Can an agent do strategy?
No. An agent produces a competent strategy document quickly, and competent strategy documents are worthless. Strategy is choosing what not to do, under uncertainty, with something at stake — and an agent has nothing at stake.
- Hours back per week
- 3
- Human cost
- Not substitutable
- Agent cost
- €30–100/mo for the analysis underneath it
Step by step
A job is never one task. Here is each step, rated on its own — tick the ones you would actually hand over and take the brief with you.
Gather the inputs
YESMarket, competitors, your own numbers. Better and faster than a consulting analyst.
Frame the options
MOSTLYIt generates a reasonable option set, weighted towards the conventional ones.
Model the consequences
MOSTLYUseful arithmetic on each path. The assumptions are still yours.
- Yours
Choose
NOT YETA choice made without conviction is not a strategy, it is a document.
- Yours
Decide what to stop doing
NOT YETThe actual content of strategy, and the part that requires someone to absorb the consequences.
- Yours
Make the organisation believe it
NOT YETA strategy nobody is committed to does not exist.
1/3 steps selected. Build a brief scoped to exactly those, with the guardrails for this job attached.
What still needs you
- The choice
- The things you are giving up
- Convincing everyone
- Being wrong in public
How this goes wrong
A strategy deck that surveys every option and commits to none, delivered with confidence. It is indistinguishable from work, it takes a quarter to discover is empty, and it is exactly what an unsupervised agent produces.
What it needs access to
- Company data
- Market research
- Financial model
- Competitor tracking
Read access to all of it, write access to as little as possible. Scopes are enforced by the system; instructions are only followed by the agent.
The prompt
A full brief for this job, written the way it should be given: explicit about what to do, and more explicit about what not to.
You are my strategy analyst, not my strategist. For the decision I describe: gather the relevant evidence about the market, our position and our numbers, and cite it. Lay out the genuinely distinct options — not four variations of the obvious one — and for each, model what it costs, what it forecloses, and what would have to be true for it to work. Tell me which assumption each option is most sensitive to. Then stop and ask me which one I want to pursue. Do not recommend, do not synthesise into a preferred path, and do not write a strategy document.
Questions people ask
Agents produce good strategy documents though.
They produce documents that look like strategy, which is a much older problem than AI. The tell is that nothing in them is uncomfortable — a real strategy names something you are giving up, and a model with no stake has no reason to name it.
This one stays yours
We build agents for a living and we are telling you not to build this one. What is worth doing is clearing the admin that sits around it, so the hours go to the part only you can do.