How can AI help with product strategy at a startup?

AI changes strategy less than it changed the sprint, and more than teams expect. The strawman is now generated from the evidence in minutes. Every document leaves the room from one operating brief. The facilitator becomes a context engineer, maintaining the brief that produces the documents rather than writing them. What does not change: the mission still comes from conviction, the vote still has to lose to the evidence, and someone still has to say what the company will not do.

What is an operating brief and how does it generate strategy documents?

AI changes strategy less than it changed the sprint, and more than teams expect. The strawman is now generated from the evidence in minutes. Every document leaves the room from one operating brief. The facilitator becomes a context engineer, maintaining the brief that produces the documents rather than writing them. What does not change: the mission still comes from conviction, the vote still has to lose to the evidence, and someone still has to say what the company will not do.

What parts of strategy can AI not do?

AI changes strategy less than it changed the sprint, and more than teams expect. The strawman is now generated from the evidence in minutes. Every document leaves the room from one operating brief. The facilitator becomes a context engineer, maintaining the brief that produces the documents rather than writing them. What does not change: the mission still comes from conviction, the vote still has to lose to the evidence, and someone still has to say what the company will not do.

With AI · Product Strategy · Product Thinking

§The third whitepaper ended by saying AI had made the middle of the nearly free and left the ends alone. Strategy is one of the ends. Less changed here. What changed is worth being precise about.

§ 15.1The strawman is generated#

§Chapter 10 asked the facilitator to arrive with filled from the evidence, every row cited. In 2022 that was a long evening: reading the tension map, the synthesis pages and the two-act memo, drafting six levels, checking each against its source. It is now a prompt with the evidence attached, and the first draft arrives in minutes with citations to the interview codes.

§The draft is not the strawman. It is the raw material for one. A model given a will produce a mission that is fluent and a strategy row that includes, faithfully, every good thing the evidence supports, which is to say it will produce Rumelt's bad strategy, a list of goals with adjectives. The facilitator's job moves from drafting to cutting: which of these bets are in sequence, which are dependencies of others, and above all, what goes in the "we will not" row, which the model will leave nearly empty because nothing in the evidence tells it to exclude.

§ 15.2Every document from one brief#

§In the hospitality engagement, every document that left the room, the first-version definition, the roadmap, the technical path recommendation, the brand handoff, was generated from a single operating brief that the facilitator maintained and a model read. The brief held the cast, the calendar, the ladder as corrected, the open items with owners, the voice, and the rules. When the team amended the pilot restaurant, the brief changed first and the four documents were regenerated consistent with each other.

§This is the one genuinely new thing. Before, four documents written by hand on four evenings drifted: the roadmap said month three, the handoff said month four, nobody noticed until the partner asked. Now the documents cannot drift from each other because they have one source, and the risk moves to the source. If the brief is wrong, everything is wrong, consistently and confidently. The facilitator's precision in maintaining the brief is now the quality of the whole output.

§ 15.3The facilitator as context engineer#

§The seventh whitepaper develops this at length; here is the strategy-specific version. The old facilitator produced documents. The new one maintains the context that produces them: the brief is updated before any document is touched, decisions are entered as they are made so that every later document already knows them, and the smallest true set of things the next document needs is what gets uploaded, never everything.

§Ethan Mollick's frame for working with models applies exactly: give the tireless collaborator the tireless work, drafting, cross-referencing, regenerating for consistency, and keep the judgment. In strategy the judgment is concentrated in three places, and none of them moved.

§ 15.4What did not change#

§The mission still comes from conviction. A model can propose ten missions from the and all ten will be plausible. None of them will be the founders'. The top of the ladder is the one level not filled from the field, and it is therefore the one level a model has nothing to fill it from.

§The vote still has to lose. The rules for five people, write before you speak, vote before you debate, one tie-breaker, close with what you have, are about human beings in a room letting evidence overrule their preference. No tool does that. In the field sprint a room of ten let twelve strangers put payments in month twelve because one person said it aloud and the rest committed. The prototype took two days to generate; the sentence took one person.

§Someone still has to say what the company will not do. This is the whole of strategy and it is the one thing generation is worst at, because generation is additive. The "we will not" row is written by a person who is willing to be wrong about an exclusion, in front of colleagues who wanted the excluded thing. There is no prompt for that.

§ 15.5What comes out#

§A ladder, cited, corrected, cut, and confirmed by the team in writing, with the first unit of work at the bottom and the reason it exists at the top. The fifth whitepaper asks whether the thing at the bottom closes as a business. The sixth owns the experiments that move the goals once it exists. This one asked what the company will not do, and the answer is on the page.