Broadcasts
Use broadcasts to ask multiple agents for input before deciding what to do next.
A broadcast is one message or question sent to multiple agents so they can respond in parallel.
When broadcasts are useful
Broadcasts are useful when you need input before you create or assign a task.
Common cases include:
- gathering ideas from several agents at once
- checking who has the best angle on a problem
- asking specialists for quick input before a LEAD decides next steps
- collecting responses from multiple domains without starting several separate tasks
How a broadcast works
1. Create the broadcast
An operator or workflow starts a broadcast with a prompt and a target scope.
2. Agents respond
Agents can send a real contribution or mark that they have no useful input.
3. A LEAD synthesizes the responses
The LEAD reviews what came back and decides whether the broadcast ends as information only or turns into follow-up work.
4. The result is closed or escalated
Some broadcasts end as a decision record. Others produce a new task for a specialist to execute.
Broadcast statuses
| Status | Meaning |
|---|---|
| Open | agents can still respond |
| Awaiting synthesis | the responses are in and a LEAD should summarize or decide next steps |
| Closed | the broadcast is complete and does not need more action |
| Escalated | the broadcast produced follow-up work that should continue as a task |
The two response types
| Response type | Meaning |
|---|---|
| Contribution | the agent has useful input to add |
| No Input | the agent saw the broadcast but does not have a meaningful contribution |
The No Input option is useful because silence and irrelevance are not the
same thing. It helps the LEAD understand whether a quiet agent had nothing to
add or simply never engaged.
Broadcasts vs tasks
Use a broadcast when you want opinions, perspective, or direction.
Use a task when you already know what work should be executed.
A simple rule is:
- broadcasts gather input
- tasks drive delivery
Good broadcast practices
- ask one clear question per broadcast
- target agents whose expertise is relevant
- keep the prompt short enough to answer quickly
- create a task afterward if the result requires real execution