Meta’s Embrace of A.I. Is Making Its Employees Miserable

There are two ways to bring AI into your work life. In the first version, an AI agent runs quietly in the background — handling the tasks you hate, flagging what needs your attention, and staying out of your way. In the second version, you spend your day proving to your employer that you’re using AI correctly, hitting token quotas, and wondering whether the software on your laptop is watching you right now.

Meta’s 78,000 employees are living version two. And the details are instructive for anyone building or choosing a personal AI setup — because the lesson isn’t really about Meta. It’s about who controls the agent.

If you’re thinking through what agentic AI looks like in practice — for your work, your business, your life — Meta just ran a very expensive real-world experiment. Here’s what happened, and what it means.

What Meta Actually Did to Its Employees

Earlier this year, Meta began installing software on US employee laptops that tracks keystrokes, mouse movements, clicks, and on-screen activity. The stated goal, according to Business Insider: train Meta’s AI models on how humans actually complete everyday computer tasks. When an engineering manager asked how to opt out, Meta CTO Andrew Bosworth replied directly — there is no opt-out on company laptops.

At the same time, the company started factoring AI tool usage into employee performance reviews. It built internal dashboards that track how many AI tokens individual employees consume per day. It ran mandatory ‘AI Transformation Weeks’ to push adoption across every division. And on April 17, 2026, Meta announced plans to cut roughly 10% of its workforce — around 8,000 people — with the first wave scheduled for May 20, 2026.

Workers on the anonymous workplace forum Blind described the culture as ‘dead,’ ‘depressing,’ ‘joyless,’ and ‘in survival mode,’ according to reporting by Metaintro. Several employees noted that managers had stopped pretending the AI rollout was optional. Many felt reduced to what one internal report described as ‘AI supervisors’ — watching systems do work that used to require their judgment, while wondering whether they were training their own replacements.

The Recursive Loop Nobody Expected

Here’s the part the headlines mostly missed.

Meta’s dashboards track AI token consumption as a productivity metric. The more you use AI tools, the better your numbers look. So some employees — engineers, naturally — responded by building AI agents to manage their other AI agents. Automated systems consuming tokens on their behalf, running in the background, gaming the leaderboard. Digital Trends reported that the whole thing started to resemble a feedback loop eating itself.

This is darkly funny. It’s also the most important signal in the entire story.

Those engineers didn’t build recursive agent loops because they were lazy. They built them because the metric was wrong. When you optimize for AI consumption instead of actual outcomes, you get consumption — not outcomes. The workers who gamed the system accidentally demonstrated the core principle of agentic AI done right: agents should produce results for humans, not generate activity to satisfy a scorecard.

Even the brightest light can’t fix a workplace that’s lost its way.

Why Meta’s AI Misery Is a Warning for Personal AI Users

We’re genuinely skeptical of the ‘enterprise AI success story’ narrative right now, and Meta is a big reason why. The company’s approach — mandatory adoption, no opt-out, surveillance baked into the hardware, layoffs running parallel to the rollout — isn’t a cautionary tale about AI. It’s a cautionary tale about power. AI tools deployed with those constraints don’t make anyone more capable. They make everyone more anxious.

The contrast with personal AI agent deployments is sharp. When you control your own AI agent — choosing what it accesses, what it does, what data it touches — the dynamic inverts entirely. The agent works for you. Nobody is watching your token count. Nobody is using your behavioral data to train a competitor’s model. The surveillance flows in the direction you set.

Meta’s story also raises a question worth sitting with: if you’re adopting AI tools pushed by an employer or platform, what data are you generating, and where does it go? Meta’s employees spent weeks being told to embrace AI, integrate with AI, and learn from AI — and are now watching their computer behavior harvested for training data. The AI agent platform you choose, and who controls it, is not a minor technical detail.

There’s also a ‘deskilling’ dimension worth taking seriously. Meta’s own internal reporting describes employees frustrated that AI systems assume responsibilities that once required their specialized judgment — including code review systems that flag issues inconsistently and require human intervention to correct. The concern isn’t robots replacing humans wholesale. It’s humans becoming worse at their craft because AI handles the practice reps. Personal AI setups that keep humans genuinely in the loop — deciding, judging, applying expertise — avoid this trap. Setups designed to maximize AI output metrics don’t.

What to Do With This Information

• Audit what your AI tools collect. Before adopting any AI platform — especially employer-mandated ones — find out what behavioral data it captures. Keystrokes, screen content, and usage patterns are now fair game in enterprise deployments. For personal AI setups, choose platforms where you can see exactly what data is stored and how it’s used.
• Ignore token metrics. If you’re evaluating your own AI agent usage, measure outcomes: tasks completed, time saved, decisions improved. Token consumption is the Meta trap — it measures activity, not results. Your AI agent’s job is to make your life better, not to generate impressive usage statistics.
• Watch the layoff-plus-mandate pattern. Meta’s combination of mandatory AI adoption and simultaneous workforce reduction is deliberate. When organizations use AI to justify headcount cuts, the employees doing the integrating often have the shortest runways. If you’re in that situation professionally, it’s worth understanding what data you’re generating and who owns it.
• Build agents that amplify your judgment, not replace it. The engineers at Meta who feel like ‘AI supervisors’ are experiencing something real — AI systems handling decisions that used to require human expertise, with humans reduced to oversight. The antidote, for personal AI at least, is intentional design: your agent handles logistics, research, and repetitive tasks. You handle anything that requires your actual expertise or relationships. Explore how agentic AI companies are building the future in 2026 to see what human-centered deployment actually looks like.

What Meta’s AI Rollout Actually Tells Us

• Meta deployed keystroke and mouse-tracking software on employee laptops with no opt-out option — confirmed by CTO Andrew Bosworth — specifically to generate AI training data from real human work patterns.
• The company is tying AI tool usage to performance reviews and cutting around 8,000 jobs simultaneously, creating a climate where workers feel they’re training their own replacements.
• Internal dashboards gamifying daily AI token consumption produced an unintended outcome: employees building AI agents to manage their other AI agents, defeating the original purpose entirely.
• The core lesson for personal AI users: the direction of control determines whether AI makes you more capable or more surveilled. Agents that work for you, with data you control, are categorically different from AI deployed on top of you by an institution with different incentives.
• When evaluating any AI platform or tool, the questions that matter most are: Who controls the data? What’s being tracked? Can you opt out? Meta’s employees couldn’t answer those questions in their favor.

The question isn’t whether AI changes how work gets done — it clearly does. The question is whether the AI in your life is working for you or collecting data about you for someone else’s model. Meta’s employees don’t get to choose. You do.