There is a particular kind of dread in being asked to document exactly how you do your job, in detail, for a system you suspect is being built to do it without you. For thousands of Meta employees in 2026, that suspicion turned out to be correct.
The clip above captures something a lot of workers are feeling but few companies will say out loud. The specifics behind it are worth understanding, because they point to a shift in how some of the largest employers in the world are thinking about the people who work for them.
What actually happened at Meta
In April 2026, Reuters reported that Meta had begun installing tracking software on employees’ computers to record mouse movements, clicks, and keystrokes, and to periodically capture images of workers’ screens. The stated purpose was to train AI agents to perform specific work tasks, including the small, fiddly things humans do without thinking, like navigating drop-down menus. The internal program had a name: the Model Capability Initiative.
The point most workers drew from this was direct enough. You do not need a detailed recording of how a human performs a task unless you intend, at some stage, to have something other than that human perform it.
Then came the layoffs. On May 20, 2026, Meta began terminating roughly 8,000 employees, about 10 percent of its global workforce, starting with its Singapore hub and moving across the UK and US as each region’s morning arrived. What made this round different from previous Meta layoffs was the detail employees had pieced together in the preceding weeks: many had been required to generate the training data for the AI agents designed to do their jobs.
The financial backdrop sharpens the story. The cuts arrived in Meta’s most profitable quarter on record, with net income of $26.8 billion on $56.3 billion in revenue for the first three months of the year, revenue up 33 percent year over year. This was not a company cutting to survive. It was a company cutting because it had decided it could.
Why this is bigger than one company
It would be comforting to file this under “highly paid tech workers, not my problem.” That would be a mistake.
Silicon Valley has a long record of setting workplace norms that spread outward: the open-plan office, hustle culture, the rapid pivot to remote work and back. If Meta successfully normalizes treating its own workforce as involuntary training data, it establishes a template that any cost-conscious employer can copy. The mechanism is not exotic. Most modern office work already happens on monitored systems, which means most modern office work can, in principle, be recorded and modeled.
This is also not limited to one firm. Meta separately moved to have AI handle up to 90 percent of its internal privacy and societal risk assessments, work previously done by people. Companies including Klarna, Salesforce, and Duolingo have all experimented with replacing roles through automation. Amazon customer service workers reported similar fears as far back as 2024, describing a growing sense that they were training the software meant to replace them.
The pattern that connects these cases is the quiet part said out loud. The idea that workers train their own replacements when they use AI tools has circulated for years, usually as something employers would not admit to directly. Meta’s surveillance program dropped the pretense.
What workers can actually take from this
Anger is a reasonable response, but it is not a strategy. A few things follow from the Meta episode that are worth holding onto regardless of your industry.
First, assume your work is observable. If your job runs on company software, treat the steps you perform as potentially recordable. This is not a reason to work worse; it is a reason to be clear-eyed about where the value of your role actually sits.
Second, the tasks most exposed are the most routine and rule-based. The parts of a job that resist easy documentation, such as judgment in ambiguous situations, relationships built over time, and knowing which rule to break and when, are harder to capture in a training set. Moving toward that kind of work is a genuine, if imperfect, hedge.
Third, know the legal landscape where you are. Workplace surveillance and the use of employee data to train AI are regulated very differently across jurisdictions. The European Union’s rules on employee monitoring and automated decision-making are considerably stricter than those in the United States, and several countries are actively revisiting these protections. If your employer introduces something like the Model Capability Initiative, your rights depend heavily on where you work.
None of this makes the underlying shift less real. The honest summary is that a profitable company recorded how its employees worked, used that record to help build systems to do their jobs, and then let thousands of them go in its best quarter ever. Whether that becomes an isolated case or a standard playbook is, for now, an open question. The video at the top of this piece is one worker’s way of saying she already knows which way she expects it to go.
Related Stories
- Meta to track workers’ clicks and keystrokes to train AI
- AI likely to displace jobs in a way similar to the Industrial Revolution, Bank of England governor warns
- “Godfather of AI” Geoffrey Hinton warns of a new wave of AI job losses in 2026
- Amazon customer service workers fear AI will replace them
- Meta plans to turn employees’ clicks and keystrokes into AI training data