The Way You Work IS Training Data For AI
The one thing you need to know in AI today | AI-Ready CMO
Last week, Meta started recording its own employees. Mouse movements, keystrokes, screen snapshots. All captured silently on every U.S. work laptop, with no opt-out. The tool is called the Model Capability Initiative. The purpose is to train AI agents to perform computer work the way humans actually do.
Meta says the data collected will not be used in performance reviews. Of course, they won’t. They also say they have safeguards for sensitive content. Of course, they have. Who wouldn’t believe Meta? But this dystopian surveillance-state-like extreme ignorance of personal privacy is just one side of the picture. And, of course, this could only happen in the US. Good luck doing this in Europe.
But we are not some geopolitical magazine, and we have a very extreme view on privacy as well. We also believe these things are kinda inevitable anyway, so let’s focus on what we can learn from this today.
In 2026, we started automating all knowledge work.
Because when something becomes training data for AI, it is just a matter of time before it gets automated.
Meta is reportedly spending up to $135 billion on AI infrastructure in 2026. They will lay off roughly 8,000 people and leave another 6,000 roles unfilled (this has already started in smaller batches). And in the middle of all that, they are paying engineers to generate workflow data by working. That tells you something specific: the bottleneck for the next generation of AI agents isn’t compute, isn’t model architecture, isn’t talent. It’s data. Specifically, agent-quality data. Examples of skilled humans completing real work on real software, in sequence, with judgment. That data does not exist at scale. So Meta is manufacturing it.
What marketers can learn from this
The companies that win the agent era won’t be the ones with the best tools. They’ll be the ones who own the best record of how their teams actually do the work.
Right now, your marketing team’s expertise lives in three places: the heads of your senior people, the half-finished SOPs in Notion, and the residue in Slack threads nobody can find. None of that is structured. None of it is reusable. None of it can train an agent to brief a campaign, screen a vendor, or write to your brand voice.
In twelve months, that’s going to be the difference between teams that scale and teams that get scaled. If you can hand a model a structured library of how your team plans a launch, qualifies a lead, or reviews creative, you compound. If you can’t, you’ll rent that capability from a vendor whose agents were trained on someone else’s data, at someone else’s price. Meta is doing the extreme version of this with surveillance.
You don’t need surveillance. You need intent. If you haven’t started documenting your workflows with extreme precision, today is the day to start.
What to do next
Three moves, in order, this quarter, starting today: