Introducing Spawn Workflows: automation that thinks
Chain agents into multi-step pipelines with triggers, conditions, and parallel execution. The building blocks for autonomous operations.
For the last year, Spawn agents have been running one-shot tasks — you describe a job, the agent does it, you get a result. That model is clean, but it has a ceiling: real businesses don't run on one-shot tasks. They run on chains of tasks that fire on triggers, branch on conditions, and hand off between people and systems.
Today we're shipping Spawn Workflows — the building blocks for chaining agents into autonomous pipelines that actually do the work your team does every week.
What a workflow is
A workflow is a graph. Nodes are agents (or webhooks, or schedulers, or human approvals). Edges carry data between them. You define the graph once and it runs forever — on a cron, on an incoming email, on a webhook from Stripe, on a Slack message matching a pattern.
Every node is a full Spawn agent with its own tools, memory, and sandbox. That means a workflow step can browse the web, query a database, write a spreadsheet, post to Slack — whatever the job requires. Steps pass structured data or unstructured text between them; the next node reasons over what came in and decides what to do.
The four primitives
- Triggers — cron schedules, webhooks, inbound email, Slack events, database row changes, or a manual "run now."
- Agent steps — any Spawn agent with any toolset. Parameterized by the data coming in.
- Branches — conditional routing based on the output of a previous step. If the lead is enterprise, route to sales; otherwise route to the self-serve flow.
- Fan-out / fan-in — run 50 agents in parallel against 50 inputs, then collect the results in a single step downstream.
Those four primitives are enough to express almost every back-office process we've seen customers try to automate.
“A workflow is a graph. Nodes are agents. Edges carry data. You define it once and it runs forever.
Why agents instead of Zapier-style glue
Traditional automation platforms work when every step is a deterministic API call. Most real work isn't deterministic. "Read the inbound RFQ and extract the line items" is an agent task, not an API call. "Summarize the last 30 days of support tickets and flag the top three root causes" is an agent task, not an API call.
Workflows let you put agents where the judgment lives, and keep the connective tissue simple. You're not writing if/else trees in a visual builder — you're telling an agent what to do and letting it decide how.
What's shipping today
- Visual workflow editor with drag-and-drop nodes and live preview.
- Cron, webhook, and email triggers — plus Slack, Gmail, and GitHub event sources.
- Parallel fan-out (up to 200 concurrent agents per step).
- Per-step memory and artifacts — every run is inspectable end-to-end.
- Template library — 40+ starter workflows for common operations tasks.
What's next
We're building out human-in-the-loop approvals, SLA-aware routing, and a lower-level SDK for teams who want to define workflows in code. If you're running operations that look like a graph — and they almost always are — we want to hear what you'd build with this.
Workflows are live for every Spawn account today. Head to /workflows to try it.