AI companies have been scouring the internet for any training data that they can land their hands on, and Meta is now looking to generate its own data in some creative ways.
According to a Reuters report, Meta is installing new tracking software on U.S.-based employees’ computers to capture mouse movements, clicks, and keystrokes — all of which will be used to train its artificial intelligence models. The tool, called Model Capability Initiative (MCI), will run across a list of work-related apps and websites and will also take periodic snapshots of employees’ screens for additional context.
The goal, per internal memos seen by Reuters, is to help Meta’s AI models get better at the kinds of computer interactions they still struggle to replicate — things like navigating dropdown menus, using keyboard shortcuts, and other nuanced human–computer interactions. “This is where all Meta employees can help our models get better simply by doing their daily work,” one memo read.
Part of a Broader Push
The MCI announcement follows a separate memo from Meta CTO Andrew Bosworth, who told employees that the company would be stepping up internal data collection as part of its AI for Work efforts — now rebranded as the Agent Transformation Accelerator (ATA). Bosworth’s vision is pointed: “The vision we are building towards is one where our agents primarily do the work and our role is to direct, review and help them improve.”
Meta spokesperson Andy Stone confirmed that MCI data would feed into those efforts, while stressing that the data would not be used for performance reviews. Safeguards are in place to protect “sensitive content,” he said, without elaborating on specifics.
Meta’s AI Ambitions — and Stumbles
The move is part of Meta’s aggressive push to become a serious player in AI — a race it has found itself scrambling to catch up in. The company’s Llama 4 model was widely considered a flop, and the fallout was significant: Meta’s entire GenAI organization was sidelined, hundreds of researchers were laid off, and the company undertook a sweeping overhaul of its AI division.
To reset its AI trajectory, Meta made a dramatic move — acquiring a 49% stake in Scale AI for $14 billion and bringing in Scale’s founder, 28-year-old Alexandr Wang, as its Chief AI Officer. Wang now heads Meta Superintelligence Labs (MSL) and has been aggressively poaching top researchers from rival labs like OpenAI, reportedly with joining bonuses in the tens of millions. Meta recently released its Meta Muse Spark model, which surprised most observers with its strong performance on benchmarks.
The company has also created a new Applied AI (AAI) engineering team focused on improving its coding AI capabilities, and has been pushing employees internally to use AI agents for everyday work tasks — even when it slows them down in the short term.
Workforce Overhaul Underway
The employee tracking initiative isn’t happening in a vacuum. Meta is planning to lay off 10% of its global workforce starting May 20, with the possibility of additional cuts later in the year. It has also been collapsing traditional job functions into a single new role called “AI builder.”
This pattern isn’t unique to Meta. Amazon has trimmed approximately 30,000 corporate employees, and fintech company Block cut nearly half its staff. Across Silicon Valley, the automation of white-collar work is increasingly shaping hiring and firing decisions.
The Data Problem
At the heart of all of this is a fundamental challenge: AI agents that can autonomously operate computers need vast amounts of real-world interaction data to train on. Synthetic data and web scraping can only go so far. By tapping its own employees’ daily workflows, Meta is essentially turning its workforce into a live data labeling operation — without anyone having to do anything extra.
It’s a clever solution, even if it raises questions about employee privacy and the direction in which AI is reshaping the employer-employee relationship. For now, Meta insists the data is strictly for model training. But the line between “training AI to replace tasks” and “training AI to replace people” is one that employees across the industry are watching very closely.