One lens, six traditions
The method has not changed since 1995. It treats a person using a machine as situated, interactional conduct — and looks for the moments where understanding breaks and gets repaired, because that is where the design reveals itself.
It first took shape in a conversation-analytic usability study of an early agent-based PDA, and it is the same frame behind PRISM, a present-day lab that records a session as parallel, time-aligned tracks. Six traditions each contribute one commitment. Under all of them runs a single operating rhythm I spent my doctorate on — Plato’s dialectic of collection and division: gather scattered observations into the pattern that shows only across them, then cut it at its natural joints (Collection and Division).
A note on dates: the commitments are older than the fieldwork. They first formed in a 1994 master’s thesis that reopened Claude Shannon’s model of communication to argue that meaning is relational and statistical, not a payload transmitted between minds — the view transformers would later vindicate at scale (Returning to the Source). The 1995 usability study turned that philosophy into an empirical method. The six traditions below are how the same commitment shows up, one tradition at a time.
Conversation analysis
Sacks, Schegloff, and Jefferson. The stance is “why that now?” rather than “what category does this belong to.” Predetermined coding schemes presuppose the phenomenon they claim to find; the discipline is to return to the record and let structure show itself turn by turn. Meaning is displayed in the next action — which is exactly why an agent should be designed for repair, not one-shot intent capture. And because the categories are behavioral, they port: repair success, adjacency completion, and sequence closure work as evaluation metrics for multi-agent systems, read off a transcript the way an analyst has always read them.
Situated action — Lucy Suchman
Human action is not the execution of a pre-formed plan; it is moment-to-moment coordination with a physical and social environment. Plans are resources for situated action, not determinants of it. The methodological consequence: the moderator’s question, the participant’s gesture, and the screen’s state are all jointly producing what happens — none is privileged as “the real data.”
Distributed cognition — Edwin Hutchins
Cognition is not sealed inside a head; it is enacted across people, artifacts, and representations. A reflective instrument the participant manipulates is part of the act of thinking, not a readout of a thought. The same claim, scaled up, is the argument that the intelligence is in the coordination — in the system of people, agents, tools, and records, not in any single model.
Frame analysis — Erving Goffman
Interaction is organized by frames that participants jointly maintain, and breakdowns happen when a frame fails and must be repaired. The work is to catch the overshoot-and-walk-back — the moment a participant senses a frame breaking before they can say so. Consistency of frame, the 1995 study concluded, is paramount to usability.
Breakdown and unconcealment — Winograd & Flores
A breakdown is not a failure to be hidden but a moment of non-obviousness: when a tool stops working silently, the whole nexus of tools and expectations becomes visible. Good design anticipates the forms of breakdown and provides a space for action when they occur. Build the instrument to expose those moments — and build the interface as an inspectable contract with visible seams rather than a false-smooth surface.
Talk and gesture — Goodwin & Goodwin
Talk and gesture are mutually contextualizing: the words guide the reading of the gesture while the gesture elaborates the words. This is what lets a physical action — a hand tracing a frozen progress bar, a knob walked back — be read as a turn in the conversation, carrying meaning that eluded several turns of speech.
What it commits me to
Five things, in every project from the PDA to the AI interface: meaning is situated; breakdown is where the truth is; the surface lies, so attend to the gap between saying and doing; representation is itself a design act with consequences; and consistency of frame is the ground everything else stands on. The five are unpacked on The Practice.
And the question of who you talk to
The same lens turns from the user onto the machine. If a persona is real only when it is realized in behavior — Goffman’s frames and Goodwin’s gesture pointed at the system — then “what am I talking to?” becomes an empirical question rather than a metaphysical one. The philosopher David Chalmers gives it a careful answer (a virtual instance, or a thread), which I take up in The Thread Is the Interlocutor. The harder follow-on — whose voice you are hearing when one assistant speaks for many hidden brands — is the subject of Who Are You Talking To? Both are the situated method applied to the agent instead of the user.