Sequential research is slow. The agent fetches three sources one by one, then a fourth, then summarises. Most of that time is network wait, not work, and a serial loop wastes it.
Independent reads or computations whose results combine at the end. Multi-source research, parallel API checks, batch document summarisation.
When step N depends on step N-1's output. Parallelism adds nothing if the data is causally chained.
Drop this into an app.yaml. Adjust the credential refs and module names to fit your existing setup.
1schema_version: 223app:4 app_id: fan-out-join5 name: "Fan out, join"6 version: "1.0.0"78modules:9 web: {}10 agent_spawn: {}1112runtime:13 mode: one_shot14 entry_agent: lead1516agents:17 - id: lead18 role: coordinator19 modules: [{agent_spawn: [agent]}]20 brain: { provider: anthropic, model: claude-sonnet-5, credential: { ref: anthropic_main, scope: per_user, provider: anthropic } }21 system_prompt: |22 For research tasks, dispatch three explorer sub-agents, one per source23 domain (news, academic, vendor), via24 agent(agent="explorer", task="research the <domain> angle").25 Once all three have reported back, synthesize their findings.2627 - id: explorer28 role: specialist29 modules: [{web: [search, fetch]}]30 brain: { provider: anthropic, model: claude-haiku-4-5, credential: { ref: anthropic_main, scope: per_user, provider: anthropic } }31 system_prompt: "Find 3-5 sources for the assigned angle. Return facts and citations."Walking through the YAML one block at a time so the design is clear, not memorised.
Three calls to agent(agent="explorer", task="...") go out, one per research angle. Each spawns its own turn against its own (cheaper) brain.
Each explorer only sees its own angle and the web module. No shared state to coordinate, no risk of one explorer's findings polluting another's search.
Each spawned agent's result comes back into the coordinator's context as a tool result, ready to synthesize.
The lead's next turn writes the final answer using all three results. Wall-clock time is close to the slowest single explorer, not their sum.
The pattern above is not the only answer. Here is when something else is the right call.
Easier for the model to reason about, slower. Good when each step's output narrows the next query.
Run one explorer multiple times in sequence with different prompts. Cheaper to reason about, sequential cost.
Engineering notes from the Digitorn team. No marketing, no launch announcements, no "10 prompts that will change your life". Just the things we write that we'd want to read.