Metamate is the model Meta uses to measure productivity: it tracks lines of code, closed tasks, and project involvement to produce scores that feed into decisions about staff cuts. The lawsuit filed by twenty-six former employees in a California federal court argues that this tool never differentiated between unproductive workers and people undergoing cancer treatment or on protected leave.
The process is easy to describe but hard to accept. The model generates ratings without adjusting for medical leave, disability leave, or family caregiving. Someone receiving chemotherapy and someone who simply stopped trying end up classified the same way.
The layoffs were set to begin around July twenty-second, twenty twenty-six. The former employees are asking that they be halted while the case moves through the courts. Meta denies the allegations and insists that all decisions were made by people, not artificial intelligence.
That claim reveals the blind spot. How do you pursue a discrimination case when an algorithmic score stands between the employee and the human decision-maker?
The answer lies in the nature of existing law. Statutes like the FMLA require reasonable accommodations on a case-by-case basis. A model that doesn't incorporate protected status as a variable cannot make those accommodations by design.
I've seen in different contexts how numbers generated by these tools become unquestionable. Managers prefer to follow the score rather than defend an exception the model doesn't record.
This recalls production quotas in ancient civilizations, where clay tablets showed targets that made no allowance for droughts or illness. Stones don't lie, and those archaeological records confirm that ignoring human context in metrics is nothing new. The subject deserves further exploration, since it reveals patterns that outlast today's technology.
Meta's argument, while technically true, doesn't resolve the underlying problem. If people saw the low scores of colleagues on medical leave and decided to proceed anyway, then both the model and the organizational culture contribute to the outcome.
The pattern keeps repeating. The European Union softened parts of its AI law through the Digital Omnibus package, pushing back requirements until twenty twenty-seven. As explored in The Generosity in the Doorway, when examining who drafts the terms, regulation often arrives late to the real debate over control.
I still don't have a clear sense of the exact balance between useful metrics and effective protections. A solution worth exploring would force human review whenever protected status is detected, so that the score doesn't even enter the overall layoff calculation.
What does this mean for labor protections when decisions are filtered through algorithmic layers?
This question forces us to reconsider whether laws written for a world of direct oversight can adapt to dynamics where accountability is fragmented between code and human judgment. The gap between declared oversight and actual oversight is where most of the harm happens.
Stones don't lie, but historians sometimes do. Should we be more worried that people trusted the number without questioning where it came from?
