Julian Orr's Xerox Technician Studies
Julian Orr's ethnographic research on Xerox field technicians, published as Talking About Machines: An Ethnography of a Modern Job (1996), documented how the most effective repair workers developed competence not by following procedures but by accumulating improvisational expertise through encounters with specific machines in specific conditions. Orr's technicians shared what he called 'war stories' — narrative accounts of difficult repairs that encoded tacit knowledge about how this specific model failed under these specific conditions in ways the service manuals did not describe. The work confirmed and extended Suchman's situated-action framework from the user side to the repair side, providing one of the richest empirical bodies of evidence for how technical competence actually develops.
The Enclosure of Repair Work — Contrarian ^ Opus
There is a parallel reading that begins not with what the technicians knew but with what Xerox was attempting to accomplish through knowledge capture. The war stories represented accumulated collective intelligence that existed outside management control — knowledge that created technician autonomy, increased their bargaining power, and made them structurally difficult to replace. The knowledge management systems were not naïve technical failures; they were instruments of enclosure, attempts to convert community-maintained expertise into corporate-owned assets that could be deployed without the community that produced them.
From this starting point, Orr's documentation becomes evidence of capital's persistent need to break craft knowledge — the same pattern that played out in textile mills, steel foundries, and every other domain where worker expertise created structural power. That the systems failed does not mean the enclosure attempt ended; it means management learned to pursue the same objective through different instruments. Contemporary AI tools marketed to 'assist' field technicians — providing diagnostic suggestions, documenting repairs, 'learning' from collective practice — continue the enclosure project under a different technical regime. The question is not whether situated knowledge can be extracted (it cannot) but whether the work can be deskilled enough that extraction becomes unnecessary, and whether technicians retain enough collective power to resist that transformation.
In the AI Story
Orr conducted his fieldwork as a corporate anthropologist at Xerox through the 1980s and early 1990s, riding along with service technicians as they diagnosed and repaired photocopiers in customer sites. The work required extended field observation — weeks and months spent with specific teams — and produced detailed ethnographic portraits of how technicians actually performed their jobs. What emerged was a systematic divergence between the engineering model of repair (diagnosis from error codes, procedure from manual) and the actual practice (improvisation from accumulated experience, diagnosis from listening and feeling).
The technicians' war stories were the key discovery. When a technician faced an unfamiliar problem, she rarely consulted the manual first. She consulted her memory and her colleagues — recalling similar situations, asking whether anyone had seen this specific symptom pattern, swapping narratives that encoded the particular ways specific machines failed in specific conditions. The stories were the medium through which situated knowledge circulated among technicians. The manual was a resource of last resort, consulted when the narrative tradition failed to provide a match.
Orr's work was foundational to the community of practice literature — Etienne Wenger and Jean Lave drew on it extensively — and to the broader development of situated learning theory. It provided empirical confirmation of what Suchman's framework predicted: that competent practice is improvisational, that expertise is tacit and domain-specific, and that the knowledge through which practitioners function cannot be extracted from the community that produced it.
The relevance to AI is direct. When Xerox deployed centralized knowledge management systems intended to capture the technicians' expertise in searchable databases, the systems failed — not because the technicians refused to participate but because the knowledge could not be extracted from the narrative practice that held it. The situated knowledge that enabled effective repair was not storable; it was a feature of the community's ongoing practice. Contemporary efforts to use AI to 'capture' expert knowledge encounter the same structural problem: the knowledge that matters is not a corpus of text but a relationship between practitioners and domains that continues to develop only as long as the practice continues.
Origin
Orr's fieldwork was conducted through the 1980s as part of a broader research program on workplace practice at Xerox PARC and the Institute for Research on Learning. His PhD dissertation and subsequent book (Talking About Machines, 1996) became foundational texts in the ethnography of work and in situated learning theory.
The research paralleled Suchman's PARC studies and influenced her framework extensively. Where Suchman focused on the user side of the human-machine encounter, Orr focused on the technician side; together the two bodies of work produced one of the most comprehensive ethnographic accounts of how people actually engage with computational technology.
Key Ideas
War stories carry expertise. The technicians' narrative tradition encoded situated knowledge that formal documentation could not capture.
Improvisation beats procedure. The most effective technicians were not the most procedure-faithful but the most responsive to specific circumstances.
Expertise is community-maintained. Individual technicians developed competence through participation in a practice community that circulated knowledge through narrative exchange.
Listening and feeling. Competent diagnosis required embodied engagement with specific machines — the tilted head, the hand on the frame — not abstract reasoning from error codes.
Capture resists automation. Attempts to extract the technicians' knowledge into searchable databases failed because the knowledge existed as a relationship, not as content.
Appears in the Orange Pill Cycle
Situated Knowledge Under Pressure — Arbitrator ^ Opus
The weight of evidence supports Orr's core claims about how repair expertise actually functions: war stories do carry knowledge that manuals miss (100%), improvisation does beat rigid procedure in complex diagnostic work (95%), and competence does develop through community participation rather than individual study (90%). The ethnographic documentation is too rich and too consistent with parallel findings across domains to dismiss. Where the contrarian reading gains force is in its analysis of institutional context and power relations (70%).
The knowledge management systems were both technical failures and enclosure attempts — these readings are not contradictory. Management genuinely believed capture was possible (the engineering model of knowledge as extractable content was widespread) AND the systems served institutional interests in reducing dependence on skilled workers. The AI question requires holding both: current tools cannot capture situated expertise (Orr is right), but they can restructure work such that less situated expertise is required (the contrarian pressure is real). The relevant variable is not technical capacity but institutional power — whether technician communities retain enough leverage to resist deskilling, whether service contracts allow time for improvisational diagnosis, whether profit margins permit the apprenticeship relationships through which war-story knowledge circulates.
The synthesis is not 'AI will fail' or 'AI will succeed' but rather: AI encounters the limits Orr documented when it attempts direct knowledge extraction, but it reshapes the terrain when it changes what counts as acceptable repair, what customers expect, what service contracts permit. The situated knowledge remains real; the institutional context that allows it to matter is what's under pressure.
Further reading
- Julian Orr, Talking About Machines: An Ethnography of a Modern Job (Cornell University Press, 1996)
- Jean Lave and Etienne Wenger, Situated Learning: Legitimate Peripheral Participation (Cambridge University Press, 1991)
- Lucy Suchman, Human-Machine Reconfigurations (Cambridge University Press, 2007)