Issue No. 10 - THE CLARITY MEMO: Unfiltered.

The Misdiagnosis Cost.

On Human Impact & the Questions AI Can’t Ask.

Founder’s Note: This issue is landing a day later than it should. Life – responsibilities, travel disruptions, and conflicting priorities – adjusted for work this week, work adjusted right back, and that’s the argument I’m making in this issue. The average AI productivity tool would flag this lateness as a dip, with no room for the why. You, a human reading this, can relate to it. Read on to see why this is a bigger deal for your organization than you think.

The Memo.

Many of you know Derek Mobley's name by now via the nationwide class action against Workday, after he applied to more than 100 jobs ran through its AI screening software and was rejected every time. He's Black, over 40, and lives with anxiety and depression. In June 2026, Judge Rita Lin let most of the discrimination claims move forward, holding the vendor that built the tool accountable, and not just the employers who used it.

The case is still active, and it's deciding something that should worry most executives: how much distance a software company, or an employer, can put between itself and the decisions made.

That distance is short, and getting shorter. Workday itself estimates roughly 1.1 billion rejected applications during the relevant time period, which confirms that "the machine made the recommendation"is no longer a defense for the vendor, or for you.

Then there's Meta. In July 2026, 26 current and former employees sued in federal court, alleging that Metamate (Meta's internal AI tool) scored and ranked employees for a layoff that disadvantaged workers with disabilities, workers on protected medical and bereavement leave, and pregnant employees.

Metamate has no capacity for judgment. It can't ask why someone's productivity dipped, it can only clock that it did, reading the absence of keystrokes and token activity as low return on investment. It didn't know an employee was recovering from surgery, newly postpartum, or grieving the loss of a loved one.

That missing common sense context is the difference between a medical leave and a termination notice. And a termination notice is a mortgage payment, a COBRA deadline, or a tuition bill suddenly at risk.

If this can happen inside Meta – with its engineering expertise, legal resources, and financial ability to absorb the cost of getting it wrong – what would your organization's version of Metamate have done with the same blind spot? Would anyone have caught it before the notice went out?

And what would be the misdiagnosis cost of that, the price of treating a human circumstance as a productivity problem?

There are two tables these issues sit on – the one that decided what the tool would measure, and the one now explaining what it did. Neither is the one that pays the price.

A line from Pope Leo XIV's first encyclical, published two months before the Meta suit was filed, now reads less like theology and more like a forecast: technology is never neutral, because it takes on the characteristics of those who devise, finance, regulate, and use it.

Metamate's engineers likely never set out to penalize maternity, disability or bereavement leave.

They just built a productivity measure out of keystrokes and token counts, without the humanity to catch what it would fail to capture.

The Unfiltered Take.

That's the misdiagnosis cost.

And what gets buried in coverage of both lawsuits is the asymmetry of time attached to it. Lawsuits take years, whereas the impact of job loss is immediate.

Meta's lawyers can litigate this across a full product cycle, settle quietly, or win on a technicality. The parent who lost their paycheck to a flawed algorithm doesn't get to wait for a judge's ruling to cover this month's mortgage. Neither do medical bills nor collection agencies grant clemency because a machine made a life-altering, costly mistake. That asymmetry is the real anchor underneath both cases, and it points to something more than "AI bias" or "lack of oversight."

The first table decides what productivity measurement looks like, under the assumption that keystrokes and token activity are a fair proxy for value, with no room for what a real, lived human experience entails. Google's own AGI Safety and Alignment Team didn't trust their internal hiring AI enough for their own applicants – a leaked memo warned candidates of a "non-trivial probability" of being incorrectly screened out, then routed them to an additional form so a human would actually read their resume. The real, lived human experience behind that memo is the frustration of having filled out the nth application of the day, only to be told "here's more work, because our tool might have already failed you."

I know this blind spot personally. My leadership style is to stay quiet in meetings, trusting my team to make the calls, and speaking up only when my perspective is needed. Most productivity tools would read that as low engagement, but they can't account for the why. My silence is a deliberate means of empowering my team, not dominating the room. So, a leadership strength is now penalized by a machine that only knows how to capture word count.

The second table is the one that accounts for what the first one built. Meta's legal team, HR, and executives are now scrambling to explain a biased output from a performance metric they never personally reviewed. This is the accountability table, where governance and risk exposure live, and the scrutiny no company or leader wants.

What's missing from both is a third table.

One that seats the twenty-six plaintiffs whose livelihoods are on hold pending a lengthy lawsuit. The employees who've learned to game the system and stopped trying to do meaningful work. And the Derek Mobleys of the world, refreshing a job board at 2 a.m., deciding whether to voluntarily disclose their cultural identity or demographic on questionnaires that allegedly won't affect their candidacy.

This is the human impact. The one leaders sacrifice when they take the easy way out of designing human judgment into their tech processes and organizational strategies.

I've made this argument before, in Issue No. 02, about workforce data, and in Issue No. 04, about algorithmic damage. This is the sacrifice that happens when productivity systems are built on a Euro-American baseline and rolled out globally as though neutral. It happens when training data learns that success looks like one type of person and treats everyone else as risk.

An algorithm that assumes the "default" employee doesn't get pregnant, take leave, carry a disability, or sit outside Eurocentric norms, will always read deviation as deficiency.

A more moral AI isn’t enough to fix this. Neither is a more scrutinized vendor process. What fixes this is moving the third table to the front of the room. Put human impact first, and accountability, governance, and the cost of future misdiagnosis all start to shrink.

The Action.

Pull whatever activity or productivity dashboard your organization runs, and hold it up against your own calendar from the last month. Look at what it would have flagged – the meetings you sat quiet in or skipped entirely, the personal days you took for doctors’ appointments and errands, , the blocked times that don't show up as output. Then ask yourself, honestly, would said tool have penalized you for any of it?

If the answer is no, ask why not.

If your title is the only difference between you and the misdiagnosis cost your employees are absorbing right now, then that’s a decision with your name all over it.

Lawsuits can take years to resolve. Your next AI-influenced decision about your people won't wait that long, and neither will they.

If you know a leader who owns a decision an algorithm is currently making for them, forward this their way. The Clarity Memo: Unfiltered drops bi-weekly. Subscribeto get it directly to your inbox.

This content is for informational purposes only and does not constitute professional, legal, financial, or organizational advice. For guidance specific to your organization, contact Fadéké Strategic Consulting, LLC at admin@fadeke.com