Meta is collecting internal data on employee mouse movements, keystrokes, and screen activity to train artificial intelligence agents capable of performing complex computer tasks for future workplace applications.

Key Points

• Meta’s Model Capability Initiative will record employee interactions, including clicks and navigation, to improve AI agent performance.
• CTO Andrew Bosworth confirmed the initiative, now rebranded as the Agent Transformation Accelerator, aims to automate routine workplace workflows.
• The company stated that collected data will be used exclusively for model training rather than employee performance evaluations.
• Spokesperson Andy Stone noted that real-world examples of human computer usage are necessary to teach AI how to navigate software interfaces effectively.

Why it Matters

This initiative signals a shift toward developing autonomous AI agents that can observe and replicate human workflows to increase corporate productivity. By leveraging internal data, Meta aims to create systems that transition from simple chatbots to active participants capable of managing complex digital tasks.