The Event That Matters

Last week, President Trump hosted a lunch with tech leaders including Elon Musk that produced what participants described as a "morally binding" safety accord on AI development, along with a proposed rebranding of artificial intelligence itself. The details that have surfaced are worth pausing on. A safety agreement described as morally binding rather than legally binding is not a governance instrument. It is a folk theory of governance, dressed in the language of institutional commitment. This distinction is not semantic. It points directly to a structural problem in how organizations are currently attempting to coordinate around AI risk.

Moral Binding as Folk Theory

In my dissertation research on Algorithmic Literacy Coordination, I distinguish between folk theories and structural schemas. Folk theories are the impressionistic, often accurate-feeling beliefs that individuals develop about how systems work. Schemas are accurate structural representations of constraint architectures. The problem is that folk theories can generate confident behavior without generating effective behavior. A "morally binding" accord among competitive technology executives fits this pattern precisely. Each party at that lunch likely holds a different internal model of what the accord requires, enforces, and permits. There is no shared schema, only shared language.

This matters because Kellogg, Valentine, and Christin (2020) documented extensively how algorithmic governance creates coordination problems that surface-level awareness cannot resolve. Knowing that an algorithm exists, or in this case knowing that a safety norm exists, does not translate into knowing how to respond when the norm conflicts with competitive incentive. The awareness-capability gap applies to institutional actors just as readily as it applies to individual platform workers.

The Rebranding Signal

The reported proposal to give AI a new name deserves separate attention. Renaming a technology during a governance conversation is a topographic move, not a topological one. Topography describes the surface features of a landscape. Topology describes its underlying structure. Renaming AI changes what the conversation looks and feels like without altering the underlying structure of algorithmic agency, competitive dynamics, or information asymmetry between developers and regulators. Sundar (2020) argues that machine agency creates attribution challenges that fundamentally alter human communication and decision-making. A new label does not resolve those challenges. It relocates them.

What Clio's Acquisition of Learned Hand Tells Us in Comparison

Consider a different AI governance story from the same news cycle. Clio, a legal practice management company, acquired Learned Hand, a startup building AI tools specifically for judges and law clerks. This is a structurally different approach to the problem. Rather than producing a voluntary accord among competing principals, it embeds AI coordination into the institutional layer of the court system itself, where constraints are formal, roles are defined, and accountability mechanisms already exist. The schema is not left to individual interpretation. It is encoded in institutional procedure.

Hatano and Inagaki (1986) distinguish between routine expertise, which performs well in familiar contexts but fails under novel conditions, and adaptive expertise, which transfers across contexts because it is grounded in structural understanding. The Trump lunch accord optimizes for routine expertise. It assumes that existing competitive relationships and reputational incentives will hold under novel AI-specific pressures. The Clio-Learned Hand model is attempting, at least structurally, to build schema-level coordination into a domain that already has enforcement architecture.

The Organizational Theory Gap

What both stories expose is the absence of what I would call competence-assuming governance. Classical coordination theory, whether market-based, hierarchical, or network-based, assumes that actors arrive at the coordination problem with pre-existing competence about what the system requires of them. Rahman (2021) and Schor et al. (2020) both document how platform environments invert this assumption, generating competence endogenously through participation. AI governance as currently practiced makes the opposite error. It assumes competence that does not yet exist, then treats voluntary agreement as a substitute for the structural scaffolding that would actually build it.

A morally binding accord among executives who hold fundamentally different internal models of AI risk is not a coordination mechanism. It is a coordination gesture. The organizational theory question worth asking is not whether the accord will hold, but what institutional architecture would need to exist for a genuine shared schema to emerge in the first place. That question did not appear to be on the lunch menu.

The Practical Implication

For researchers studying organizational responses to AI, the Trump lunch and the Clio acquisition belong in the same analysis, not separate ones. They represent divergent bets on where coordination competence comes from. One bets on moral commitment among powerful individuals. The other bets on institutional embedding. The variance puzzle in platform outcomes suggests that identical access produces dramatically different results when structural schemas are absent. There is no obvious reason why AI governance should be exempt from this finding.