James G. March

In the [YOU] on AI Field Guide

The cycle's account of the AI adoption dynamic is, in structural terms, a description of March's exploitation trap operating at unprecedented scale and speed. When twenty engineers in Trivandrum achieve twenty-fold productivity gains in a single week, the learning system does not deliberate. It updates: this works, do more of this, allocate everything here. The updating is rational. The trap is that the update is also irreversible. The routines that encoded pre-AI competence begin to decay the moment they are displaced. The feedback loops that would signal the decay—the failed debugging session, the architectural error that only a practitioner with deep experience would detect—are eliminated by the same productivity gains that made the adoption obvious. The quarterly numbers remain excellent until they do not, and by the time they do not, the human capabilities that could address the underlying problem have atrophied beyond quick recovery.

March's framework explains why this dynamic is not a management failure but a structural feature of how learning systems operate. The garbage can model of organizational choice explains why AI adoption happens without strategic decision: solutions find problems through temporal coincidence, and AI is a solution of such generality that it attaches itself to every problem it encounters. The ratchet explains why adoption is irreversible: each local adaptation individually rational, collectively transforming the organization into something that can no longer function without the tool, without anyone having chosen that outcome. And the myopia of learning explains why the trap is invisible until it springs: exploitation returns are visible, measurable, and immediate; exploration returns are distant, uncertain, and unattributable. The learning system sees what happened last quarter. It does not see what will happen when the game changes.

The most productive application of March's framework to the AI moment is the concept of the technology of foolishness: the deliberate cultivation of organizational and individual practices that enable action without prior justification. AI is a technology of reason, and it makes reason overwhelmingly productive. The twenty-fold exploitation returns mean that every hour not directed at the exploitation machine carries an opportunity cost of twenty hours. In this environment, the technology of foolishness does not merely look wasteful. It looks irresponsible. March's argument—that the organizations which survive are not the most rational but the ones that managed to remain productively foolish in an environment that made foolishness expensive—is more urgent now than when he made it in 1971.

The cycle's prescription for leaders—protect exploratory time, build AI Practice frameworks, sequence workflows to preserve the development of human perceptual skills—is, in March's vocabulary, the deliberate construction of organizational slack: protected space within which the exploration that exploitation will never spontaneously fund can occur. The construction is structural, not cultural. Culture follows structure. And the structure must be built against the relentless gravitational pull of exploitation returns that make every alternative allocation of time and resources feel like a failure of management.

Origin

James March was born in Cleveland in 1928 and spent most of his academic career at Stanford, where he held appointments in management, political science, sociology, and education simultaneously—a disciplinary pluralism that was not incidental to his work but its method. Organizations, in March's view, could not be understood from within any single discipline; they were the intersection of all of them, and any framework that tidied them into a single theoretical language was missing what was most interesting. His collaborations across disciplines produced a body of work of unusual durability: the behavioral theory of the firm with Richard Cyert (1963), the garbage can model of organizational choice with Michael Cohen and Johan Olsen (1972), the exploration-exploitation framework (1991), and a late career series of papers on leadership, learning, and the nature of organizational wisdom.

March was a poet and fabulist as well as an organizational theorist—he published poetry throughout his academic career, and his fables about leaders and organizations drew on a literary sensibility that his more analytically rigorous colleagues sometimes found puzzling and his students found essential. The sensibility was not decorative. March believed that the deepest truths about organizations were not formalizable in equations; they required the kind of attention to ambiguity, irony, and the gap between intention and outcome that literary practice develops. His 1986 paper on the ambiguity and engineering of identity treated organizational decision-making as a problem in narrative, not optimization.

He died in 2018 at eighty-nine, having spent a career insisting that the most important things about organizations are the things that rational-planning frameworks cannot see: the way good outcomes emerge from bad processes, the way bad outcomes emerge from good processes, and the way the organizations most committed to rationality are often the ones most reliably trapped by it. His last published interview, in 2017, was characteristically direct: “The notion that magically, through learning, we will end up with an optimum set of rules, I think is fanciful.”

Key Ideas

Exploration and exploitation. The foundational distinction: exploitation is the refinement and extension of existing competencies, producing proximate, predictable, measurable returns; exploration is the search for new alternatives, producing distant, uncertain, frequently negative short-term returns. The two compete for the same resources, and the competition is rigged: exploitation wins because its returns are immediate and legible. Left unmanaged, organizations drift toward exploitation and away from exploration, not through bad strategy but through the accumulated weight of individually rational decisions, each favoring the near over the far. The framework is the primary diagnostic for what AI is doing to organizations.

The garbage can model. Organizations are not problem-solving machines. They are arenas in which four loosely coupled streams—problems, solutions, participants, and choice opportunities—flow independently and collide more or less at random. Solutions are looking for problems. Problems attach themselves to available choice opportunities. The people present when a decision is made are not necessarily the people best qualified to make it; they are the people who happened to be available. The garbage can explains why AI adoption happens without strategic decision: AI is a solution of extraordinary generality that finds problems everywhere, in every coincidence of available tool and present problem and willing participant.

The myopia of learning. Learning systems are structurally biased toward the near, the certain, and the measurable through three mechanisms: temporal (immediate returns outcompete distant ones), spatial (local improvements outcompete distant possibilities), and failure-averse (successes reinforce current strategies while failures discourage alternatives). The myopia is rational at each individual decision point and systematically damaging at the system level over time.

The technology of foolishness. March's 1971 argument that organizations need the capacity to act without reasons: to experiment without knowing what the experiment is testing, to play without knowing what the play is for. The technology of foolishness is the necessary complement to the technology of reason—and the complement that AI's overwhelming exploitation returns make most expensive to maintain and most necessary to protect.