The preceding nineteen modules wove the AI transition through frameworks that predate it, on the conviction that classical strategy is not made obsolete by AI but is made more demanding by it. This final module confronts AI strategy directly, because the AI transition introduces a structural condition that no prior framework fully anticipated: the core capability on which an AI product depends is, for most builders, rented from a small number of providers and available to every competitor on identical terms. When the source of a product’s intelligence is a commodity, the question of where durable advantage resides becomes the central strategic question of the era, and answering it well is what separates the products that will compound from the many that will be competed to zero. This module develops that answer through four lenses, namely the distinction between AI-native and AI-enabled products, the strategic significance of agentic systems, the commoditization risk that defines the competitive environment, and the sources of defensibility that survive it.
AI-Native Versus AI-Enabled: A Distinction of Architecture, Not Marketing
The first distinction a product leader must reason about precisely, because the market uses it loosely, is between AI-enabled and AI-native products. An AI-enabled product is one to which AI has been added as a feature while the product’s fundamental architecture and value proposition would survive the removal of that feature, whereas an AI-native product is one whose value proposition collapses entirely without AI because the intelligence is constitutive of what the product is rather than additive to it (Andreessen Horowitz, 2024). The test is counterfactual and clarifying: if the product would function essentially as before with the AI removed, it is AI-enabled, and if its core reason for existing disappears, it is AI-native. The strategic significance of the distinction is that AI-enabled features, being additive, are the most easily replicated and therefore the least defensible, since a competitor can add the same feature to its own product, while AI-native products, being architecturally reorganized around intelligence, can pursue value propositions and workflows that incumbents bolting AI onto legacy architectures cannot match.
The deeper implication, which connects to the blue ocean and disruption logic of earlier modules, is that the most consequential AI opportunities are typically AI-native reconceptions of a category rather than AI-enabled enhancements of an existing product, because the native architecture can eliminate the assumptions the legacy product was built around. A product leader who frames the AI question as how to add AI features to the current product is, by the structure of the question, confined to the less defensible position, whereas one who asks what the product would be if rebuilt around intelligence from the foundation is positioned to find the durable opportunity.
Agentic Systems: From Assistance to Execution
The most significant capability shift within the AI transition is the move from systems that assist a human performing a task to systems that execute the task autonomously, which is the domain of agentic products. An agentic system is one that pursues a goal through a sequence of actions, invoking tools, making decisions, and adapting to results with limited human intervention, and the strategic consequence of this shift is that the unit of value moves from augmenting human productivity to performing the work itself. This relocates the competitive battleground in two directions at once, as the analysis of the current environment suggests: defensibility shifts downward toward the data models, permissions, workflow logic, and compliance infrastructure that autonomous execution requires, and upward toward the networks, proprietary data generation, and real-world execution that agents enable (Andreessen Horowitz, 2024).
For the product leader, the rise of agentic systems poses both the largest opportunity and the largest substitution threat reasoned about in Module 10, since a sufficiently capable general-purpose agent is the substitute that can absorb the job a specialized product performs. The strategic response is to determine whether the product is positioned to be the agent that executes the valuable workflow, in which case the opportunity is to own end-to-end execution, or whether it is positioned to be a capability the agent invokes, in which case the opportunity is to become an indispensable tool in the agent’s repertoire, accessible through the interfaces that autonomous systems use. The error is to assume the product can remain a human-operated interface in a domain where execution is migrating to agents, since that is the position most exposed to substitution.
Commoditization Risk: The Defining Condition of the Environment
The condition that organizes all AI strategy is that foundation model capability is commoditizing, which means that the raw intelligence a product can access is becoming abundant, broadly available, and decreasingly differentiated across providers and over time. The strategic consequence, stated bluntly by the analysis emerging from the venture community, is that using a capable model or adding an agent is easy to replicate and therefore is not a moat, however impressive the capability may be (Andreessen Horowitz, 2024). A product whose differentiation rests solely on access to model capability is building on a commodity, and as the commodity becomes more capable and more widely available, that differentiation erodes toward zero, which is the fate awaiting the large population of products that are thin wrappers over a rented model. The product leader must internalize that the model is the engine but not the moat, and that strategy in this environment is precisely the work of building advantage that does not reside in the commoditizing layer.
This condition also reframes the supplier-power analysis of Module 10 into a defining strategic risk, since dependence on a commoditizing capability supplied by concentrated providers means that the value a product adds must be located somewhere the provider does not capture, lest the provider, by improving its base offering, absorb the product’s reason for existing. The history of platform shifts is replete with applications rendered obsolete when the platform incorporated their function, and the AI equivalent is the product whose entire value is a capability the next model release will provide natively.
Where Defensibility Actually Resides
The constructive core of AI strategy is the identification of the layers where advantage survives commoditization, and the convergent finding across the most credible analyses is that defensibility lives not in the model but in the assets and relationships that surround it. The first such asset is the proprietary data feedback loop, which becomes more valuable in an AI world rather than less, because when the model is the commodity the scarce input becomes the proprietary data that specializes it, and a product whose usage generates data no competitor can observe builds a model advantage that widens with scale (V7 Labs, 2024). This is the data network effect of Module 14 expressed as the central AI moat, and it is the single most reliable source of durable advantage available to an AI product.
The second is workflow ownership and embedding, the depth to which a product is woven into the systems of record, the daily practices, and the surrounding tools of its customers, since a product that owns an end-to-end workflow and serves as the system of record accrues switching costs that a competitor renting the same model cannot overcome (Andreessen Horowitz, 2024). The third is the network effect among the product’s users, which an AI agent cannot replace, because the value that arises from the presence of other people, as in collaboration and communication products, is not a capability a model can synthesize away (Bain & Company, 2025). The fourth is trust and the surrounding evaluation, governance, and reliability infrastructure, which becomes a durable differentiator precisely because AI outputs are probabilistic and consequential, so that the product which has earned trust and built the mechanisms that sustain it holds an advantage that is slow to accumulate and therefore slow for competitors to replicate.
A sober note belongs here, since the enterprise reality tempers the enthusiasm: a substantial fraction of agentic projects are forecast to be abandoned without clear value, guardrails, and change management, and the majority of firms have struggled to scale AI beyond pilots (Bain & Company, 2025). The strategic implication is that the defensibility which matters is not demonstrated capability but realized, governed, trusted value in production, which is a far higher bar and one that the surrounding infrastructure, rather than the model, determines.
The synthesis that closes both this module and the series is that AI strategy is the discipline of building durable advantage in an environment where the core capability is rented and commoditizing, which returns the product leader to the classical frameworks rather than away from them. The agenda is to pursue AI-native reconceptions rather than AI-enabled additions, to position deliberately within the shift toward agentic execution rather than be substituted by it, to internalize that the model is an engine and never a moat, and to build defensibility in the layers that survive commoditization, namely proprietary data loops, workflow ownership and embedding, network effects, and trust infrastructure. The product leaders who will define the next decade are not those with access to the best model, since that access is becoming universal, but those who understand that in an age of rented intelligence, durable advantage is built in everything that surrounds the intelligence, which is exactly the strategic work the preceding nineteen modules were preparing them to do.
References
Andreessen Horowitz. (2024). Who owns the generative AI platform? https://a16z.com/who-owns-the-generative-ai-platform/
Bain & Company. (2025). Will agentic AI disrupt SaaS? https://www.bain.com/insights/will-agentic-ai-disrupt-saas-technology-report-2025/
V7 Labs. (2024). Are data moats dead in the age of AI? https://www.v7labs.com/blog/data-moats-a-guide
Agrawal, A., Gans, J., & Goldfarb, A. (2022). Power and prediction: The disruptive economics of artificial intelligence. Harvard Business Review Press.