Multi-Agent Collaboration
A complex task is distributed across multiple specialist genies, coordinated by a recipe — in sequence, via Assign Task to Genie.
Why This Pattern
Splitting work across specialists isn't about parallelism — that's the previous pattern. It's that a single genie asked to be expert in five domains at once becomes unreliable, because every added tool competes for the LLM's attention. Splitting the work is a bet that narrower, more reliable specialists plus recipe-level coordination overhead beats one broad, failure-prone generalist.
The bet only pays off if the sub-tasks are genuinely separable — and if headless coordination is acceptable, since Assign Task to Genie can't carry user identity or approvals across that boundary.
When to Use It
- Deep expertise is needed across multiple distinct domains at once (HRIS, IT provisioning, communications, and so on).
- A single genie would have too many tools to select from reliably.
When NOT to Use It
- The task is sequential but simple enough for one genie — the coordination overhead isn't worth it.
- The specialist genie calls need Verified User Access, Business Approvals, or permission-aware knowledge bases — Assign Task to Genie supports none of these in a headless invocation.
How It Works in Workato
A recipe calls Assign Task to Genie for each specialist, collecting structured output from each call — in sequence, per Workato's own documentation, not simultaneous autonomous dispatch. Where genuine concurrency is needed for similar items, that's Parallelization, a different pattern.
Real limitations
- Assign Task to Genie can't be used with Verified User Access — there's no user context in a headless invocation.
- It can't be used with Business Approvals — approval notifications are delivered through the user's chat interface, which a headless call doesn't have.
- It can't enforce permission-aware knowledge bases without a user context.
Source: Workato Docs: Agent orchestration · Agentic Design Cookbook V2.0, pp. 31–34