Zoom and Couchbase target AI agent context and cost bottlenecks

The Zoom layer combines conversation intelligence with documents and information from connected systems including Salesforce Inc.’s CRM, ServiceNow Inc.’s workflow platform and Workday Inc.’s human capital management application. Zoom builds a graph linking people, topics and decisions and updates those relationships in near real time so agents do not have to reconstruct context for each request. The layer is available now and provides shared memory and agent coordination across Zoom Workplace, customer experience products and connected applications. Chief Product Officer Russell Dicker said during a briefing that the context layer has “already synthesized and distilled that information.” Zoom said internal benchmarks showed the approach reduced costs by 45% to 65% compared with standalone agents parsing information individually; the company did not provide independent validation of that comparison.

ZoomMate Agent Teams, also available now, assigns specialized agents to different parts of a project at the same time, with a lead agent coordinating tasks and combining output for human review. Agentic search draws answers from Zoom and more than 40 third-party applications. The My Notes note-taker is expanding from meeting capture into workflow automation, with users able to send summaries to Salesforce, Zoom Chat and other destinations using templates or natural-language instructions. A forthcoming standalone mobile app and recording devices from DTEN Inc. and GalaxySailLab Inc.’s Bota extend capture to in-person conversations. Zoom is also adding personalized transcripts for invited users who miss meetings, links from screenshots to transcript passages, and a Zoom for Business Owners product that connects meeting context with its Bonsai business management and client engagement platform.

Within Workplace, Zoom is upgrading translation so a speaker can broadcast translated audio to participants without requiring each listener to hold a separate license, and it is adding voice cloning intended to preserve the speaker’s vocal identity. A planned live meeting assistant will privately suggest answers and guide pacing and engagement. A redesigned Chat feature lets agents participate alongside employees, while a Threads page aggregates replies and an Activity page organizes mentions, reactions and threads. Zoom is also opening the same intelligence to outside applications through expanded Model Context Protocol servers and Search and Ask APIs, and it announced 15 new AI services plus a Platform Studio portal that combines AI Services, its Video SDK and Co-Browse SDK. TechRadar reported that the new APIs would let external AI tools such as ChatGPT, Claude and Gemini query the Unified Context Layer directly.

Couchbase’s expansion includes native Amazon Bedrock integration with semantic caching, a fully managed Agent Memory and Model Context Protocol server on Amazon Web Services, and vector search in its analytics engine, now called Operational Insights. During a pre-briefing, Tim Finley, Couchbase’s senior vice president of AI, said agents are a different kind of user. He cited a Cisco Systems Inc. study finding that agents generate 450% more traffic per task than humans performing the same work, and Cloudflare data showing that machine traffic surpassed human traffic on the web in June, about 18 months earlier than forecast. “Capacity plans built for human growth curves are likely to break,” Finley said.

Finley also said customers are “actively sacrificing intelligence by downgrading models because of the bill once they hit scale,” a trade-off he linked to wasted spend on repeated questions, bloated prompts and agents that start every session from scratch. Couchbase says semantic caching cuts large language model calls by 40% to 60%; Agent Memory reduces token usage by 33% to 45% compared with alternatives on the LoCoMo long-term conversational memory benchmark; and vector-based context retrieval produces prompts about 70% smaller than full-context approaches. Those are Couchbase’s own figures, and results will depend on how repetitive a workload is. Finley described convergence as different from consolidation because a converged system shares a single copy of data across capabilities. In Operational Insights, he said a phone manufacturer could determine which factories are linked to the most positive battery-life reviews in one SQL query and write the result back for agents to read. “There has been a fallacy for a long time with analytics that a really great insight was the finish line,” Finley said. “We think it’s the action.”

Editor's Summary

Zoom’s Unified Context Layer and Couchbase’s AI Data Plane expansion both address a production problem for AI agents: they need shared context and cheaper access to data. Zoom is opening its conversation data to external agents, while Couchbase is pitching caching, memory and vector search to cut model calls and token use. Both announcements rely partly on company-reported performance figures.