Skip to main content
Beta

AI Foundry is in beta. We are actively shaping the product, so things may change as we iterate. Your feedback is welcome.

Playbook

A Playbook is a catalog resource that describes a multi-step agentic workflow. It composes agents, prompts, skills, and spec templates into a directed graph where each node performs a discrete task and edges define the flow of control and data between them.

Playbooks move the AI platform from single-agent interactions toward agentic pipelines: complex, stateful workflows where the output of one agent feeds the input of the next, loops repeat until a condition is met, and parallel branches execute concurrently.

AI Foundry Playbooks

Playbook reference

FieldRequiredDescription
TitleYesDisplay name shown in the UI.
NameYesUnique identifier. Immutable after creation.
DescriptionYesShort description of the workflow's purpose.
Show on HomeNoWhen enabled, this playbook appears as a quick-launch button on the home screen.
Launch ModeNoDefault UI to open when this playbook is launched.
Home PromptNoOptional pre-filled prompt auto-sent when the user clicks this playbook from the home page.
Source Template IDNoOptional source template linked to this playbook on the home screen.
Project Template IDNoOptional project template linked to this playbook on the home screen.
PromptsNoBuilt-in Agents (reference)
SkillsNoSkills available to all agents in this playbook.
Spec TemplatesNoSpec templates available to all agents in this playbook.
Built-in Agents (reference)NoBuilt-in external AI agents referenced by this playbook.

Playbook flow model

A Playbook workflow at its core comprises a set of orchestrated agents working sequentially, in parallel or loop.

You can either deploy a single built-in agent (an external AI agent not managed by the Catalog, like a model provider or third-party assistant) or a combination of agents managed by the Catalog.

The visual builder

AI Foundry Playbook Builder

The Playbook Builder in the AI Foundry UI provides a drag-and-drop canvas for designing and editing without writing JSON:

  1. Overview step: set name, title, and description.
  2. Agentic Flow step: add nodes from a palette, draw edges between them, and configure node properties.
  3. Resources step: attach playbook-level prompts, skills, and specs using multi-select pickers.

Switching to JSON mode replaces the visual builder with a full-spec JSON editor, which is useful for bulk edits or for copying playbooks across environments.

Resource scoping

Resources (prompts, skills, specs) can be attached at two levels:

  • Playbook level: available to every node in the flow.
  • Node level: available only to that specific node. Node-level lists are merged with the playbook-level lists.

This allows a common baseline (e.g. a shared style guide) while letting individual nodes use specialized resources.

See also

  • Agent: the execution unit referenced by playbook nodes.
  • Prompt: reusable text injected into nodes.
  • Skill: reusable capabilities attached at playbook or node level.
  • Spec: structured documents referenced by nodes.