The Specific Event

This week, approximately 1,900 unionized workers at Blizzard Entertainment ratified their first collective bargaining agreement with Activision Blizzard, securing explicit protections requiring management to bargain with staff before deploying generative AI in ways that affect their work. The contract does not ban AI. It does not freeze current tooling in place. What it does is insert a procedural checkpoint into the otherwise unilateral authority employers have historically exercised over technology adoption decisions. That is a structurally interesting development, and I think most of the coverage is misreading what it actually represents.

What the Contract Does and Does Not Do

The coverage has framed this as workers "securing protections against AI." That framing is understandable but imprecise. What workers actually secured is a right to bargain over AI deployment, not a veto over it. The distinction matters enormously. A bargaining obligation creates a mandatory coordination mechanism between management and labor before implementation. It does not determine outcomes. Blizzard can still deploy generative AI tools that reduce headcount or restructure workflows. It simply has to negotiate the terms before doing so. This is less a shield against AI and more a formalization of the governance channel through which AI deployment decisions flow.

This distinction between access to a coordination mechanism and control over its outputs maps directly onto a problem I study in platform contexts. Kellogg, Valentine, and Christin (2020) documented how algorithmic systems create accountability gaps precisely because workers interact with outputs - task assignments, performance scores, content moderation decisions - without access to the decision architecture producing those outputs. The Blizzard contract does not solve the accountability gap. It opens a door into the room where the architecture is being designed, without specifying what workers are allowed to do once inside.

Why Governance Structure Is the Real Variable

The organizational theory question buried in this story is not whether AI will cost jobs at Blizzard. It is whether a bargaining obligation changes the epistemic conditions under which AI deployment decisions get made. Rahman (2021) describes how platform firms use what he calls "the invisible cage" - policy structures that constrain worker behavior while appearing neutral and technical. The Blizzard contract is an attempt to make that cage partially visible, to require that its dimensions be disclosed and debated before workers are placed inside it.

That is a meaningful governance innovation, but it carries a significant structural risk. Bargaining over AI deployment presupposes that union representatives have sufficient technical understanding to evaluate the proposals management brings to the table. If that understanding is shallow, the bargaining process becomes performative. Management presents a deployment plan using technical framing the union cannot effectively interrogate, the union raises procedural objections, and the outcome converges on whatever management intended anyway. The contract creates a coordination mechanism without guaranteeing the competence needed to use it.

The Competence Problem Inside the Governance Channel

This is where the parallel to my own research becomes direct. The awareness-capability gap I work on in platform contexts holds that knowing an algorithm exists, or knowing that a deployment decision is being made, does not translate into the ability to respond effectively. Gagrain, Naab, and Grub (2024) show that algorithmic media literacy - the ability to understand how systems shape outcomes - does not automatically follow from exposure or even from formal disclosure. The same logic applies here. Blizzard workers now have formal notice and a seat at the table. Whether they have the schema-level understanding needed to evaluate AI deployment proposals in technical and organizational terms is a separate question that the contract does not address.

Hatano and Inagaki (1986) draw a useful distinction between routine expertise, which is the ability to execute established procedures well, and adaptive expertise, which is the ability to reason from structural principles when conditions change. A union representative with routine expertise can flag obvious violations and escalate familiar grievances. An AI deployment negotiation requires adaptive expertise: understanding what classes of system design choices tend to produce which categories of labor impact, and being able to transfer that understanding across different implementation scenarios. That kind of expertise is not acquired through contract language alone.

What This Signals About AI Governance More Broadly

The Blizzard contract will be cited widely as a template, and Canadian bank executives are already fielding questions about job displacement as their AI investments accelerate. What the contract reveals is that the field is currently building governance infrastructure without building the epistemic infrastructure required to operate it. Procedural rights without structural comprehension produce coordination theater rather than coordination. The more interesting organizational design question for the next several years is not which industries will unionize AI oversight, but what kinds of training and knowledge transfer would need to accompany those oversight mechanisms to make them function as intended.