Meta has started capturing detailed employee computer activity, including keystrokes, mouse movements, clicks, and screen interactions, under a new internal program designed to train workplace AI agents. The initiative, called the Model Capability Initiative (MCI), is being deployed across U.S.-based employees and records how work is performed inside company systems in real time.
The data is being used to train AI systems that can navigate software, execute workflows, and handle routine office tasks with minimal human input. Instead of relying only on internet-scale datasets, Meta is now building its models using internal behavioral data generated during actual work.
MCI Operational Framework
The MCI system runs in the background on company-issued devices and captures how employees interact with digital tools during their daily work.
It records typing behavior, mouse activity, navigation patterns across applications, and periodic screen context. This data is structured into training inputs that help AI systems understand how tasks are completed across software environments.
This approach directly addresses a well-documented limitation in AI systems. According to research from McKinsey & Company, most enterprise AI deployments struggle not with intelligence, but with execution inside fragmented digital workflows. Systems can generate outputs, but often fail when required to operate across tools, interfaces, and multi-step processes.
Autonomous Workplace AI Systems
Meta is using this data to develop AI agents that can function inside enterprise environments with a higher level of operational accuracy. These systems are being designed to navigate software interfaces, complete structured tasks, and reduce dependency on manual execution.
Industry forecasts suggest this is a rapidly growing area. Gartner estimates that by 2028, nearly one-third of enterprise software interactions could be handled by AI agents, particularly in roles involving repetitive digital workflows.
Meta’s strategy differs in one key way: it is grounded in live employee behavior rather than simulated datasets, giving the company access to highly contextual training inputs that are difficult to reproduce synthetically.
Privacy and Data Concerns
The rollout has triggered internal concerns, particularly because the tracking system is mandatory on work devices and does not offer a clear opt-out mechanism.
Employees have raised questions about continuous monitoring and the long-term use of behavioral data. Meta has stated that the data will not be used for performance evaluation and includes safeguards to filter sensitive content.
However, governance challenges remain. Research from the Organisation for Economic Co-operation and Development highlights that workplace data collection at this level introduces risks related to consent, data reuse, and behavioral distortion, especially when employees are aware they are being monitored.
AI Growth and Job Cuts
The initiative comes as Meta accelerates its investment in artificial intelligence while preparing for significant organizational restructuring.
Reports indicate that the company is planning workforce reductions affecting around 10 percent of its global workforce, or approximately 8,000 employees, as it moves toward a more AI-driven operating model.
This overlap has drawn attention because the same workflows being captured today are likely to contribute to automation systems that reduce reliance on similar roles in the future. At the same time, Meta continues to increase spending on AI infrastructure, positioning itself more aggressively in the global competition to build advanced AI systems.
Workflow Data Economy
Meta’s latest move reflects a broader change in how competitive advantage is emerging in artificial intelligence. Public internet data powered the first wave of AI models, but enterprise-grade systems now depend on access to real operational workflows.
Research from Stanford Institute for Human-Centered Artificial Intelligence shows that models trained on structured human interaction data perform more reliably in applied environments than those trained only on static datasets.
By capturing how work is actually executed inside its systems, Meta is positioning its AI development around real-world performance rather than theoretical capability.