Meta is installing monitoring software on the work devices of thousands of employees and contractors across the US. The program, called the Model Capability Initiative, captures keystrokes, mouse movements, and screenshots while workers use common applications. The goal: train AI agents that can replicate how humans interact with internal systems.

Meta is simultaneously planning to cut roughly 8,000 jobs, creating a situation where employees are effectively helping build the tools that could make their roles redundant.

What the Model Capability Initiative actually does

MCI is a structured telemetry collection system that records granular details about how employees work on their machines. The software activates when approved applications are in use, operating on a whitelist system. The data types being collected include keystrokes, mouse movements, and periodic screenshots. All of this feeds into training internal AI agents designed to mimic human interaction patterns across Meta’s software infrastructure.

Employees cannot opt out of the program. When participation is mandatory, the power dynamic between employer and employee shifts in ways that labor advocates have been warning about for years.

There are also legitimate concerns about personal data exposure. Even with a whitelist approach that limits recording to approved work applications, the nature of keystroke logging and screenshot capture means personal information could easily get swept up in the collection process.

The AI adoption mandate

Meta has set aggressive performance targets tying employee evaluations to AI tool adoption. The target is significant AI usage across software engineering workflows by 2026.

Meta has been investing heavily in AI infrastructure under CEO Mark Zuckerberg’s direction, positioning the technology as central to the company’s future across products, internal operations, and developer tools. The MCI program represents the internal-facing side of that bet, essentially using Meta’s own workforce as both the training data source and the eventual benchmark against which AI agents will be measured.

Layoffs and the uncomfortable math

The roughly 8,000 planned job cuts add a layer of urgency to the conversation. Critics have pointed out that this approach fundamentally changes how employee contributions are valued. Traditionally, institutional knowledge and workflow expertise were assets that made individual workers harder to replace. Under MCI, that same expertise becomes an extractable resource, something that can be captured, encoded, and deployed without the person who originally developed it.

What this means for the tech workforce

The risk is reputational and legal. Mandatory telemetry collection without opt-out will almost certainly face scrutiny from regulators, particularly in jurisdictions with strong data privacy frameworks. And the optics of training AI on employee work while simultaneously laying off thousands of workers could make recruiting harder at a time when Meta still needs top talent to build the very AI systems it’s betting its future on.