AI maturity: how people and agents share the work¶
Studio OS has three AI maturity levels. They describe who starts work and when an agent asks a person to decide. They do not measure how capable your company is, and level 1 does not turn AI off. A new workspace starts human-led so its team can build context and trust before giving agents more initiative.
| Level | Who starts the work? | What the person does |
|---|---|---|
| 1 · Human-led | A person requests work, approves a proposal, or an incoming business event needs to be recorded. Agents observe, organise and prepare within that request. | Set the direction, request next steps and decide what to do with issues agents surface. |
| 2 · Collaborative | Agents can also notice a need, make company-wide suggestions and start bounded work in their own areas. | Give judgment, resolve missing facts and review actions that require approval. |
| 3 · AI-led | Agents act more independently within their configured mandate and keep people informed. | Set objectives and boundaries, supervise outcomes and make decisions that still require a person. |
Level 1: human-led, with AI at work¶
You can ask an agent to research a question, prepare a proposal, organise a project or follow an assigned task. The agent can continue that requested work across handoffs. Incoming enquiries and replies can still be read and recorded; agents can maintain the client and project picture, draft a response and surface a material issue. The operator decides whether a newly detected need becomes another piece of work.
For example, if a delivery date slips, Studio can record the change and tell you what it found. It does not independently launch a new recovery plan or assign colleagues to one. If you ask Madis to replan the project, he can do that within your request.
This level is different from the separate pause autonomous work switch. Pausing background cycles still lets you chat with agents and request work manually. Human-led allows observation and authorised work to run; it limits self-started interventions.
Level 2: collaborative initiative¶
Agents can identify gaps and opportunities without waiting for a new request. They may prepare suggestions about sales, finances or operations, follow their standing responsibilities and carry out routine work within their permissions. They involve a person when a preference, missing fact or consequential decision changes the outcome.
For example, Sven may connect a full delivery calendar with an active sales pipeline and suggest a later start date. The agent can assemble the evidence and propose the next step; the right person still decides on a new customer commitment.
Level 3: the direction of an AI-led company¶
Our longer-term aim is a company where agents can carry more of a previously established operating routine across sales, marketing, delivery and administration. They would remember approved ways of working, coordinate across roles and handle repeatable steps with less prompting, while people set the mandate and intervene at genuine decision points.
Imagine an ordinary week in that future: an enquiry is qualified against agreed criteria, a familiar offer is prepared from known prices, the delivery calendar is adjusted before capacity is exceeded, a proven campaign continues, and the next invoice is ready when the work is accepted. The owner reads one coherent account of what happened and steps in for exceptions and new commitments. That is the direction, not a description of what selecting level 3 automates today.
The current level 3 is a step toward that aim, not a promise of a business that runs itself. Today it tells agents to use their judgment within their mandate and report back more often after routine work. Levels 2 and 3 still share many of the same execution gates. The available tools, integrations, approvals and explicit permissions determine what any agent can actually do.
Boundaries that apply at every level¶
Changing the level does not itself authorise sending a letter, posting an invoice, publishing a campaign or making a new commitment. Those actions follow their own approval and permission rules. A specifically authorised workflow step may have its own approval in advance; raising the maturity level alone is not that authorisation. The active agent roster and connected systems also limit the work available in a given workspace.
An administrator can see or change the level under Admin → AI. Some workspaces have the level fixed in their company profile, in which case the on-screen choice is disabled. Start with a narrow, observable workflow, review what the agents record and propose, and widen their mandate when the routine and its limits are clear.
For a wider view of the controls, see Operating Studio OS and Features & agent tools.