Strategy
Build Once, Deploy Many: The Case for Separating Identity from Interface
The old way: one persona per audience
Most attempts at an AI version of an expert treat each use as its own project. You want something on your website for prospects, so you write a prompt and paste in your bio. You want something for the team, so you write another. A third for the clients on retainer.
Each one starts from scratch. Copy the bio, paste the instructions, adjust the tone, test it, fix what sounds wrong.
The maintenance is the smaller problem. The bigger one is drift. You change your mind about how you handle late payers, and you update the team assistant. The client one still gives the old line. Three assistants now hold three slightly different versions of your judgement, and nobody, including you, knows which is current.
The alternative: one model of you, several twins
Imora separates the two things that approach tangles together: what you think, and who you are talking to.
In Imora's terms, an Imprint is the model of you; twins are versions of it for different audiences. The Imprint holds your reasoning graph: the evidence from your answers, corrections and documents, distilled into principles with the conditions they apply under. A twin adds only an audience, a focus and a tone.
So a twin is cheap to make and hard to get wrong. You are not describing yourself again. You are saying who this one is for.
What this looks like in practice
Take the founder of a twelve-person agency. This is a hypothetical, but a typical one. From one graph she might run three twins:
- A team twin that account managers check before replying to a client. "They want two extra landing pages inside the current fee. How would she handle it?"
- A client twin on a shared link, for retainer clients with routine questions between calls.
- A new-starter twin that explains how she makes hiring, pricing and quality calls, with more patience and more background.
Ask all three about scope creep and the position is the same: name it early, price the change, never absorb it silently, make an exception once for a client in their first month. The team twin says it bluntly. The client twin says it courteously. The new-starter twin explains why.
The reasoning does not vary by audience. The presentation does. And a twin's configuration never limits what you have a view on: a twin focused on pricing will still answer a hiring question from your graph if you have recorded a view.
Teach one, and all of them know
Because every twin reasons from the same graph, a correction made anywhere applies everywhere.
Suppose the team twin tells an account manager to absorb a small change request. You read the reply and disagree. You mark it "Not what I'd say" and write what you would have said. That correction becomes evidence, and it outranks everything else in the graph. The client twin and the new-starter twin answer from it the next time the subject comes up.
It is worth being exact about what does not happen. Only you teach your twins. Nothing a visitor says changes how they think. A client can argue with your client twin for an hour and your graph is untouched.
What visitors do contribute is questions. When a shared twin has no recorded view on something, it says so, offers an approach clearly framed as an approach and not your position, and puts the question in your inbox. You answer it once, and every twin has it.
The case for small firms
For one person the arithmetic is simple: teach one graph, keep it current in one place.
For a firm it matters more, because the alternative is worse. A consultancy with five senior people has five sets of judgement, and they are not interchangeable. Blend them into one "company assistant" and you get an average nobody would sign. Give each person their own graph, and every answer is traceable to one person's reasoning. A junior asking about a pricing objection knows whose call they are getting.
On Imora, Pro covers 5 twins and Business covers 10, with team management. The plans are on the pricing page.
Why separation helps quality
When the model of you stands apart from any one twin, you can check it on its own terms.
Mirror Score tells you how much your twin has to go on: how deep and broad your evidence is. It is a completeness measure, not an accuracy score. Accuracy is checked separately. We hide some of your real answers from your twin, then check which version gets closest to what you actually said.
Short calibration sessions, about five minutes each, add new scenarios aimed at the gaps. You are maintaining one graph, not tuning several assistants. How the graph is built is covered in the reasoning graph post.
A question to ask any tool
If you are weighing up ways to put your judgement in front of more people, ask one thing: when I change my mind, how many places do I have to change it?
If the answer is "every prompt you have written", the approach will not survive a busy quarter. If the answer is "one", it might. Our comparison page sets out where Imora fits and where a custom GPT is enough.
That is all "build once, deploy many" means. One record of how you decide, several audiences, and corrections that reach all of them.