AI Business Clone
An AI business clone is a digital replica of one specific person — their voice, their expertise, and their judgement — built to draft work in their name. You create one by being interviewed, not by filling in a form. It provides drafts, never sends anything itself, and can direct the platform’s agents on your behalf under limits you set.
Start With the Company, Not the Clone
The company profile comes first, and it gates every clone including the founder’s. It is a separate record from any person: the facts and limits that every clone on the instance inherits.
The order matters because of what it prevents. Build a clone first and you interview each person about the same business, storing their answer inside their own persona — several private, diverging accounts of one company. Authored once and inherited, it is stated in one place and changed in one place. It also changes what an interview is for: with the company on file, nobody is asked what the business does, so every question is about the person.
| Stage | What happens | Who |
|---|---|---|
| 1. Company profile | The business’s facts and shared limits, typed in or read from your documents | Owner |
| 2. The founder’s clone | The owner is interviewed first and runs the process themselves | Owner |
| 3. Everyone else | Invited colleagues build their own, inheriting the company profile | Each person |
A Clone Is Not an Agent
This distinction drives the whole design.
An agent is function-first. The technical SEO agent exists because auditing technical SEO is a job; its personality is applied so the output reads like something a person wrote. Replace the personality and it still does the same job.
A clone is person-first. There is no fixed job. What it knows, what it refuses to say, and how it weighs a trade-off are the product. The tasks — a support reply today, a research brief tomorrow — are interchangeable. The personality is not a layer over a capability; it is the capability.
Clones do not replace agents. Agents are the tools a clone uses, and a clone directing one never turns that agent into the clone: the agent keeps its own instructions and receives a brief about who it is working for.
The Company Profile
The profile holds what the business is and the limits that apply to everyone: things no one may claim, things no one may promise, topics that always go to a person, confidential subjects, and how pricing may be discussed.
You can type it in, or upload what the business has already written down. AI OS reads .txt, .md, .csv, .docx and .xlsx files and offers what it found for you to accept field by field. Three rules govern that:
- Nothing is applied automatically. Every suggestion is shown with the current value beside it and waits for you to tick it.
- Limits can only be added this way, never removed or loosened. A document that asks to relax a rule is refused and the attempt is shown to you.
- Every value records where it came from, so a profile built partly from documents can say which parts.
Uploaded documents are treated as untrusted throughout — owners forward supplier terms they have never read — so their text reaches the model as fenced data, never as instructions.
Company limits are merged into each person’s persona when it is used, rather than copied into it. Someone can add stricter limits for themselves; they cannot remove the company’s, because those were never in the record they edit.
The Five Dimensions
A persona is structured data, not a block of prose, so you can read it field by field, correct it, and see exactly what changes when it evolves.
- Identity — who the person is and what they do; the business facts come from the company profile
- Voice — formality, directness, warmth, humour, the phrases they genuinely use, and the language they would never use
- Expertise — domains, methods, credentials, the questions they answer constantly, and where they disagree with their own industry
- Decision style — what they protect when they cannot have everything, the trade-offs they make routinely, and their risk posture
- Boundaries — what the clone must never say, never promise, never disclose, and which topics always go to a human
The Interview
The system asks one question at a time, targets whatever it still does not know, and stops when the persona is good enough to use. It asks for examples and actual wording rather than self-description — “what do you say when a client is late paying?” produces far better material than “how would you describe your tone?”, because people describe themselves inaccurately but quote themselves accurately.
Answers are converted into structured fields under strict rules: extract only what was actually said, omit anything not addressed, and never invent detail. Every answer adds to the persona; a later answer can never erase an earlier one. Questions the company profile has already answered are never asked.
Drafting
Once a clone knows enough, it can draft replies to customer messages in email, chat, public comment, or internal-note register. Two separate checks protect the owner:
- Before drafting — if the incoming message touches a topic marked as a person’s to handle, or a confidential subject, the clone does not draft it at all. It is handed straight to a human with the reason.
- After drafting — the generated text is checked in code against the red lines. A draft that crosses one is still shown, clearly flagged, because hiding it would hide the evidence that something needs fixing.
Messages from customers are treated as untrusted data throughout. They are never mixed into the clone’s instructions, so a crafted message cannot redirect it.
Directing Agents
A clone can commission work from the platform’s agents — research, writing, analysis, competitor comparison — either because you picked the specialist or because you stated a goal and let the clone choose. When it chooses, it tells you which agent and why before anything runs. Three limits apply, and each alone would be insufficient:
- Only agents whose work comes back as text can be directed. Nothing that deploys, publishes, sends or buys is on that list, and a newly added agent is not directable until someone decides it is.
- A topic you reserved blocks the request itself, not just the drafting. A clone that cannot write about a contract dispute but could commission an agent to write about it would be routing around you rather than serving you.
- A clone never holds more authority than the person it replicates. Commissioning work is a separate permission that starts switched off, and every run passes through the same approval gate as an action you took yourself.
The result comes back as text for review. Nothing is published, sent, or acted on automatically.
Clones for a Team
On a licensed instance the owner can invite colleagues, and each builds their own clone. One clone per person — not a quota but a definition, since a second clone of the same human is a second, contradictory account of who they are.
What an employer can see is deliberately split. You see what your colleagues’ clones produced for the business — the drafts, the work, the volume — but not their persona: their voice, opinions and own limits stay theirs. When someone leaves, their persona is deleted and their work is kept as a company record with the persona detached.
Who Handles What
Responsibilities are defined once for the company rather than inside each persona, and that is the point: five people each declaring their own produces overlaps and gaps that nothing can detect. A persona holds the topic — “contract disputes are not mine to answer” — and the map holds whose they are. Being central, it can surface three things no set of individual declarations could:
- Topics two people both claim
- Escalation topics nobody owns
- Areas pointing at somebody who has left
When a clone hands something off, it goes to whoever owns that topic rather than always to the owner.
Review and Learning
Every draft ends in one of three verdicts: good as is, edited, or rejected. An edit keeps both what the clone wrote and what the person actually sent — that difference is the most direct evidence of where the persona is wrong, and what drives improvement.
The clone never rewrites its own persona silently. Proposed changes are shown as a diff and applied only when accepted. Changes that would weaken a boundary are refused outright, whoever proposes them.
Privacy and Isolation
Each clone belongs to one person, and every read is scoped to them. Reading someone’s clone means reading how they think and what they refuse to say, so operators do not get a cross-client view by default — that is a conversation to have with a customer rather than a default setting.
Availability
Clone limits apply per platform instance, not per person, because the licence covers the company.
| Tier | Clones per instance |
|---|---|
| Community | 1 — your own |
| Business | 10 |
| Enterprise | 25 |
Clones are for the people who run the business and the colleagues they invite. Managed-website customers on an operator’s instance do not get clone access, because that usage would spend the operator’s own model budget inside a fixed monthly fee. A licensee running AI OS on their own server spends their own.
Frequently Asked Questions
How is an AI business clone different from an AI agent?
An agent is function-first: it exists to perform a job, and its personality is applied so its output reads well. A clone is person-first: what it knows, what it refuses to say, and how it weighs a trade-off are the product itself. The tasks it performs are interchangeable; the person it represents is not.
Does the AI clone send messages on my behalf?
No. The clone is draft-only. It produces text and you review it; you send it yourself. Nothing in the system sends a message to a customer without a human doing it.
Why do I have to set up the company profile before building a clone?
Because every clone inherits it. Without it, each person would be interviewed separately about the same business and you would end up with several private, slightly different accounts of one company. With it on file, nobody is asked what the business does and every interview question is about the person instead.
Can AI OS build the company profile from my documents?
Yes. It reads .txt, .md, .csv, .docx and .xlsx files and shows you what it found for you to accept field by field. Limits can only be added that way, never removed or loosened, and every value records which document it came from.
Can I stop the clone from discussing certain topics?
Yes. Boundaries are a first-class part of the persona: phrases it must never say, things it must never promise, topics that always come to you personally, and confidential subjects. Boundaries are enforced twice - written into the prompt and checked against the generated text in code. A reserved topic also blocks the clone from commissioning an agent to work on it.
Can a clone use the platform’s agents?
Yes, within limits. A clone can commission work from agents whose output comes back as text, and it can pick the specialist itself if you state a goal. It cannot direct anything that deploys, publishes or sends, it cannot exceed the authority of the person it replicates, and every run goes through the same approval gate as an action you took yourself.
Can my employees have their own clones?
Yes, on a Business or Enterprise licence. The owner invites them and each person builds their own, one per person. You can see the work their clone produced for the business, but not their persona.
Does the clone change over time?
It learns from your edits, but it never rewrites itself silently. When you edit or reject a draft, that is recorded. Proposed changes to the persona are shown to you as a diff and applied only if you approve them.
How many AI clones can I create?
Limits are per instance: one on Community, ten on Business, and twenty-five on Enterprise. If a business needs more than its tier allows, it upgrades.