Zoom adds shared business context to help AI agents turn meetings into work

Zoom is promising more than better meeting notes: AI agents that carry decisions into finished work across business apps. At Zoomtopia on October 20, 2026, it launched a shared context layer alongside coordinated agent teams. Those foundations are available now, but several tools meant to turn captured conversations into follow-up actions are still scheduled for later delivery.

One shared record instead of repeated reconstruction

The Unified Context Layer combines conversations from meetings, phone calls, chat and contact-center interactions with documented company knowledge and connected business systems. Zoom describes two different inputs: fast-changing conversation intelligence and stored records that remain important even when they become outdated.

The system builds a graph—a set of links between people, topics and decisions—and refreshes those relationships in near real time. Connected systems include Salesforce, ServiceNow and Workday. Zoom says this supplies shared memory across its workplace and customer-service products, so each agent does not have to reconstruct the background for every request.

There is an economic claim behind that design. Chief Product Officer Russell Dicker said internal benchmarks showed costs falling 45% to 65% compared with standalone agents parsing information individually, according to SiliconANGLE. Zoom provided no independent validation of that comparison; the figures are company-reported results, not an independently established saving.

Agents divide the project; people review the result

ZoomMate Agent Teams puts specialized agents on different parts of a project at the same time. A lead agent assigns and coordinates tasks, then combines the outputs into one deliverable for human review. The advertised workflow is parallel work with a coordinating agent, rather than a single assistant handling every part in sequence.

Search is another available piece. Zoom says its agentic search can find answers across Meetings, Phone and Chat, plus more than 40 third-party applications. Named connections include Google Drive, Microsoft OneDrive, ServiceNow and Workday, giving the search tool sources beyond the conversation itself.

The context is also intended to travel outside Zoom's client. Through Model Context Protocol connectors and application programming interfaces—software connections that let other tools retrieve information—Zoom says platforms including Claude, ChatGPT and Codex can access its conversation intelligence. Starting in November, users are also due to launch Zoom workflows from Claude and Codex, including meeting, chat and phone tasks and updates to customer records.

Broader context comes with handling rules

Zoom paired the wider access to business information with governance plans. Sensitivity labels can synchronize with Microsoft Purview or use native classifications to enforce handling policies. Planned centralized AI management would add visibility, audits and administrative controls, while custom guardrails are designed to restrict sensitive information in outputs and require human review where specified.

Dicker said administrators could inspect agent activity, shut down agents that violate policy and review sources behind responses. Zoom also allows human review before high-stakes actions. Separately within the same release, planned deepfake detection would analyze video and audio and assign confidence scores to suspected external manipulation. These are different controls: some govern what agents may do with information; another assesses whether incoming media may have been manipulated.