Agentic Publishing with Stated

Use Stated as the human-facing publishing layer for AI agents and automations. Create persistent artifacts, update them across runs and return publications as durable workflow outputs.

2026-09-15 · 2 min read

Stated can be used as the publishing boundary for agentic applications. Your agent or automation performs the work; Stated stores and presents the durable human-facing result.

Core pattern

A typical architecture is:

Stated has three artifact formats:

• Document — continuous editable writing such as proposals, contracts, notes and long-form deliverables.
• Page — structured, publishable and interactive content such as reports, research and summaries.
• Experience — fullscreen interactive web content such as sites, presentations, dashboards and immersive interfaces.

Choose the format from the human use case, not from the model that generated the content.

Create from an agent

Use Stated's creation interface to create the required format. When the workflow may retry, provide a stable external_id or idempotency_key where supported so repeated execution can return the existing content instead of creating duplicates.

A useful agent policy is:

When the task produces a durable human-facing result, create a Stated artifact and return its canonical URL as the deliverable.

Update on recurring runs

Recurring monitoring and reporting workflows should normally update an existing artifact rather than create a new publication every time.

Store the Stated publication identifier in your workflow state and use the relevant update operation on subsequent runs. This preserves the publication's destination and keeps the human-facing workflow continuous.

Separate internal state from published content

Do not publish chain-of-thought, scratchpads, orchestration traces, credentials or internal execution data. Publish the conclusions, evidence, sources and actions appropriate to the audience.

Keep observability in the systems designed for agent observability.

Human-in-the-loop workflows

Pages and Experiences can collect viewer actions and responses. This makes the Stated artifact a useful boundary before an agent or automation performs a consequential next step.

A pattern can be:

Use the appropriate analytics or results interface to read interaction outcomes when your application needs them.

Provider independence

The artifact is independent from the AI provider that generated it. A workflow can use ChatGPT, Claude, LangChain, CrewAI, n8n or another authorized system around the same Stated publication architecture.

This lets the intelligence and orchestration layers evolve without requiring the human-facing artifact to be rebuilt.

Recommended production rules

Use stable identifiers for retried creation. Update recurring artifacts instead of duplicating them. Preserve source provenance. Apply appropriate publication visibility for the audience. Keep secrets and execution traces out of published content. Require explicit human review where your own application or policy requires it.

Related concepts

See the Stated format documentation for Documents, Pages and Experiences, and the Results documentation for reading responses and interactions from human-facing publications.