The Specific Event

California's SB 947, the No Robo Bosses Act, has cleared the state legislature and landed on Governor Gavin Newsom's desk with a 25-day window for signature or veto before the September 30, 2026 deadline. The bill's core provision is straightforward: codify human accountability in algorithmically-mediated workplace decisions. The framing assumes that inserting a human into the decision loop resolves the governance problem. That assumption deserves serious scrutiny before it becomes law.

What the Legislation Actually Proposes

The No Robo Bosses Act targets automated systems that direct, evaluate, or discipline workers without meaningful human oversight. On its face, this sounds like reasonable consumer and labor protection. The instinct behind it is not wrong. Rahman (2021) documented how algorithmic control systems function as what he calls an "invisible cage" - structuring worker behavior through automated constraint rather than direct supervision. The research problem is that mandating human sign-off on algorithmic decisions does not automatically produce meaningful oversight. It can just as easily produce a compliance theater where a human formally approves decisions they have no structural capacity to evaluate.

The Competence Problem the Bill Does Not Address

Here is the coordination failure the legislation sidesteps: a human supervisor placed in the loop between an algorithmic recommendation and a worker outcome is not automatically competent to evaluate that recommendation. This is precisely the awareness-capability gap that appears throughout algorithmic literacy research (Gagrain, Naab, & Grub, 2024). Supervisors can become aware that an algorithm flagged a worker for low productivity. Awareness alone does not give them the schema to assess whether the flag reflects genuine underperformance, a measurement artifact, a data quality problem, or a systematic bias in the training data. The legislation creates the structural position for human judgment without creating the conditions for that judgment to be meaningful.

This maps directly onto the distinction Hatano and Inagaki (1986) draw between routine and adaptive expertise. A supervisor trained to follow a checklist when reviewing algorithmic outputs has routine expertise. Routine expertise fails when the algorithmic environment shifts, when the model updates, or when edge cases fall outside the checklist's scope. What is actually needed for the legislation's intent to be realized is adaptive expertise - the capacity to reason from structural principles about what the algorithm is optimizing, where it can err, and what class of errors is most likely in a given context.

Algorithmic Coordination as a Distinct Mechanism

Classical organizational theory - markets, hierarchies, professional governance - assumes that coordination participants arrive with pre-existing competence relevant to the decisions they are making (Kellogg, Valentine, & Christin, 2020). A manager in a traditional hierarchy is presumed to understand the work being managed at some functional level. Platform and algorithmic coordination inverts this assumption. The algorithm develops competencies endogenously, through data, that no individual human in the organization necessarily possesses. This is not a temporary knowledge gap that training can close overnight. It is a structural feature of how these systems are built and deployed.

Schor et al. (2020) frame this in terms of dependence and precarity: workers become dependent on systems whose logic is opaque to them, and that opacity is not incidental but often deliberate. Extending Schor's analysis to the supervisory level, SB 947 risks creating a parallel form of managerial dependence. Human overseers become formally responsible for decisions whose generative logic they cannot access, which shifts liability without shifting actual control.

What Meaningful Accountability Would Require

The topology of the problem is different from the topography the legislation draws. Topography here means the visible surface: a human signs off on the algorithmic decision. Topology means the underlying structure: what that human can actually know, what the algorithm is optimizing, and where their respective competencies meet or fail to meet. Legislation that addresses topography without topology produces accountability in name without the conditions for accountability in practice.

A more structurally coherent approach would mandate explainability at the level of schema, not just individual decision logging. Not "the algorithm flagged this worker on date X for metric Y," but "the algorithm optimizes for these structural features, produces these known error types under these conditions, and here is what a qualified reviewer would need to know to override it responsibly." That is a harder standard to write into law. It is also the only standard that produces what Hancock, Naaman, and Levy (2020) would recognize as genuine human-AI collaborative oversight rather than human-laundered automation.

What Newsom's Decision Will Signal

Newsom signing SB 947 as written will tell us something important about where regulatory thinking currently sits relative to the actual coordination problem. The bill treats algorithmic accountability as a principal-agent problem solvable by inserting a human principal into the chain. The research literature on algorithmic work suggests it is a competence problem first. Until that distinction enters legislative reasoning, human oversight mandates will redistribute formal responsibility without redistributing real capacity to exercise it.