Companies want real work data to train AI agents. Employees want productivity, but they don't want to become raw material. Caio and Marina break down the Meta case, the tension between surveillance and consent, and how to do this without torching your team's trust.
In this episode
01 The Hook: You Are the Dataset
- Marina opens with a provocation: what if your company had a camera trained on the way you work, just to teach a robot to replace you? How would that feel?
- Caio brings up the Reuters story: Meta had an internal plan to track mouse movement and keystrokes to train AI, and backed off after the team pushed back
- Why this hits everyone: nobody wants to be productive just to feed the machine that's learning to do their job
- Name the core conflict: the company wants real data, the employee doesn't want to become raw material
02 Why Companies Want This Data
- Caio explains the practical side: a good AI agent needs real examples of how the work actually gets done, not some idealized manual
- The difference between training on synthetic data and real operational data, and why the real stuff is worth so much
- A concrete example: customer service, sales, support, where the history of conversations and decisions becomes gold for training an agent
- Marina asks the obvious: so does the company really need to spy on me for this? Caio separates process data from individual surveillance
03 The Line Nobody Should Cross
- The turn: there's a difference between capturing the result of the work and capturing the person working, keystroke by keystroke
- Marina raises the audience's point: is this surveillance dressed up as innovation? Where's the real consent
- Caio is blunt: tracking individual mouse and keystrokes torches trust and isn't worth the technical payoff
- The hidden cost: when the team finds out they've become guinea pigs, productivity drops, people leave, and the AI project dies with it
04 How to Do It Right Without Burning the Team
- Caio gives the how-to: start with the data that already exists and that the company clearly has the right to use, like CRM records and tickets
- Transparency as a rule: tell the team what's being collected, what for, and give them some control
- The idea of collecting the process, not the person: anonymize, aggregate, focus on the workflow and not on who did it
- Marina asks: what if an employee refuses? Caio talks about incentive and participation, making the team win with AI, not lose to it
05 Practical Wrap-Up
- Caio sums up the thesis: real work data is valuable, but trust is even more valuable, and it doesn't come back once it's broken
- Three questions every leader should answer before collecting: does the person know, does the person agree, and does the person gain something from it
- Marina ties it back to the title's provocation, now with the answer in hand
- A final invitation to reflect and to reach out for anyone who wants to apply AI without becoming a bad headline