Most AI founders are technical, so they instinctively hunt for their edge upstream, in the model: make it faster, fine-tune it harder, squeeze out another point of accuracy. Niraj Dawar's downstream strategy argues this is exactly the worst place to plant your flag, and in AI the argument is sharper than in almost any industry he was writing about.
Upstream advantage in AI is structurally fragile, because the upstream is owned by a handful of foundation providers whose capabilities improve on a schedule you don't set and who can vaporize your fine-tuning edge with a single release. Betting your moat on being marginally better at the model layer is betting against the people who own the model layer, and they have more compute, capital, and talent than you will ever assemble. Every increment of upstream advantage you build is one Tuesday release away from being erased.
Downstream is where the durable margin lives, doubly so in AI. Downstream advantage is about cutting the customer's total cost and risk of adopting and running your product: the implementation friction, the integration effort, the trust barrier, the risk of the thing melting down in production. These are problems the foundation providers have no interest in solving for your specific vertical, because they are building horizontal infrastructure, not vertical solutions. The messy, industry-specific, trust-dependent work of making the model genuinely usable and safe in one particular context is precisely the ground the giants cede without a fight.
So reorient. Spend disproportionately on the customer's side of the transaction rather than the model's. Not "how do I make the AI better" but "how do I make adopting this lower-risk, lower-effort, and higher-trust for this specific customer than any alternative." That work, the onboarding, the guardrails, the integration, the explainability, the reliability tuned to one industry's ugliest edge cases, compounds, is specific to you, and bores the foundation providers to tears. Stop competing on the axis where you are structurally weak. Start competing on the axis the giants won't follow you onto. Downstream isn't where the glory is. It's where the margin is, and the margin outlasts the glory.