The Specific Event

Marc Lore, the billionaire founder behind Jet.com and various other ventures, has implemented an AI algorithm at his restaurant company Wonder to determine which employees receive promotions. As reported this week, Lore describes the system as "very objective," framing algorithmic evaluation as a corrective to the biases and inconsistencies of human managerial judgment. Wonder is not piloting this approach cautiously - it is using algorithmic output as a primary input into career advancement decisions. This is a meaningful organizational experiment, and it deserves serious scrutiny rather than the breathless coverage it has largely received.

What the Objectivity Claim Actually Means

The word "objective" is doing significant work in Lore's framing, and it is worth unpacking precisely what that claim entails. An algorithm can be procedurally consistent - it applies the same rules to the same inputs without fatigue or favoritism. But consistency is not the same as validity. The question is whether the inputs the algorithm measures are the inputs that actually predict effective leadership, creative problem-solving, or adaptive performance under novel conditions. Kellogg, Valentine, and Christin (2020) document extensively how algorithmic evaluation systems at work tend to capture what is easily quantifiable rather than what is organizationally consequential. The metric becomes the target, and the target displaces the underlying construct.

The Expertise Problem Lore Has Not Solved

There is a distinction in the expertise literature between routine expertise and adaptive expertise (Hatano and Inagaki, 1986). Routine expertise involves executing known procedures reliably under stable conditions. Adaptive expertise involves diagnosing novel situations and generating responses that the situation has not previously demanded. Algorithmic promotion systems are well-suited to reward the former and structurally blind to the latter. An employee who consistently hits measurable targets is visible to the algorithm. An employee who recognized an emerging problem before it became measurable, or who restructured a workflow in ways that benefited colleagues without generating individually attributable data, is largely invisible. Promoting the former systematically over the latter does not eliminate bias - it institutionalizes a specific bias toward legible performance.

The Organizational Theory Parallel

Palantir has been making a related argument in its enterprise AI pitch: that layers of middle management exist primarily to move information up and down hierarchies, and that AI can perform that function more efficiently, rendering those layers unnecessary. What both Wonder and Palantir are describing is a reorganization of organizational coordination around algorithmic intermediaries. This connects directly to Rahman's (2021) concept of the invisible cage, in which algorithmic systems constrain worker behavior through opaque rules that workers can observe in their effects but cannot directly inspect or contest. When a promotion algorithm determines career trajectories, workers face exactly this structure: they can see outcomes but cannot interrogate the logic producing them.

The Schema Deficit in Workforce Response

What is absent from coverage of Wonder's system is any account of how employees are supposed to understand and respond to it. Gagrain, Naab, and Grub (2024) distinguish between surface-level algorithmic awareness and structural schema: knowing that an algorithm exists is categorically different from understanding the structural features of how it weights inputs and generates outputs. Employees at Wonder may know they are being evaluated algorithmically. That knowledge alone does not tell them which behaviors the algorithm rewards, how it handles ambiguity, or what happens when measurable outputs conflict with organizational needs that are harder to quantify. Absent that schema, workers are likely to develop folk theories - individual impressions based on observed correlations - that may or may not map onto the algorithm's actual logic (Kellogg et al., 2020).

The Governance Question That Is Being Skipped

Mental health workers are currently pushing back through collective bargaining and legislation against algorithmic triage systems that determine patient care prioritization. The concerns being raised there - about accountability, contestability, and the displacement of professional judgment - apply with equal force to algorithmic promotion systems in corporate settings. The difference is that mental health workers have organized a visible response, while Wonder's employees have not, at least not publicly. Schor et al. (2020) note that platform-mediated dependence tends to suppress collective action because workers internalize the algorithm's authority as a structural fact rather than a governance choice. Lore's framing of the system as "objective" performs exactly this function: it presents a design decision as a neutral feature of reality, foreclosing the question of whether a different design was possible.

Why This Matters Beyond Wonder

Wonder is not a platform company in the conventional sense, but it is importing platform governance logic into a brick-and-mortar organizational context. If algorithmic promotion produces measurably better outcomes at Wonder, that result will be cited as evidence for broader adoption. If it produces worse outcomes, the failure will likely be attributed to implementation rather than to the underlying model. Either way, the governance structure - who designs the algorithm, what it measures, and who can contest its outputs - deserves more scrutiny than it is currently receiving. Framing the question as "AI versus human bias" forecloses the more productive question: what institutional structures ensure that algorithmic evaluation systems remain accountable to the organizational goals they are meant to serve.