CONCEPT

Training Never Ends

Janah's operational principle — learned through repeated Samasource failures — that training is not a state that is achieved but a process that is maintained, because standards, tools, and contexts evolve faster than any one-time training event can anticipate.

The principle emerged from a specific and instructive Samasource failure. In 2016, a Nairobi team trained for an autonomous-vehicle annotation contract performed well initially, then experienced quality erosion six months later when the client updated its taxonomy. The workers had not regressed; the target had moved. The training had been designed as a completion event; the work required training as continuous process. Janah formalized the lesson into an operational principle that extended across every Samasource contract: training is not an upfront investment to be amortized but an ongoing operational expense, consuming between twenty and thirty percent of productive capacity on a permanent basis. The cost was not a startup inefficiency. It was the price of keeping collective understanding current with evolving demands — the price that every organization serving a dynamic technology landscape must pay or accept the quality erosion that its absence produces.

In The You On AI Field Guide

Every technology transition in the history of work has demonstrated the principle, and every transition has produced institutions that initially resist it. The factory system of the nineteenth century treated training as a one-time event, with the predictable result that as machines evolved, the gap between what workers had been trained to do and what the work required grew until it produced quality failures, safety incidents, and the labor unrest that forced the development of ongoing training infrastructure.

The professions learned earlier. Medicine formalized continuing education in the mid-twentieth century, recognizing that a physician trained in 1960 who did not update her knowledge would be practicing obsolete medicine by 1970. Law followed. Engineering followed. Each profession developed institutional mechanisms — continuing-education requirements, professional-development programs, recertification cycles — that acknowledged the obvious truth that the world changes faster than any single training event can anticipate.

The digital-work economy has been slow to develop equivalent mechanisms, and the AI transition is accelerating the consequences of this slowness. AI capabilities change with each model update. Prompting strategies that work today may be mediocre in months. Architectural patterns generated reliably today may be superseded by patterns that do not yet exist. A developer trained on current tools is not trained for the tools she will use in six months.

At Samasource, continuous training took specific institutional forms. Quality reviewers functioned as embedded trainers, providing real-time feedback that combined assessment with instruction. Team leads held weekly calibration sessions. Mentorship relationships transmitted professional judgment that formal training could not fully capture. The infrastructure consumed substantial resources and produced the benefit that only becomes legible when it is withdrawn: the prevention of quality erosion that would have occurred in its absence.

Origin

The principle crystallized through operational experience rather than theoretical derivation. Samasource's early experiments with compressed training produced the specific failure patterns that revealed the need for continuous investment.

The autonomous-vehicle contract described in Give Work and in the Janah book's fifth chapter served as the clearest single case that informed the organizational practice thereafter.

Key Ideas

Every technology transition in the history of work has demonstrated the principle, and every transition has produced institutions that initially resist it

State vs process. Training is not a threshold crossed but an ongoing relationship between learner and evolving work, mediated by institutions designed to keep the relationship productive.

Twenty-to-thirty percent. Samasource estimated that continuous training consumed 20–30% of productive capacity as a permanent operational expense — a line item the organization refused to treat as optional.

Invisible benefit. The benefit of continuous training is counterfactual — the quality erosion prevented rather than the quality visible at any given moment — and therefore systematically undervalued by measurement regimes that focus on current output.

AI-era urgency. The pace of AI capability evolution makes continuous training more structurally necessary than it has ever been, and its absence more costly, at the precise moment when the institutional infrastructure to provide it is least developed for the populations that most need it.

In The You On AI Field Guide

Origin

Key Ideas

Related Entries

Further Reading