Your coordinator gives a fresh subagent the task "summarize what we found so far." It returns nothing useful. Why?
Why D is the answer
It starts without the conversation history, so the findings must go in its task
A fresh subagent does not inherit the conversation or files already read. Anything it needs has to be passed in explicitly.
In detail
- A new subagent has its own context window. It does not receive the coordinator's conversation or the files already read.
- "What we found so far" points at information the subagent cannot see, so it has nothing to summarize.
- The fix is in the coordinator: put the findings, or the location of the findings, directly in the task text.
- Avoid words such as "above" or "so far" in a task, because they point at context the subagent lacks.
Use case
An illustrative scenario, not a real incident.
A coordinator asks a fresh subagent to "summarize what we found so far". The subagent has seen nothing, so it returns a vague reply. Passing the notes in the task produces a real summary.
See it in different ways
flowchart LR A["Coordinator"] -->|"Task plus findings"| B["Research subagent"] B -->|"Summary"| A A -->|"Task plus summary"| C["Synthesis subagent"] C -->|"Report"| A
sequenceDiagram participant C as Coordinator participant S as Subagent C->>S: Task text, with everything it needs S->>S: Reads, searches and thinks in its own context S-->>C: Final summary only
sequenceDiagram participant P0 as Coordinator participant P1 as Fresh subagent P0->>P1: Summarize these notes, notes included P1-->>P0: Summary of what it was given
Key point: Write each task as if the reader has seen nothing before.
Why the other options are wrong
- A. The subagent used a smaller model that cannot read earlier conversation turns
- Model size is not the cause. Any subagent starts without that history.
- B. Summaries are only produced when structured output has been turned on
- Structured output controls the shape of a reply, not what information the subagent has.
- C. Subagents are unable to summarize text and must call a separate summarizer tool first
- Subagents can summarize text directly. No separate tool is needed.