# Human in command AI: three levels of autonomy

> Human in command AI means people set how far AI may go in each workflow. Three levels for a sales and service operation, and who should choose the level.

AI in operations · The Intellsys team · 18 November 2026 · 6 min read

Human in command AI means people decide how far AI may go in each workflow, and stay answerable for every outcome. A sales and service operation has three workable levels: the AI recommends and a person acts; the AI prepares an action and a person approves it; the AI acts within limits a person has set. Choosing the level for each workflow is a management decision, not a software setting.

## The short answer

- Human in command means the level of autonomy is chosen by people, workflow by workflow.
- Three levels: recommend, prepare for approval, act within limits.
- The level is set by whoever owns the workflow's outcome, not by IT or the vendor.
- A workflow earns a higher level with evidence, and can be moved back down.
- In Intellsys all three levels are available, and the business sets the mode for each process.

## Three levels of autonomy in a sales operation

Talk of AI at work often jumps between two pictures: a chat window that answers questions, and software that runs the business alone. Neither helps a sales head decide anything. Three levels in between do.

| Level | What the AI does | What a person does | An example in a dealer network |
| --- | --- | --- | --- |
| Recommend | Suggests what to do and why | Decides, and does it | Suggests which of today's leads a salesperson should call first |
| Prepare for approval | Drafts the action, ready to go | Reviews, edits, approves or rejects | Drafts a follow-up message to a customer, waiting for the salesperson to send it |
| Act within limits | Takes the action itself, inside rules a person set | Sets the limits, reviews what was done | Moves a lead untouched for two days to another salesperson in the same dealership |

The levels differ in one thing: when the person sees the action. At the first level, before it is even shaped. At the second, before it happens. At the third, after it happens, in a log.

## Who sets the level for each workflow

The level is not one setting for the whole company. A business might be happy for software to reorder a call queue and unwilling for it to touch a price. So each workflow needs an owner who chooses its level, and that owner should be whoever answers for the outcome.

- **Sales head** Lead follow-up, call order, reassignment inside a team.
- **Service head** Service reminders, booking slots, parts suggestions.
- **Finance head** Prices, discounts and anything that changes what a customer pays.
- **Marketing head** Which campaigns to change, and what may be said to customers.
- **Technology head** Which records and data the AI may read at all.
- **Chief executive** Anything that commits the business, and the rule for raising a level.

*Who should choose the level of autonomy for each kind of workflow: the person who answers for its outcome.*

Two things follow. The vendor never chooses the level for you. And the technology head decides what the AI may see, not what the sales team allows it to do. Who may see which record in the first place is its own question, covered in [role-based access control for dealer networks](https://intellsys.cloud/blog/role-based-access-for-dealer-networks).

## When a workflow is ready for a higher level

A workflow should move up a level only on evidence from the level below. The path is the same each time.

- **Recommend and watch** The AI suggests; people act; someone records how often they agree with it.
- **Agree the limits** The workflow's owner writes down what the action may and may not do.
- **Prepare and approve** The AI drafts; people approve each one; rejections are read, not ignored.
- **Review the record** The owner checks approvals, edits and rejections over a fixed period.
- **Act within limits** Only then does the AI act alone, inside the limits, with a log people read.

*How one workflow moves from recommend to act within limits. Each step needs evidence from the one before, and can be reversed.*

Before raising any workflow, the owner should be able to tick every line below.

- [ ] **A named owner** One person answers for this workflow's outcome.
- [ ] **Written limits** What the action may change, for whom, and how often.
- [ ] **A record of the level below** Weeks of suggestions or approvals, with how often people agreed.
- [ ] **A way to undo it** Every action can be reversed, and someone knows how.
- [ ] **A log people read** Actions are visible after the fact, and someone looks.
- [ ] **A way back down** The owner can return the workflow to a lower level the same day.

*Six conditions to meet before a workflow moves up a level of autonomy. If any line is missing, stay where you are.*

## What stays with people at every level

Some things do not move up, whatever the evidence. A customer relationship belongs to the person whose name is on the record. A price is a promise the business makes. An exception, such as a fleet customer asking for terms nobody has offered before, needs judgement about the customer, not only the data.

The test is simple. If a mistake would have to be explained to a customer, a dealer or a regulator by a person, a person should have seen the action before it happened. That keeps prices, discounts and anything binding at the first or second level for good.

The difference between rules that run on their own and software that decides is set out in [agentic AI vs workflow automation](https://intellsys.cloud/blog/agentic-ai-vs-workflow-automation). What a dealer-led sales team can use from AI today is in [AI for sales teams](https://intellsys.cloud/blog/ai-for-sales-teams).

> **How Intellsys OS handles it.** Our principle is "Agentic-led. Human-in-command." Intellsys AI runs in three modes, Smart, Hybrid and AI Ops, and the business sets the mode for each process.

## How Intellsys AI applies human in command today

Intellsys AI works at all three levels, as three modes. In Smart, people run the process and Intellsys AI suggests. In Hybrid, it prepares the action and a person approves. In AI Ops, agents act inside limits the business sets, and every action is logged, explained and reversible. AdGPT is an analyst a person questions in plain language about connected advertising data: it shows its reasoning step by step, uses a sub-agent for each advertising platform, and returns charts and downloadable files, which a person then acts on or not. Separately, sales opportunities are suggested to a team each day, and the team decides what to do with them.

The levels above apply to every mode: the owner of each workflow chooses its mode, moves it up when it has earned it, and can move it back down the same day. More is on [the Intellsys AI page](https://intellsys.cloud/intellsys-ai), and the wider system is in [what a business operating system is](https://intellsys.cloud/blog/what-is-a-business-operating-system).

## Questions people ask

### What does human in command AI mean?

It means people decide how much AI may do in each workflow and remain answerable for the outcome. The AI may recommend, prepare an action for approval, or act within set limits, but a named person chooses that level and can lower it at any time.

### What are the levels of AI autonomy in a business?

Three are useful in a sales and service operation. At the first, the AI recommends and a person acts. At the second, the AI prepares an action and a person approves it. At the third, the AI acts inside limits a person has set, and people review a log.

### Who should decide how much autonomy AI gets?

The person who answers for the workflow's outcome: the sales head for lead follow-up, the finance head for prices, the service head for service reminders. The technology head decides what data the AI may read, and the chief executive sets the rule for raising any level.

### Should AI ever change prices or discounts on its own?

For most businesses, no. A price is a promise to a customer, and a mistake must be explained by a person. AI can recommend a price or prepare a quotation for someone to approve, but the decision stays with people.

### Does Intellsys AI take actions on its own?

Only where the business allows it. Each process runs in Smart, Hybrid or AI Ops. In AI Ops, agents act inside limits set by value, volume, customer and hour, and every action is logged, explained and reversible.

---

[Request a demo](https://intellsys.cloud/demo) · [Blog](https://intellsys.cloud/blog)

This is the Markdown version of https://intellsys.cloud/blog/human-in-command-ai. Intellsys is built by GrowthJockey, a full stack venture architect.
