Create your first AI Agent
Choose the provider, model and reasoning level that suit the work you want to hand off.
An Agent supplies the intelligence for a Task. It does not need a permanent job title or a separate personality for each business function. The Job and its Tasks tell it what to do.
Create a practical everyday Agent first. You can add another with stronger reasoning when a particular kind of work calls for it.
Before you start
- You are in the correct client workspace.
- An Owner or authorised Admin has activated an AI provider.
- You have permission to create Agents. Operators can use available Agents but cannot create or edit them.
Create the Agent
- Choose a name your team can recognise
For example, Agent L. The name helps you identify the Agent when assigning Tasks and following work. It does not limit the Agent to a particular role.
- Select the provider and model
Choose an enabled provider, then a model from the available list. Start with the complexity of the work rather than automatically selecting the cheapest or most powerful model.
- Choose a reasoning level
Where available, this controls the reasoning effort the model applies. Use a practical starting level and evaluate the result on a small, representative Job.
- Select Create AI Agent
Return to the Agent registry and check its availability. An inactive Agent or unavailable provider needs attention before you can use it for new work.
Match intelligence to the work
| The work you need done | What to consider |
|---|---|
| Summarise emails or organise known information | An everyday model may be a good starting point. Check that it captures the important details. |
| Compare competing options or investigate an unusual case | Try stronger reasoning and give it clear criteria for making the decision. |
| Repeat a narrow, well-defined action at volume | Test efficiency and consistency. A lower token price is useful only if the work is completed reliably. |
The same Agent can contribute to a sales Job today and an accounts Job tomorrow. You can also assign different Agents to different Tasks within one Job.
Try a small Job before relying on it
Ask the Agent to produce something you can check easily, such as a summary of a known document. Review the result, tool activity and usage. If it misses the brief, improve the instructions or try another model before using it for ongoing operations.