Steering AI to a Non-Obvious Strategy
AI gives predictable strategy when asked lazily — enumerate every input first, then make it argue back
- Difficulty
- Moderate
- Time to result
- ~days to results
- Steps
- 3
- Confidence
- 88%
A method for getting genuinely useful strategic thinking out of an LLM instead of the predictable answer your competitors would also get. The core move is to do the hard upfront work of listing every real-world input that must be considered, steer the model to weigh all of them, and force a critical back-and-forth conversation. Use when using AI as a strategy thinking-partner.
Origin
Mosseri says AI 'should' be great at strategy given it can hold the whole market, but in practice it isn't unless you steer it hard — not toward an answer, but toward the constraints. He lists the inputs a real strategy must weigh.
Core principles
- 01A lazy strategy prompt yields a predictable answer the competition already expects
- 02Steer toward constraints and inputs, not toward a predetermined answer
- 03Strategy requires weighing many inputs at once; make the model consider all of them
- 04Pick a model willing to push back, and explicitly tell it to be critical
How to run it
- 1
Enumerate every input
Before prompting, think long and hard about all the inputs a strategy must weigh: state of the technology, personnel and what motivates them, talent you can attract, competitive landscape, regulatory/compliance landscape, and the brand's identity and reason to exist.
Pro tip An idea 'on the bubble' can itself attract the best talent — so talent attraction is a strategic input, not just an output.
- 2
Steer the model to weigh them all
Prompt so the AI is explicitly considering each of those inputs, not optimizing for a clean single answer.
Watch out Ask lazily and you'll get something predictable that the competition would expect you to do.
- 3
Make it a critical conversation
Run it as an iterative back-and-forth, and tell the model to be critical. Choose a model with a 'vibe' willing to disagree.
Pro tip Prefer a model that behaves like 'a jerk' and pushes back over a people-pleaser that says 'you're so right.'
Watch out Different models differ a lot in willingness to be pushed back on.
In the wild
Ask an AI for a strategy lazily and you get something predictable a competitor would expect; put in the work to name and steer every input and it becomes clarifying and genuinely helpful.
→ A more effective, less obvious strategy emerges from the upfront input work.
Common mistakes
Steering toward an answer instead of the constraints
Pushing the model to a conclusion defeats the point; steer it toward the inputs so it reasons within real bounds.
Is it for you?
Best for
Operators using an LLM as a strategy sparring partner on a real competitive decision
Not ideal for
Quick tactical questions where a predictable answer is fine
From the transcript
“I think if you want a really more effective one, you need to think long and hard about what are all the different inputs that…”
“Um, particularly if you tell it to be critical.”
“I recommend picking one that likes pushing back.”
From the episode
Adam Mosseri: AI is a tailwind for authenticity