Expert agents

Automate the routine work

Expert agents handle the repetitive, rules-based parts of content production so your team can focus on work that requires human judgment. Set up as many as you need: multiple agents can run simultaneously, each working on different content independently. Here are some things you can automate:

• Automatically translate content into your target languages when a new language variant is created.
• Fill in SEO and GEO metadata before content moves to marketing review.
• Check content against your brand or compliance guidelines.
• Populate content automatically whenever a new content item is created.

How to set up expert agents

Upon activation, go to Environment settings > Expert agents to create a new agent.

1. Describe your agent

Provide details to help your teammates identify and understand the agent’s purpose.

1. Agent name – the name appears in content item version history and in the audit log. Use a descriptive name that makes the agent's purpose clear in any of these contexts, such as "Blog post translator" or "SEO metadata filler".
2. Purpose – a short summary of what the agent does, such as “Translates blog posts to French.” This purpose is for your team’s reference and doesn’t affect how the agent works.

2. Write clear instructions

In Instructions, tell the agent exactly what to do. Because expert agents can’t ask follow-up questions, your instructions need to be complete and self-contained. Describe the input, the steps to take, and the expected output. Your instructions can be as simple or as detailed as the task requires, including conditional logic, structured sections, and formatting to keep long instructions readable. If your instructions are too long, simplify them or split the work across multiple agents with more focused tasks.

3. Specify when to run

Under Automation, you can set various activities and filters to start the agent when a specific content change occurs. For events: 

• Content item language variant created – an event when you create a new item or other language variants of the default content item.
• Workflow step changed to – an event that occurs when your content moves to a specific workflow step.

For filters:

• Content type – specify which content type the agent will work on.
• Collection – select which collection the content will be under.
• Language – scopes the agent to content items in a specific language.

4. Set permissions

In Permissions, select what the agent is allowed to do in your environment. You need to select at least one permission. Under Manage content items:

• Read content items – lets the agent read content items and their variants, along with the project configuration they depend on: content types, taxonomies, workflows, collections, languages, roles, and spaces. Checked automatically when any other content item permission is selected.
• Update content items – lets the agent edit content items and their variants, move them through workflow steps, and publish or unpublish them. Checked automatically when Create content items or Delete content items is selected.
• Create content items – lets the agent create new content items and variants.
• Delete content items – lets the agent delete content items and their variants.

Under Manage assets:

• Read assets – lets the agent read asset metadata and folder structure. Without this, the agent can only see the IDs of assets referenced in content items, not the actual asset details. Checked automatically when when Upload assets is selected.
• Update assets – lets the agent edit asset details (including taxonomy and collection), manage folders, and move assets between them. Doesn't include uploading or replacing files.

5. Monitor your agents’ work

To see a specific agent’s run history, click on the right side of its row. Each run shows the content item it acted on, its status, and when it ran. Click a specific run to see a summary of what the agent did and why. You can also track the expert agent's changes directly in the content items the agent worked on. In the content item's version history, each change made by the expert agent appears as a separate version labeled with the agent's name. The audit log records a similar attribution, so you can always tell expert agent changes apart from changes made by a user or by Aiko.

Example

Here’s an example of how you might set up an agent that helps translate blog post content to a specific language. You can adapt this example to your project.

Blog post translator

Step 1: Write these details to describe your agent:

• Agent name: Blog post translator
• Purpose: Translates blog posts into [your target language] when the content item reaches the [your workflow step] step

Step 2: Copy and paste the following text into Instructions and adjust what’s in your project: Step 3: Customize the automation to start the agent:

1. For events: select Workflow step changed to
   • Set the workflow step to [your workflow step for translation-ready content]
2. For filters, select the following:
   • Content type: your content type, for example, Blog post
   • Collection: your collection, for example, Marketing
   • Language: [your target language]

Step 4:  In Permissions, select Create content items under Manage content items. This also checks Read content items and Update content items automatically.

Considerations

When using expert agents, keep in mind the following:

• Only project managers can access and manage expert agents.
• There are no hard usage limits, but a safety measure activates if many complex agentic runs occur simultaneously and risk overloading the infrastructure.
  • If the safety measure activates, you receive an error message. Contact Support to identify the cause before the restriction is lifted. 
• Not all instructions require the same processing time – many short parallel tasks are typically fine; many long, complex simultaneous tasks may trigger the safety measure.
• Changes made by an expert agent are labeled with the agent's name in both version history and the audit log, so you can tell them apart from changes made by a user or by Aiko.
• Run history shows only recent runs.
• Test your agent instructions on a few items manually before enabling automation at scale.
• Expert agents consume AI credits.

These considerations are meant to keep things running smoothly, not to limit what you can automate.

Help shape what comes next