Artificial Intelligence at Zoom: An Overview

Artificial Intelligence at Zoom: An Overview

Artificial Intelligence is rapidly evolving and becoming an integral part of everyday life, powering everything from large and small language models to specialized algorithmic functions. While generative AI enables seamless interaction between humans and machines for content creation, problem-solving, and decision-making, AI also exists in more specialized applications. Automation tools streamline workflows, while services like transcription and translation break down communication barriers, making collaboration more efficient and accessible. As AI continues to advance, its applications are transforming industries by enhancing efficiency, accuracy, and innovation, fundamentally reshaping the way businesses operate.

Given AI’s expanding role in the modern workplace, the AI Bluepaper serves as a detailed guide to AI features within the Zoom platform. By exploring some of Zoom’s key AI deployments, functions, and capabilities, it provides businesses with the insights needed to harness AI effectively—helping streamline workflows, enhance collaboration, and drive greater productivity across teams.

Agentic Artificial Intelligence: Maximizing the self-driven, autonomous power of AI

Agentic AI refers to artificial intelligence designed to do more than just answer questions—it’s built to take action on a user’s behalf. As an AI-first open work platform built for human connection, Zoom is actively developing agentic AI capabilities to help users seamlessly move from information to execution.

Rather than stopping at suggestions or meeting summaries, the vision for agentic AI is to evolve into an autonomous assistant that can help users manage next steps, complete tasks, and keep workflows moving. As AI technology continues to improve, agentic AI will increasingly function as a personalized digital assistant—helping users stay organized, follow through on priorities, and reduce the manual effort of moving information between conversations, meetings, and work tools.

Zoom AI Companion: Your Intelligent Workplace Assistant

Zoom AI Companion is an intelligent, conversational, digital assistant at the core of the Zoom Workplace platform, designed to enhance productivity, streamline collaboration, and reduce employee workload—all at no additional cost for customers with select paid services assigned to their Zoom user accounts.

As AI tools become increasingly integrated into the modern workforce, they are no longer limited to performing specific tasks—like automating data entry, drafting emails, generating reports, or performing basic algorithmic functions. Rather, emerging AI tools represent a transformative shift, fundamentally changing the way we work. Zoom AI Companion goes beyond simple task automation; it can act as a true companion in a user’s daily work, capable of answering general knowledge questions, generating content, offering insights, summarizing documents, messages, voice messages, and meetings, assisting with decision making, and helping users achieve new goals or unlock higher levels of performance.

As an AI-first company, Zoom AI Companion is thoughtfully integrated across Zoom’s suite of products, including Zoom Meetings, Team Chat, Phone, Contact Center, Docs, Whiteboard, and more. As Zoom continues to innovate and expand its capabilities, so too will Zoom AI Companion, offering new features and functionalities. Below is a list of key features available today, showcasing how this AI assistant can help enhance every aspect of a user’s workday.

Current key features and functionalities include:

Category	Key Capabilities	Business Value
Centralized AI Workspace	Contextual query processing, smart scheduling, cross-platform data synthesis, AI-generated follow-ups.	Unified productivity hub with intelligent workflow automation.
Meeting Intelligence	In-meeting real-time assistance, automated documentation, smart recording analytics, preparation automation.	Enhanced meeting effectiveness from preparation to follow-through.
Communication Enhancement	Intelligent chat assistance, transcript analytics, email composition, predictive writing.	Accelerated communication quality across all channels.
Contact Center Optimization	Real-time sentiment analysis, conversation intelligence, performance analytics, smart responses.	Elevated customer experience with AI-driven insights.
Content Creation	Document intelligence, visual content generation, collaborative tools, media organization.	Streamlined content development and ideation processes.
Task & Workflow Management	Automated task creation, cross-platform integration, event management tools.	Reduced administrative overhead with intelligent automation.

How Zoom AI Companion Uses the Power of Agentic AI

Zoom AI Companion brings the value of agentic AI to life by helping users easily turn information into action. When AI Companion identifies follow-up tasks—whether from a meeting, a Contact Center interaction, or another insight identified by AI—it can automatically add those action items into Zoom Tasks for follow-up or assignment. This helps ensure that key next steps don’t get lost in a meeting summary or conversation recap. Instead, they become clear, trackable tasks that can move forward. By connecting insights directly to action, AI Companion helps users stay on top of priorities, manage responsibilities, and keep their workday moving without unnecessary friction.

Algorithms: How artificial intelligence helps make communication frictionless

Beyond the intelligent, conversational LLM capabilities of Zoom AI Companion, Zoom also utilizes other AI services (i.e., algorithms) across the platform. These services often operate seamlessly in the background, powering features like real-time voice transcription, live translation, personal audio suppression, and more.

Together, these features help create a cohesive, frictionless experience that can enhance both the quality and efficiency of every interaction on the Zoom platform. Current key AI-service features and functionalities include:

• Transcription

• Translation

• Closed Captioning

• Personal Audio Isolation

Model Context Protocol (MCP): A Standard for Connecting AI to Tools to Accomplish More

As part of its AI architecture, Zoom uses the Model Context Protocol (MCP)—an open standard that enables secure connections between models, tools, data sources, and workflows. MCP plays a key role in advancing agentic AI by replacing the need for one-off integrations that each company must build and maintain separately. Instead, it provides a shared, structured framework that exposes capabilities to AI models in a consistent way. This foundation allows AI systems to act more like intelligent agents—by not only answering questions, but also taking action directly within the systems users rely on. Zoom currently supports MCP as part of its Custom AI Companion add-on, enabling organizations to build custom agents that connect with their unique data sources and applications, automating routine workflows and delivering tailored, accurate responses based on teams' actual working content.

How MCP Works in Practice

At its core, MCP is built on a client–server model. Think of the AI environment (like ChatGPT or Claude) like a client that wants to get work done, and external systems (like Jira, Confluence, or a database) as servers that publish what they can do. Each server provides a clear list of functions—like “search knowledge base” or “create task.” The AI client then decides, in the middle of a conversation, when and how to use those functions.

In other words, connecting AI with an MCP server is just like ordering at a restaurant. You (the AI) walk into a restaurant, receive a menu (a list of services or functions available via the MCP server), and you (the AI) tell it what you want, and it responds. There is no guesswork of what is or isn’t available—everything is provided in the menu up front.

The following sections provide additional details for how MCP works:

Step 1: Expose Capabilities Through Servers

Any system can run an MCP server. That server acts as the “menu,” publishing a structured description of the actions it supports. For example, Jira might publish functions like “search issues” or “update ticket.” Because these functions follow MCP’s shared format, the AI can understand them immediately without custom engineering.

Step 2: AI Acts as a Client

On the other side, the AI environment plays the role of the client. It reads the menu of available functions, remembers them, and decides which to call when responding to a user. This means the AI doesn’t need to be programmed ahead of time with thousands of possible integrations—it simply learns what’s available when the connection is made.

Step 3: Securely Pass Context

MCP also defines how context and permissions are passed along. This helps ensure that when the AI uses a function, it does so only within the boundaries of what the user is allowed to access. For example, if a user has permission to see only their team’s Jira tickets, MCP makes sure the AI respects that scope. This security layer is what makes MCP practical for enterprise use, where sensitive data and access controls are non-negotiable.

Why MCP Matters

By standardizing how AI connects to external systems, MCP removes the friction of custom integrations and ensures safety and consistency. Users benefit because the AI can not only answer questions but also take informed actions across a variety of tools in a way that feels seamless and secure.

Agent-to-Agent (A2A) Protocol: How AI Assistants Communicate With Each Other

In addition to MCP, Zoom will also use the Agent-to-Agent (A2A) Protocol as a shared language for collaboration between autonomous agents. A2A is an open standard that allows AI agents—potentially built by different vendors or running in different environments—to discover each other, share context, delegate tasks, and exchange results securely. Zoom plans to support third-party AI agents with AI Companion by using A2A to pull in context from your Zoom conversations and take action across other business apps on your behalf. (Third-party agent for ServiceNow Now Assist coming soon.)

If the Model Context Protocol (MCP) connects AI models to the tools and data they need, A2A connects the agents to each other. Together, the two standards form the backbone of interoperable, multi-agent systems, where intelligence isn’t confined to one model but distributed across a network of cooperating agents.

How A2A Works in Practice

At its core, A2A is built on the idea that each agent can act as both a client and a server. One agent may request help or delegate a task, while another may respond and execute that task. The communication between them follows a common structure and security model, so no matter who built the agents—or where they run—they can understand each other.

You can think of it like a team of specialists working together: each agent has a clear job, a résumé describing what it can do, and a shared way to hand off work to others.

The following sections describe this process step-by-step:

Step 1: Agents Publish Their Capabilities

Every A2A-compatible agent exposes a small, structured “Agent Card.” This card acts as the agent’s identity and capability profile—it lists what the agent can do (for example, “summarize text,” “schedule a meeting,” or “query data”), what formats it supports, and how it can be reached.

Because this card follows the A2A standard, any other agent can read it and immediately understand how to interact without needing custom code or configuration.

Step 2: Agents Discover and Connect

When one agent wants to collaborate, it looks up another agent’s Agent Card—often through a directory, registry, or a well-known endpoint—and establishes a secure connection. This process allows agents to find each other dynamically, even if they were built by different teams or organizations.

Discovery ensures flexibility: a task-planning agent can find a visualization agent, or a customer-support agent can locate a translation agent, all through standard discovery mechanisms.

Step 3: Agents Exchange Tasks and Results

Once connected, agents communicate through standardized task messages. A task message might include a request (“analyze this data set”) and a response (“here are the insights”). These exchanges can happen synchronously for quick operations or asynchronously for longer-running tasks.

A2A also supports streaming and notifications, so agents can send intermediate updates or partial results as they work—mirroring how humans might collaborate in real time.

Step 4: Secure Collaboration and Context Sharing

Each interaction between agents is authenticated and scoped by the user or system that initiated it, ensuring agents only access the data or capabilities they’re authorized to use.

This controlled exchange of context allows complex workflows—like one agent summarizing a document while another creates a follow-up action—without leaking information.

Why A2A Matters

By defining a universal way for agents to talk to each other, A2A unlocks a new layer of interoperability and composability. Instead of building massive, monolithic agents that try to do everything, A2A empowers organizations to design specialized agents—each focused on a specific domain—and have them collaborate through a shared protocol.

For enterprises, this means:

• Cross-vendor compatibility: Agents from different providers can interoperate securely.

• Scalable design: Teams can add or replace agents without re-architecting entire systems.

• Governed automation: Security, observability, and auditing are standardized from the start.

• Faster innovation: New capabilities can be introduced simply by publishing a new Agent Card—no rewiring of existing integrations.