it are very different things.
Is This Legal?
This is a question a lot of people are asking.
In the United States — where this is currently being rolled out — employers generally have the legal right to monitor activity on company-owned computers. Employees have very limited privacy rights on work devices.
So technically, yes — Meta is likely within its legal rights to do this in the US.
But that does not mean it is not controversial.
Employee monitoring software has existed for years. But this is different in a key way — most monitoring software is used to check if workers are productive. This software is being used to extract the knowledge and behaviour of workers to train AI.
That is a different use entirely.
Europe has far stricter rules. Under GDPR — Europe's privacy law — collecting this kind of detailed behavioural data from employees would face serious legal challenges. This is why Meta is starting with US-based employees only.
What Does This Mean For The Future Of Work?
Let us be honest about what is happening here.
Companies like Meta are in a race to build AI that can do the work of white-collar knowledge workers. Not factory workers — office workers. People who sit at computers and do things like write documents, analyse data, navigate software, and manage information.
That is an enormous number of jobs.
The challenge has always been — how do you train AI to do those things? You need data. Lots of it.
The answer Meta has found is simple and somewhat shocking:
Use the workers themselves.
Pay them to come to work. Record everything they do. Build AI that replicates what they do. Then you no longer need them.
This is not a conspiracy theory. This is what the internal memo says. This is what Bloomberg, Reuters, and Fortune are all reporting this week.
The era of companies using human workers as AI training data has officially begun.
What Are Workers Saying?
Reactions inside Meta have reportedly been mixed.
Some employees understand the necessity — they work in AI and believe in the mission. They accept that their data will be used this way.
Others are deeply uncomfortable. Bloomberg reported that some employees feel they are being asked to train the robots that will eventually throw them "overboard."
The phrase "Metamates" — a term Zuckerberg introduced a few years ago to make employees feel like a team — has been brought up with significant irony in this context.
You are a Metamate right up until the moment the AI you helped train can do your job better than you.
The Simple Summary
Here is the whole story in plain words:
Meta wants to build AI that can use a computer the way a human does — clicking, typing, navigating software. That AI does not exist yet because no one has properly trained it.
To train it, Meta needed data. Real data of real humans doing real computer work.
So they installed software on employee computers that records everything — every keystroke, every click, every screenshot.
This follows earlier efforts using haptic gloves to record hand movements for robot and VR training data.
The result of all this data collection is AI agents — software that can eventually sit at a computer and do a full day of office work without a single human involved.
And the people whose daily work is being recorded to build this AI? Many of them will be laid off before the end of 2026.
Final Thought
There is something almost poetic — and deeply uncomfortable — about what is happening here.
For decades, workers were told that AI would replace repetitive, physical jobs first. Factory workers. Truck drivers. Manual labour.
Office workers — people with degrees, specialised skills, years of experience — were told they were safe. Their jobs were too complex. Too human.
Now the most valuable companies in the world are quietly recording every keyboard stroke and mouse click of their white-collar workers — to build AI that can do exactly what those workers do.
The data collection has always been the hardest part of building powerful AI.
Meta just found a remarkably efficient way to solve that problem.
The workers themselves.
Sources: Reuters, Bloomberg, Fortune — April 21–22, 2026
#meta ai#mark zuckerberg#employee tracking#keystroke logging#ai training data#model capability initiative#future of work
