AI strategy prompts ("the model provider just launched our core feature", "where does value accrue in the AI stack?", "should we build our own model?") test whether you can reason about competition when the underlying technology improves every quarter and is available to everyone. An answer that lists AI features misses the point. An answer that makes a bet, says why it lasts and says when to stop it, scores.
The worked example: the model provider we build on just launched our core use case, free, inside its own assistant. It is a free practice question, the provider launched our core feature.
Read diagram description
Seven moves for an AI strategy answer: who captures the value, what the provider ships next, which moat survives a model swap, whether margin survives usage, how users find you, build, buy or partner, and a bet with kill criteria.
1. Who captures the value
Start by placing the product in the stack: chips, cloud, models or applications. Applications keep value when they own something the layers below cannot provide by default. For the example: "Our product is an application on someone else's model, so our value has to come from something above the model." Where does value accrue in the AI stack works through the stack.
2. What the provider ships next
Model providers ship generic use cases inside their own assistants. Segment your users by how much of your value the provider's feature replaces. "Casual users who only needed the generic use case are at risk; teams who rely on our integrations, review workflow and admin controls are not."
3. Which moat survives a model swap
Ask what remains if every competitor gets the same model tomorrow. Proprietary data that improves the product with use, deep workflow integration, distribution and trust survive; a better prompt does not. The moat ladder ranks them, and data flywheels explains when data really compounds.
4. Whether margin survives usage
AI products pay for every use. Figma's FY2025 annual report showed gross margin falling from 88% to 82%, with $49.1 million of AI and usage hosting costs in cost of revenue, and Figma introduced AI credit limits from March 2026. A strategy that wins users but loses money on heavy ones is not a strategy. Wrapper gross margin and pricing AI products give the arithmetic.
5. How users find you
Discovery is moving into AI answers and assistants. Say where your users will meet the product and what you will own there: answer-engine visibility for search, distribution inside AI assistants for plugins and connectors.
6. Build, buy or partner
Decide per layer. Buy the model unless a narrow, high-volume task justifies your own; build the parts that are your moat; partner for distribution. Build, buy or fine-tune and platform risk and lock-in cover the trade-offs, including keeping a second model provider qualified.
7. The bet and its kill criteria
End with a funded bet and the evidence that would stop it. "Over the next two quarters, put the team on the review workflow and admin controls that enterprise teams use, measure retention in that segment against the casual segment, and stop investing in the generic use case if casual retention falls below our floor for two months." Kill criteria explains how to set the floor in advance.
What good sounds like
A strong strategy answer is specific about who is at risk and who is not, honest about margin, and ends in a bet that can be proven wrong. It does not end with "we will add more AI".
Questions about this topic
A decision with a reason the competition cannot copy quickly. Name where value accrues, what your model provider is likely to absorb, which advantage survives a model swap, whether gross margin holds as usage grows, how users will find the product, and what bet you would make with criteria for stopping it.