Entrepreneurship theory has long centered on the human entrepreneur’s ability to act under true uncertainty, where judgment is paramount (McMullen & Shepherd, 2006; Alvarez & Barney, 2007). The rise of Artificial Intelligence (AI) directly challenges this paradigm, by enabling a new type of venture whose logic is predicated on “unbundling prediction from other aspects of decision making (e.g., judgment, action)” (Shepherd & Majchrzak, 2022: 2). This ‘unbundling’ poses a theoretical puzzle: When incumbent ventures whose strength resides the traditional, bundled synthesis of judgment and action compete against AI-enabled rivals, what are the specific mechanisms that govern their rivalry and determine its outcome? In answer to this question, this paper develops the theory of ‘activated disruption’. We argue that the disruptive potential of an AI-enabled venture is not constant but is often latent, its effectiveness held in check by the deep contextual capabilities of incumbents. However, an exogenous shock may unleash this latent potential by creating environmental conditions that neutralize an incumbent’s tacit, pattern-based knowledge while favoring the AI-challenger’s architectural capacity for predictive re-optimization. To develop this theory, we analyze a critical case that embodies the puzzle: the decade-long rivalry between Mumbai’s Dabbawalas, a community-based enterprise (CBE) whose metis-driven system exemplifies human pattern mastery and the AI-platforms that challenged them. Their story allows us to induce a model explaining how a pattern-destroying exogenous shock ‘activated’ the platforms’ latent threat. Our core contribution is this contingent process model, which specifies when and why the competition between these rival entrepreneurial logics becomes decisively ‘disruptive’.