Wharton School professor Ethan Mollick, who studies artificial intelligence, points out that when people say AI will easily replace workers, he thinks of the ethnography of copier repair technicians in the 1990s: the work was improvised, informal, with poor documentation, yet technology ultimately eliminated this role.

He refers to "Talking About Machines," published in 1996. Author Julian E. Orr followed Xerox copier repair technicians, concluding that getting machines running again often relied not on official manuals, but on "war stories" shared among technicians, trial and error on-site, and tacit knowledge circulating only within peer circles. Later discussions on communities of practice and organizational learning frequently cited this book.

In the book, technician Tom is fixing a paper feed issue. The parts are normal, and the paper isn't loaded incorrectly; he determines that cheap paper, a long-term dirty machine, and low usage have caused the paper to become damp. He prints one sheet at a time, causing the latch to engage and disengage repeatedly, deliberately inducing the fault. Continuous copying doesn't work because the latch will remain engaged. He also notes the billing meter reading; this machine has extremely low usage, and a smaller machine would be more suitable, but the machine is too old, and the rental cost is almost negligible, so the customer has no motivation to replace it.

Another page describes technician Bob making courtesy visits on days with few repair requests. The account is important, and he and team manager John were supposed to take the client manager out for dinner the previous day, but both forgot. The three visit Integrated Circuit Corporation to inspect each machine, check usage and error logs since the last visit, and discuss the quirks of each machine and its placement. One machine in the print center has had an external function disabled to shift some work to older, harder-to-use machines designed for high volume.

Mollick's point is that the role largely disappeared not because a system learned Tom's tricks, but because machines became more reliable and cheaper, making those tricks unnecessary. In responses, some applied the same logic to legacy codebases: the real logic resides in the minds of senior developers, and roles disappear because the copiers changed, not because tacit knowledge was captured. Others added that technology is increasingly designed to be disposable rather than repairable, and long-term tax policies encourage capital expenditure over labor costs.

The buyers are equipment manufacturers and software companies that build reliability into their products, while the sellers rely on improvisational repair labor. This is not an event-driven transaction where models read manuals to get to work; funding shifts from "sending someone to fix chaos" to "designing chaos out." Beneficiaries include low-failure equipment, remote diagnostics, and service contracts priced by reliability, while the burden falls on undocumented judgments, customer quirks, and skills circulating only among colleagues.

Mollick's new book, "Co-Existence," is set to be published on October 20. This post includes the book cover and two pages of technician field notes, discussing whether "unwritten knowledge remains the work itself" and whether intentionally disabled functions might be mistakenly repaired by agents, disrupting team processes.

Source: Public Information