The Bitter Lesson as Operational Constructivism: AI Agents, Organizational Chaos, and the Primacy of Performance

This paper applies the philosophical framework of Operational Constructivism (OC) to a central dilemmain the contemporary application of artificial intelligence to organizational work, as articulated by EthanMollick. We analyze the tension between the ”Garbage Can” model of organizations—which emphasizesmessy, undocumented processes—and Richard Sutton’s ”Bitter Lesson”—which posits that general-purposelearning on outcomes outperforms hand-crafted process encoding. We argue that this debate is not merelya technical or strategic choice but a live instantiation of OC’s core tenets. The Bitter Lesson’s approach,which trains AI agents on desired outputs rather than explicit processes, perfectly mirrors OC’s principle ofConstructive Validation, where an artifact’s performance is the ultimate arbiter of its underlying model’swarrant. The AI agent becomes the OC ”artifact,” and its successful navigation of organizational chaos is aform of earned realism. We conclude that OC provides the most coherent philosophical lens through whichto understand, and indeed champion, the Bitter Lesson’s potentially transformative impact on knowledge work.