Choose how Rovo Agents respond with subagents

[#1]

Hi agent builders!

We’re adding a new way to control how agents with subagents respond to users.

When your agent uses a subagent, you can now choose whether the main agent should combine the subagent’s output into a final response, or whether the subagent’s answer should be forwarded to the user as-is.

This gives you more control over who owns the final response, especially when different parts of your agent are designed to behave like distinct specialists.

What's changing?

Subagents are task-focused helpers that can have their own instructions, knowledge, tools, and routing description. They are useful when your agent needs to handle distinct jobs that benefit from specialised guidance.

By default, when a subagent is selected, it returns an output to the main agent. The main agent then uses that output to write the final user-facing response.

This flow covers most complex use cases and remains the default setting, especially when the main agent needs to orchestrate multi-step workflows, combine outputs from multiple specialists, and keep one consistent voice.

For some use cases, the subagent is the specialist that should answer directly. For example, when it has a specific response format, tone, support workflow, or set of instructions that should be preserved in the final answer.

With this update, builders have more flexibility to configure how their agent handles subagent responses.

Choose how your agent handles subagent responses

There are two options:

Combine subagent outputs to generate the final response

The subagent returns its result to the main agent, and the main agent writes the final response.

Use this when the main agent should act as the orchestrator. For example, this works well when your agent needs to:

• Compare options from different sources
• Summarise findings from one or more specialists
• Combine product guidance, policy guidance, and user context into one answer
• Keep one consistent voice across the whole conversation

Forward the subagent’s output as-is

The subagent owns the user-facing response, and its answer is sent directly to the user.

Use this when the subagent is the right specialist to answer directly. This is especially useful when each domain has a distinct way of handling requests, such as:

• A support specialist that follows a specific troubleshooting flow
• A service-focused subagent that needs to use a particular tone, format, or escalation path
• A formatter that must return a precise structure
• A policy or refusal specialist whose wording should be preserved

In this mode, the main agent can route the request to the right specialist, then let that specialist take over the final answer.

Note:

• Response configuration does not change tool or skill access. Subagents do not inherit tools, skills, knowledge, or instructions from the main agent. If a subagent needs a specific tool or skill, add it to that subagent directly.
• Only one subagent can be selected per turn in this setting, to avoid multiple subagents responding at the same time.

Why this is useful

• More control over final answers: Choose whether the main agent or the specialist subagent should own the response.

• Better specialist experiences: Preserve the tone, structure, and instructions of subagents that are designed to answer directly.

• Flexible orchestration: Keep using the main agent to combine specialist inputs when the user needs one synthesised answer.

How to think about it

Start with the default: let the main agent combine subagent outputs.

This is best for agents that need to bring multiple pieces together to provide deep context, and should work for the majority of use cases.

For example, comparing plans, summarising findings from several sources, or combining product, policy, and account context into one answer.

Choose forward the subagent’s output as-is when the subagent should behave like the final responder.

This is best for specialist flows, commonly used in customer support or service use cases.

For example, troubleshooting a specific service issue (VPN issue, licensing queries, hardware setup, employee onboarding, etc), following a specific support playbook, returning a structured form, or using wording that must stay consistent for policy or compliance reasons.

Frequently asked questions

1. Do I need to change my existing agents?

No, if your current agent is working as expected, you can keep the existing behaviour. This update gives you an additional configuration option for agents where a subagent should own the final response in suitable use cases.

2. Which option should I choose?

Choose combine outputs when the main agent needs to synthesise information or keep one consistent voice. Choose forward output as-is when the subagent is the specialist that should answer directly and its response instructions should be preserved.

3. Do subagents inherit tools or skills from the main agent?

No. Subagents do not inherit tools, skills, knowledge, or instructions from the main agent, regardless of which response option you choose.

The response setting only controls who writes the final user-facing answer:

• Combine subagent outputs to generate the final response: the subagent returns its result to the main agent, and the main agent writes the final response.

• Forward the subagent’s output as-is: the subagent writes the final response directly.

In both cases, each subagent only has access to the tools, skills, knowledge, and instructions configured for that subagent. The main agent does not automatically know what tools a subagent has, and a subagent does not automatically know what tools the main agent has.

If a subagent needs to use a specific tool, skill, or knowledge source, add it to that subagent directly.

4. Can more than one subagent respond directly?

When forwarding a subagent’s output as-is, only one subagent responds directly in a turn so the user receives a single final answer.

5. Can I still use subagents as helpers?

Yes. Subagents can still work like helpers that return information to the main agent. The new configuration simply lets you choose when a subagent should instead own the final response.

Try it out

If you build or manage Rovo Agents, open Studio and review the response configuration for your subagents. Try using direct subagent responses for specialist workflows where the exact tone, format, or support process matters.

We’d love to hear how this helps you build clearer, more specialised agent experiences. Share your questions and feedback in the comments.