Your boss isn't just checking if you're working anymore — they're recording how you work so an agent can do it instead.
The Summary
- Meta and other companies are now tracking employee actions on work devices not just to monitor productivity, but to generate training data for AI agents that will automate those same jobs.
- Workplace surveillance has evolved from measuring whether you're working to capturing how you work so machines can learn to replace you.
- Companies like AT&T and JPMorgan already use monitoring tools for presence tracking and productivity dashboards — the infrastructure for agent training is already installed.
The Signal
The quiet shift happening inside corporate surveillance infrastructure is this: every click, drag, and decision you make at work is becoming a training example. Companies investing in AI agents need high-quality behavioral data to model real work, and the cheapest source is the workforce they already employ.
Meta is reportedly tracking employee actions specifically to train AI agents. This isn't paranoia. It's optimization. The same monitoring tools that exploded during remote work transitions — justified as productivity safeguards — now serve a dual purpose. They watch you work, and they teach machines to work like you.
"This is the evolution of workplace surveillance — from measuring work to learning how to replace it."
The infrastructure was already there. AT&T tracks office presence. JPMorgan monitors software engineers through dashboards and AI usage metrics. According to a US Government Accountability Office report from September, employer surveillance grew because of remote work proliferation and the flood of available monitoring tools. What changed isn't the watching. It's what they're doing with what they see.
Agent training requires massive datasets of real human decision-making. How does a procurement specialist choose between vendors? What pattern does a customer service rep follow when de-escalating an upset client? How does a financial analyst prioritize which data points matter? You can't teach that by reading a manual. You teach it by recording thousands of real interactions, then feeding them into models.
Key dynamics at play:
- Surveillance tech marketed for accountability now doubles as data capture for automation
- Workers generating their own replacement training data as a condition of employment
- No new software purchase required — companies repurpose existing monitoring infrastructure
The incentive structure is brutal. A company that invests in monitoring tools gets immediate productivity oversight. Later, that same investment funds agent development. Every logged action becomes an asset. Dan Schawbel from Workplace Intelligence called it perfectly: the surveillance isn't just checking that you showed up. It's learning how to show up without you.
The Implication
If you're working at a company with device monitoring, productivity tracking, or required software on company hardware, assume your workflow is training data. The question isn't whether this is happening. It's how long until the agents trained on your work patterns are good enough to replace your role.
For workers: document the judgment calls, the context, the parts of your job that can't be captured in click patterns. Those are the elements agents will struggle with longest. For companies: this isn't just an HR policy question. It's a labor relations powder keg. People will tolerate being watched. They won't tolerate training their replacements under the guise of productivity monitoring.