Field notes / working methods
The Persona Toolkit is a ranked reference of thirty role-based thinking lenses for working with AI: what each one is good at, where it fails, and which ones cluster for real work. It is not a product, and it is not mine to hand you finished. It is a worked example plus the method: you build your version through a feedback dialogue -- challenge, questions, sign-off -- shown at the end of this piece.
A persona is an instruction, not a feature. Tell an AI model to “act as a CFO; protect cash” and it reasons about runway, margin, and unit economics. Tell it to “act as a devil’s advocate” and it attacks your plan before the market does. Same model, different lens -- the way the same person gives you different answers depending on whether you ask them as your accountant or your friend.
The toolkit documents thirty of these lenses. Each entry covers how the persona thinks, when to reach for it, its pros, its cons, the single task it does best, and which other personas it clusters with for a full workflow. The point of ranking them is attention: you will use five of these weekly and twenty-five occasionally, so the five earn the top of the list.
You already know how to do this. You do it every time you ask a different colleague the same question and expect a different answer.
The spread is weighted, not even. Half the list is finance and strategy, because that is where decisions move money. A balanced encyclopedia would be more complete and less useful. The honest advice underneath the chart: re-weight it for your own seat. A recruiter’s version of this list should be heavy on people personas; an engineer’s on technical ones. The weighting is the method.
Category spread of the 30 personas. Source: toolkit summary sheet.
These are the ten highest-leverage personas in the toolkit, with the mechanism behind each. The skew is visible on purpose: this ranking was built for a finance-and-strategy seat. Your ten will differ -- the method for building your own is at the end of this piece.
A persona is a lens, not a licence. It cannot sign an audit, file a return, or replace a regulated professional opinion -- and the toolkit says so on every card where that matters. It also decays: a competitive-intelligence readout is a snapshot, and a pricing view is only as good as the demand data behind it. Advice that hides its limits is advertising. This piece is trying to be the other thing.
This is how the toolkit was actually derived. Four rules, then the honest way to reproduce it. Nothing withheld -- including the part most write-ups leave out.
One correction to the usual pattern: this toolkit was not produced by a single magic prompt, and pretending otherwise would be the kind of advice this piece exists to avoid. The real derivation was a working dialogue: a rough ask, a challenged premise, three clarifying questions answered, and structural pushback accepted -- pairs became clusters, an even spread became a weighted one, alphabetical order became a leverage ranking. The output was curated through that exchange, not extracted by an incantation. If you want to reproduce it, start the dialogue rather than hunting the prompt:
The starter. Copy it, fill the bracket, run it.
Run that and you will not get my toolkit. You will get yours -- which is the point. The prompt is not the method. The dialogue is.