If you work at Meta's US offices, your computer just got a new passenger. The company has started rolling out an internal tool called Model Capability Initiative (MCI) that runs quietly in the background, recording how employees actually use their computers - mouse movements, clicks, keystrokes, and occasional screenshots included. The goal, as reported by Reuters and The Verge, is to feed all of that behavioral data into Meta's AI models.
Why Meta wants to watch you work
The reasoning is actually pretty straightforward, even if the execution feels a bit uncomfortable. Meta wants its AI agents to get better at navigating computers the way real people do - not just answering questions or generating text, but genuinely interacting with apps, completing tasks, and moving through software in a natural, human-like way. Watching actual humans do exactly that is, apparently, the most direct path there.
Think of it like teaching someone to drive by sitting in the back seat and taking notes. The AI isn't learning from instructions - it's learning from observation.
What this means for the bigger AI race
This kind of "computer use" capability is becoming one of the hotter frontiers in AI development. The idea of an AI agent that can genuinely operate your computer - handling repetitive tasks, navigating interfaces, filling out forms - is hugely appealing to businesses looking to automate work. Meta is clearly positioning itself to compete seriously in that space.
But training those models requires data that's grounded in real human behavior, not just synthetic examples. Using employees as a kind of living training dataset is an efficient solution, even if it raises obvious questions about privacy, consent, and the general feeling of being monitored at work.
The part worth sitting with
Workplace monitoring isn't new - plenty of companies track productivity in various ways. But there's something worth noting about the specific use here: this isn't about checking whether employees are slacking off. The data isn't being used to evaluate performance. It's being harvested to build commercial AI products. That's a meaningfully different dynamic, and one that employees and labor advocates are likely to have thoughts about as this becomes more widespread across the tech industry.
For now, the tool is limited to US-based Meta employees. But as AI agents become more capable and more central to how companies operate, it's a safe bet this kind of human-behavioral data collection will become a much bigger conversation - well beyond Meta's walls.
