Consider these four sentences.
“It’s cold in here.” “Could you close the window?” “I’ll close the window.” “The window is closed.”
In the transmission model of language, these are four descriptions of varying accuracy about a physical situation. Temperature, possibility, intention, state.
But watch what they actually do.
The first creates a shared condition, an opening, an implicit bid for response. The second is a request, it places an obligation on you, creates a social fact that didn’t exist a moment ago. The third is a promise, it commits the speaker to a future action and makes them accountable for it. The fourth is an assertion, it stakes a claim about reality that can be held against the speaker if wrong.
Same general territory. Completely different social actions. Different things created in the world. Different commitments generated. Different relationships altered.
J.L. Austin spent his career cataloguing what language does rather than what it describes. John Searle systematized it into speech act theory. And Fernando Flores saw something neither of them fully articulated, that speech acts are not just an interesting feature of language. They are the primary substrate of human coordination. The atoms of social reality. The mechanism by which organizations, relationships, and commitments come into existence and are maintained.
Flores identified a small number of fundamental speech act categories that between them cover most of what matters in human coordination.
Assertions, claims about the state of the world. Staking a position that can be evaluated as true or false. Taking on accountability for accuracy.
Declarations, utterances that create new reality by being spoken. “You’re hired.” “This meeting is adjourned.” “I pronounce you married.” The saying is the doing. No further action required for the act to be complete.
Requests, bids for future action from another. They open a space of response, the other can accept, decline, counter-offer, or commit to respond later. A request that is neither accepted nor declined leaves something unresolved hanging in the social space between people.
Promises, commitments to future action by the speaker. They create accountability. They alter the relationship between speaker and listener. A promise broken is not just a failed prediction, it’s a different kind of thing entirely. A betrayal of the social reality the promise created.
Assessments, evaluations. Judgments about quality, fit, value. These are neither purely descriptive nor purely performative, they’re grounded in the background of the assessor and cannot be fully separated from it.
Now here’s the move Flores makes that’s genuinely original. He says that organizational breakdown, the failure of coordination, the missed deadline, the dropped ball, the systemic dysfunction, is almost never a technical failure. It’s almost always a breakdown in the network of speech acts. A request that was never clearly made. A promise that was assumed rather than spoken. A declaration that was heard differently by different parties. An assessment that was private when it needed to be public.
The organization isn’t broken. The conversation is broken.
And because we were taught to look at organizations as structures, systems, processes, visible things, we keep trying to fix the wrong level.
Here’s what this means for how you interact with AI right now, today.
Every prompt you write is a speech act. Usually a request, sometimes an assertion establishing context, occasionally a declaration setting up a role or frame. And the quality of what comes back is directly related to the quality of the speech act you made.
Vague request. Vague response. Not because the AI is limited, but because you didn’t complete the social action. You left the request unresolved. Conditions of satisfaction unstated. Background context unspecified. The AI filled the gap with statistical probability rather than understanding, because understanding requires the background you didn’t provide and it doesn’t have.
This is why prompt engineering works when it works, it’s the intuitive discovery of speech act structure. Specify the conditions of satisfaction. Establish the context. Make the request complete. The response improves not because you found a magic formula but because you made a real request instead of a gesture toward one.
But here’s the deeper implication. The AI can receive speech acts. It cannot make them. Not really.
When an AI says “I’ll help you with that”, no commitment has been created. No social reality has shifted. No accountability exists. The form of a promise appeared but none of the substance. Because promises require a promiser who can be held to account across time, who has something at stake, who exists in a social reality where betrayal means something.
The speech act landed in you. Nothing landed in it.
Which means every interaction you have with AI is structurally one-sided in a way that feels bilateral. And that asymmetry, invisible, persistent, consequential, shapes everything about what AI can and cannot be for you.
@ Lee Greene