Subagents: Agents That Control Agents.
The new feature in detail. How subagents delegate tasks and why monolithic super-agents are a thing of the past.
In mid-February, Langdock launched Subagents, a feature that fundamentally changes how you work with agents. For the first time, agents can call other agents and delegate tasks to them. Sounds abstract, but it is not.
What Has Changed
In mid-February, Langdock launched a feature that fundamentally changes how you work with agents: Subagents. For the first time, agents can now call other agents and delegate tasks to them.
In this issue, we show you what the feature can do, how to use it, and where the limits are.
What Are Subagents
Until now, every agent in Langdock was an island: its own instructions, its own Knowledge Base, its own actions. Anyone who wanted to cover different tasks had two options.
Build one giant agent that can do everything (and eventually does nothing well).
Build several specialized agents and switch between them manually.
Subagents solve exactly this problem. A parent agent can now include specialized agents as subagents and automatically delegate to them.
How Do Subagents Work
The process runs in three steps.
Recognition
The main agent recognizes from your question that a task matches one of its subagents.
Delegation
It formulates a prompt and sends it to the subagent. The subagent works completely independently with its own settings (model, Knowledge Base, Actions).
Result
The subagent returns its response. In the chat, you see this as an expandable tool action, similar to other actions.
The subagent does not see the full chat history. It only receives the specific prompt formulated by the main agent. This is by design: every agent stays focused on its task.
Real-World Example: Onboarding Assistant
Here is a concrete example of what subagents can look like in practice.
"Onboarding Assistant"
An agent that helps new employees on their first day. It welcomes them, answers general questions, and delegates specialized topics to two subagents.
"IT Setup Guide"
- Knowledge Base: IT documentation (VPN setup, tool access, hardware ordering).
- Model: Claude Sonnet 4.6 Reasoning for fast, precise answers.
- Task: Everything related to technical setup.
"HR Expert"
- Knowledge Base: HR policies, benefits overview, vacation regulations.
- Actions: Can provide links to relevant forms.
- Task: Everything related to employment law and organizational questions.
What This Looks Like in Practice
More Ideas: Subagents in Practice
The onboarding example is just the beginning. Four more scenarios for inspiration.
| Main Agent | Subagents |
|---|---|
| Sales Assistant | Product Expert (with product documentation) and Market Analysis Agent |
| Project Manager | Jira Agent (status queries) and Controlling Agent (budget) |
| Customer Support | Technical Agent (technical documentation) and Billing Agent (invoice questions) |
| Strategy Sparring | Analyst Agent (Code Interpreter) and Research Agent (web search) |
How to Set Up Subagents
Setup is straightforward.
- Open the configuration of your main agent.
- Go to the "Actions" section.
- Click "Add Subagent" and select an existing agent.
- Done. The main agent can now call the subagent automatically.
You must be at least an Editor or Owner of the agent you want to add as a subagent.
Everyone who has access to the main agent can automatically use the subagents as well. If someone does not have direct access to a subagent, the request is declined, but access can be requested directly.
Limitations
Three points to keep in mind.
Subagents cannot call their own subagents. There is only one level. A subagent is always an "endpoint".
The subagent does not see the full chat history of the main agent. It works only with the prompt it receives. So if you are having a complex dialog and then trigger a subagent task, the main agent has to pass all relevant information to the subagent in the prompt.
If a subagent has actions requiring confirmation (e.g., sending a Slack post), you still need to confirm those manually. This is a safety feature.
Our Tip
Start small: take an existing agent that does too much at once and offload one clearly defined task into a subagent. This way you will quickly see whether and how the delegation works.
Until next month,
Timo