Meta Platforms (NASDAQ:META | META Price Prediction) employees are facing a double whammy. Not only is the company going through mass, successive layoffs, but the remaining employees are now supposed to forego their privacy… to train an even better replacement. Reuters caught wind of what has been happening inside Meta, and it doesn’t bode well if you work using a computer.
Meta has been installing tracking software on its U.S.-based employees, and the software records mouse movements and keystrokes. This means Meta is trying to build an AI agent using all of this data. In an internal memo, Meta said, “This is where all Meta employees can help our models get better simply by doing their daily work.” Or, in short: “Train your replacement.”
Meta wants to build a computer agent
When Claude’s computer use tool first came out, it was quickly adopted for menial tasks. The bot could go on webpages, research, fill out forms, and so on. Unfortunately, it’s extremely expensive and not reliable.
Meta wants in on this and likely wants to improve it to the point where it can start replacing large portions of its workforce.
Claude trained its model to work using screenshots, which are then analyzed by a model. The computer use tool is thus rather slow and struggles with a lot of the web.
Meta wants to go a step further and track tens of thousands of employees’ keyboards and periodic screen captures. By doing this, it might have the data it needs to build a model that can use a computer in real time while remaining cheap.
Zuckerberg will likely get what he wants
As you may have guessed, employees at Meta were not happy with this decision, and the company faced heavy pushback. Having software track all your mouse and keyboard inputs is arguably worse than being watched in this day and age. However, Meta is not giving up and has made “some” concessions.
Employees can now pause data collection for 30 minutes at a time to handle personal matters. Some can also request full exemptions, but this is tightly restricted to those handling sensitive material.
Regardless, the program is still mandatory for most of Meta’s staff. No law prevents the company from surveilling its employees, so Zuckerberg will have the millions of hours of data that a better AI model can train on.
Meta is pivoting from its advertising business
Meta Platforms has long been an advertising business. Its Family of Apps segment is its bread and butter that funds all its ambitions. Furthermore, Meta’s push into AI so far has been constrained to its own user base. The data centers it builds are for its own platforms and AI models, unlike most hyperscalers.
But this new directive to farm data off of its own employees hints that Meta could try to sell AI agents to others in the future. If Meta can prove it can build a product more superior than what companies like Claude have, it’s going to be very useful for companies that are employing thousands to fill out web forms and write emails.
Zuckerberg has been desperately trying to find “the next big thing” for Meta. The (so far) failed Metaverse investments didn’t play out well, but these AI investments have led to significantly more investor enthusiasm and are paying off in many ways.
What this means for META stock
I would not expect a sudden reversal in direction for META stock until it can build an AI agent that works. This is still an advertising business generating ~$50 billion of free cash flow per year, most of which is now going to AI.
In fact, Meta went from having ~$51 billion of net cash on its balance sheet in 2020 to having $2.3 billion in net debt by the end of 2025. I do not expect META stock to outrun its hyperscale peers in the coming years. And when it comes to AI, Meta hasn’t been able to deliver anything competitive. Its large language models are well behind state-of-the-art models by Google and Claude. It’d likely take years before Meta can start competing seriously, if ever.