# Playbooks

Playbooks are predefined sequences of **steps**—each step is instructions for a Worker to perform a task. So a playbook doesn’t “run on” a Worker; it defines what Workers do, step by step (e.g. “Executor update these products,” “Analyst produce a summary”). Merchants run playbooks **manually** or as **bulk actions** (no events or schedule), so outcomes stay intentional. You can ship these workflows to merchants or embed them in your apps for consistent, predictable automation when they choose to run them.

## What you can build

- **Repeatable operations** — Product enrichment, feed updates, content localization, or reporting. Same steps every time; merchants run them when they want (e.g. on selected products as a bulk action).
- **Cross-system flows** — Chain steps that touch the store, external APIs (via [MCP Tools](./mcp-tools)), and different [Workers](./workers) (Executor, Planner, Analyst) in one playbook.
- **Packaged automation for partners** — Define playbooks once and offer them as part of your solution so merchants get best-practice workflows under their control.

Playbooks are ideal for repetitive tasks where the logic is stable and you want reliability and consistency.

## How playbooks work

A playbook is an ordered list of **steps**. Each step assigns a task to a Worker—and **each step can be assigned to a different Worker** (Executor, Planner, Analyst, etc.): one step might direct the Executor to apply changes, the next the Analyst to produce a report, and so on. When the merchant runs a playbook, those steps are carried out in order. Workers have access to the [MCP Tools](./mcp-tools) and store context you configure for them.

You define the steps; execution is always started by the merchant (manual or bulk).

## Success Agent and playbooks

The [Success Agent](./success-agent) can **create** playbooks (e.g. help the merchant define or save a workflow). It **cannot execute** playbooks—only the merchant can run them.

## When to use playbooks vs. agents

- **Use playbooks** when the workflow is well-defined and you want the same steps every time, with the merchant in control of when it runs.
- **Use the Success Agent or Workers in agent/planning mode** when the path isn’t fixed—when the merchant or system needs to explore, decide, or adapt the plan based on context.

The Success Agent can help create or refine a playbook; the merchant then runs it when ready.

## Combining with the rest of AI Commerce

- [Workers](./workers) perform the tasks defined in playbook steps (e.g. Executor applies changes, Analyst produces reports). Configure each Worker with the tools and MCP integrations its steps need.
- [Guidelines](./guidelines) can influence how agent-driven steps behave (tone, safety, scope).
- The [Success Agent](./success-agent) can create playbooks but cannot execute them.

[Overview: AI Commerce →](./overview) | [Workers →](./workers) | [Success Agent →](./success-agent)