Meta is moving ahead with a new internal AI plan that uses data from employee work activity, including mouse clicks, keyboard use, and screen snapshots, to train its AI models. According to Reuters, the company has started installing tracking software on U.S.-based employees’ computers through a tool called Model Capability Initiative (MCI). The goal is to help Meta build AI agents that can better copy how people use computers in real work tasks.
What Meta Is Doing
The latest report says Meta’s software will collect work activity from selected apps and websites, then use that material as training data for AI systems. Reuters reported that the tool may also take occasional screen snapshots so the models can learn from real computer behavior, not just from text or stored records. The company says this is meant to improve areas where AI still struggles, such as using dropdown menus, clicking buttons, and handling keyboard shortcuts.
In simple terms, Meta wants its AI to learn from how employees actually work on computers. Instead of only teaching AI with written instructions or public data, Meta is using real workplace actions as examples. That is why the company sees this as a way to make its future AI agents more useful for everyday office tasks. This approach is part of a broader push inside Meta to build AI tools that can do more work with less human help.
Why Meta Wants This Data
Meta’s main reason is clear: AI agents still have trouble doing basic computer work in a natural way. Reuters reported that the company wants better training examples for tasks like opening menus, choosing options, and using shortcuts the way a real person would. Meta spokesperson Andy Stone said the company needs “real examples” of how people use computers, because that helps AI learn more practical behavior.
This matters because many AI tools are good at language, but weaker at computer actions that seem simple to people. A model may write a reply very well, yet still struggle to move through a work app step by step. Meta appears to believe that if it trains on actual employee behavior, its AI will become better at handling office work on its own. That is an important shift, because it shows AI training is moving beyond public text and into real workplace behavior.
How the System Works
According to Reuters, the tracking tool is called MCI, and it runs on work-related apps and websites. The tool can capture mouse movements, clicks, and keystrokes, and it may also take screenshots from time to time. Meta told staff that the data will be used to improve model training, especially for tasks where AI still behaves less like a human user and more like a basic script.
The company also said there are safeguards in place for sensitive content, and that the collected data will not be used for performance reviews or any other purpose besides model training. TechCrunch, citing a Meta spokesperson, reported the same point: the company says the data is being collected only to train its models and is not meant for any other use. Even so, the move has drawn immediate attention because the data comes directly from employees’ everyday work habits.
Part of a Bigger AI Restructure
This is not a small one-off experiment. Reuters said the new data collection is part of a wider internal shift at Meta, where the company is reorganizing around AI and changing how work gets done. Meta CTO Andrew Bosworth told employees that the company is building toward a future where AI agents do most of the work, while people mainly guide, review, and improve them. The company has also renamed one of its internal efforts as the Agent Transformation Accelerator (ATA).
Reuters also reported that Meta is pushing staff to use AI agents in coding and other tasks, even if that slows work in the short term. At the same time, the company has been reducing old job distinctions and leaning more toward a general “AI builder” identity for some roles. Reuters further reported that Meta is planning to lay off about 10% of its workforce globally starting on May 20, with more cuts possible later. Taken together, these steps show that Meta is not only training AI, but also reshaping its workforce around it.
Why Employees Are Worried
The biggest concern is privacy. Even if the data is collected on work devices, many employees may still feel uncomfortable knowing that clicks, keystrokes, and screenshots can be logged and studied. Reuters quoted legal experts who said that U.S. rules on workplace surveillance are fairly weak, while European law may be much stricter. That difference matters because it shows that the same monitoring idea can face very different legal and ethical reactions depending on where workers are located.
There is also a trust issue. Workers may wonder whether “training only” stays truly separate from other uses later on. Meta says the data will not be used for reviews or other purposes, but people inside the company may still worry about how much is being recorded and who can see it. That concern is not just about one company. It reflects a wider debate in tech: how far a company should go when collecting human behavior data to make AI more powerful.
What This Says About the AI Race
Meta’s move shows how competitive the AI race has become. Companies are now looking for better training data wherever they can find it, because high-quality data can make a model smarter and more useful. TechCrunch said the story highlights how companies are searching for new sources of training data, and Reuters showed that Meta sees real employee behavior as a way to improve AI on tasks where models still fall short.
This can be read as a sign that AI is moving from a text-first stage into a work-first stage. The next wave of AI tools may not just answer questions or write drafts. They may try to act inside software, follow office workflows, and complete tasks with less human help. If that happens, companies will need examples of real computer use, which explains why Meta is now so focused on employee activity data. That is an inference from the Reuters reporting, but it fits the direction Meta is describing publicly to staff.
How This Could Affect the Future of Work
If Meta succeeds, this kind of training could help create stronger AI assistants for office tasks. That could mean faster workflows, fewer repeated manual steps, and better automation in areas like scheduling, data entry, support, and coding. Reuters reported that Meta wants agents to eventually do most of the work while people direct and review them. That suggests the company sees human workers more as supervisors of AI than as people doing every step by hand.
But the same trend could also make workers uneasy across the tech industry. If companies start recording more of what employees do on their computers, the line between useful training and workplace surveillance could become much thinner. That is why this story is important beyond Meta alone. It is not only about one company collecting mouse movements and keystrokes. It is about how the next generation of AI may be built from the everyday digital habits of real workers.
The Bottom Line
Meta’s latest plan is simple to explain but big in impact: use staff clicks, keystrokes, and screen activity to train AI agents that can work more like humans on a computer. Reuters reported that the tool is already being rolled out on U.S. employee devices, while Meta says the data is limited to model training and is protected by safeguards. The company’s wider AI strategy, including the Agent Transformation Accelerator and its push toward more AI-driven work, shows that this is part of a much larger change inside Meta.
For readers, the key point is this: Meta is trying to teach AI with real workplace behavior, not just public data. That may help AI agents become much smarter at office tasks, but it also raises serious questions about privacy, trust, and the future of work. As of April 21, 2026, this is one of the clearest signs yet that the AI race is now moving directly into the workplace itself.
