The Signal

The Lead

The new data problem is your own employees

This week, Meta confirmed it will begin logging screenshots, keystrokes, and mouse activity on roughly eight thousand U.S. employees' laptops. No opt-out. The stated purpose is training internal AI agents on real workflows — how engineers actually use VSCode, how support teams actually move through Gmail, how managers actually stitch together five tools to close a task.

Set aside the Meta-specific optics for a moment. The pattern underneath is the one that matters, and it is coming to every enterprise in some form within the next eighteen months.

Here is the shift. For the last three years, the dominant enterprise AI question was which tool to buy. Copilot or Cursor. Glean or Notion AI. The tool question is a distraction. The useful AI inside a company is the one that has learned how that company actually works — which exceptions get escalated to whom, which steps in the approval flow are real and which are vestigial, which decisions require judgment and which can be codified. None of that lives in your SaaS documentation. It lives in the screens and keystrokes of your best operators.

Meta's program is the blunt-force version of capturing it. The thoughtful version — the one most organizations will end up running — looks different. It starts with logic extraction, not surveillance. You sit with the twenty people whose judgment you most want to scale, you map their decision criteria, you encode the rules and the edge cases, and you build the AI on top of that explicit foundation. Workflow data supplements the model. It does not substitute for the thinking.

The operators who get this right in the next year will do three things. They will treat employee workflow data as a strategic asset and build governance for it now, not after legal asks hard questions. They will invest in capturing decision logic explicitly through structured interviews and case libraries, not only implicitly through telemetry. And they will be honest with their people about what is being captured and why, because the organizations that blow trust on this transition will spend the following decade trying to earn it back.

The tool-vendor conversation will keep getting louder. Do not let it distract you from the data conversation underneath it. That one determines which AI systems actually work inside your walls.

From the Practice

Eleven tools, zero conventions

A recent Discovery Sprint with an SMB services client surfaced a familiar pattern. Eleven different AI tools in active use across thirty people. Zero shared conventions. And no one able to answer the question of which workflows had actually been changed by any of them.

The fix was not another tool. It was a one-page workflow map that identified the three decisions the organization made most often, and a rule for which of those AI would own, which humans would own, and which required both. That map is now the foundation for everything else.

If this landed, hit reply. These come out weekly and the conversations with readers make them better.

Or explore upcoming training at resonai.ai/training.