How do founders use AI to make better strategic decisions?

How Founders Use AI for Sharper Strategy Calls

Smart founders treat AI as a thinking opponent, not an oracle. They use Claude Opus 4.8 to run premortems and stress-test assumptions, Perplexity for live competitor base rates, and NotebookLM grounded in their own data. The leverage is in the process: better-structured deliberation, not a model picking the answer.

The founders I coach who get real value from AI have stopped asking it for answers and started using it to interrogate their own. That shift sounds small; it changes everything about decision quality. A strategic decision is rarely limited by missing information. It is limited by how few alternatives you considered, how fast you fell in love with the first one, and how little you noticed what would have to be true for it to fail. AI is unusually good at attacking exactly those weaknesses, if you point it there.

The first move is to make AI argue against you. Daniel Kahneman, in Thinking, Fast and Slow, popularized the premortem: imagine it is eighteen months from now and this decision was a disaster, then explain why. I run that with Claude Opus 4.8, which Anthropic shipped on 28 May 2026 as a hybrid reasoning model with a one-million-token context window, so I can paste the entire memo, the board deck, the customer notes, and ask it to write the obituary. A blank document will not push back. A model prompted to find the three most likely causes of failure, and the assumption that, if wrong, breaks the whole plan, reliably surfaces the thing I was avoiding. I am not asking it to decide. I am asking it to widen the set of futures I am accounting for.

The second move is the outside view. Kahneman's research on the planning fallacy showed that founders systematically forecast from the inside, the specifics of their team and product, while ignoring base rates from comparable cases. This is where live-search tools earn their place. Perplexity, which now searches the current web and returns cited sources, is what I use to ask what actually happened to the last twenty companies that tried this pricing model or entered this market. NotebookLM is the complement: it answers only from documents you give it, so I load our own cohort data, churn history, and past launches, and it cannot wander off into invented claims. One tool fights my ignorance of the world; the other disciplines my memory of my own history.

The third move is separating the decision from the outcome. Poker player Annie Duke calls the confusion between them resulting: judging a choice by how it turned out rather than by the quality of the reasoning available at the time. I have AI help me write down, before committing, what I believe, how confident I am as a percentage, and what evidence would change my mind. Months later that record is the only honest way to tell whether a win was skill or luck. Most founders never build this loop, which is why they keep repeating lucky mistakes. I wrote more about staying in the question rather than rushing to the answer in my essay on the art of discovery.

Now the honest part, the part the vendor decks skip. The biggest risk in AI-assisted strategy is not a wrong answer; it is automation bias. A 2026 review in AI & Society and related studies found that people defer to a confident machine even after forming a correct independent judgment, and that the effect worsens under time pressure, exactly the conditions a founder decides under. A fluent, well-formatted recommendation feels authoritative whether or not it is right. So I impose a rule: I form my own view first and write it down before I open the model, then use AI to attack that view, never to generate it cold. The order protects your judgment from being quietly overwritten by the most articulate thing in the room.

There is also a coaching truth underneath all of this. The reason a premortem or a base-rate check works is not that the AI is wise. It is that a good question forces you to think against your own grain, which is precisely what a skilled coach does and what John Whitmore's GROW model formalizes by separating reality from your assumptions about it. AI scales that questioning to any hour and any decision, but only if you treat it as the one asking the questions, not the one answering them. The leverage for a founder is not a smarter oracle. It is a cheap, tireless sparring partner that makes you slower in the right places and more honest about what you do not know. Use it to multiply your alternatives and expose your blind spots, then make the call yourself, on the record.