Inside Zoom’s AI evolution: From basic meeting tools to agentic productivity platform powered by LLMs and SLMs

Enlarge The name of the home was the name of the remote work during the pandemic.

Although the company once synonymously synonymous with a video conference, it was a quiet developed AI infrastructure to redefine the workplace in the past few years. The video conferencing is important and there are also more to the corner of the enlargement.

To move from the meeting to the stage

Everyone knows that zoom is a technology for meetings. So what is the session?

In a business context, no doubt may have no purpose, but they must be excluded. Meetings This should cause something to happen or not a movement element or another stage.

“The agent reaches the point where the Technology can finally become a stage in the AI ​​ERA,” he said.

Today, an aggressor announces an aggressive agent AI strategy that includes a number of new services. The update offers agent opportunities that link to work flows to the activities of communication events to the activities of communication events, special AI studio to create special AI agents.

Hidden Technical Evolution Behind the Zoom’s Agentic AI

Before joining the ZOOM, Huang has been 30 years in Microsoft, as well as in Microsoft in Microsoft’s AZURE Openai service 30 years. When he raised in 2023, he made many lessons learned from that experience.

Zoom in the direction of Huang, silently, began to build a AI architecture designed to simplify the tasks to summarize conversations. Zoom is clearly announced Partnership with anthropic In May 2023 – but not the only large language model (LLM) used when raising.

While Microsoft teams In general, Microsoft Openai relys on Openai through Azure service and Google meets Google has been supported by Gemini, a agnostic approach to Zoom LLMS.

Huang explained that when the zoom was launched by the first ITERATION OF AI HOME in 2023, it was not based on any LLM. Instead, the company began with a federal approach, using more than one LLM, including its own specially established small language model (SLM).

“We also built our best models there, including Openai and Antopic, but our own highly special 2 billion parametre language model,” Huang said.

Zoom’s AI Companion uses a federal approach to the larger zoom model, which is used in conjunction with industry-leading language models. The small model first acquires partial results after entry and partial results and is transmitted to larger models to produce the final result. This allows you to enlarge an approach, both small and special models and larger models, more powerful models, increasing the costs and benefit from larger, more powerful models.

The small language model is in the center of Zoom’s Agentic AI Travel

Probably the most technically stakeholder of the ZOOM’s AI strategy is the attention of its SLS. The elimination of small models has completely built 2 billion parameter models compared to larger and magnifying the industry.

The technical advantage of this approach is clear when individual domains are individually. “When customizer, the more effort, it is difficult to handle a greater ship,” said Huang said.

When leaving, the ability to correct a small model, a special agent is a critical component for the development of AI workflows. Looking at the SLOM, which allows users directly on user devices, which allows you to enhance, enlargement, as well as better privacy and more individual practices.

AI Companion 2.0: Agent AI converts meetings into stages

The AI ​​Companion is AI Companion 2.0 that changes ZOOM updates to the heart of Zoom’s updates from responding to full agent functions. 2.0 is developing from the assistant to the aid of the zoom assistant to AIT, which is able to perform memory and task.

Improved AI companion can now perform multi-step actions on behalf of users, can spoil tasks such as planning plans, video clips and create documents.

Basic updates include:

• Agent skills: Calendar management, clip generation, advanced writing aid;
• Task management: Automatic detection of movement items from meetings and conversations;
• Meeting Expansion: AI-powered agenda, live records and sound records;
• Creating a document: Advanced references and automatic information table generation in zoom documents;
• Virtual agents: Self-service opportunities for customer service with both conversations and sound support;
• Industrial solutionsSpecialized tools for Frontline staff, healthcare workers and teachers;
• Zoom driver: New central depot for assets and productivity documents;
• Special avatarsAI created video avatars to create presentation clips.

Most of the features will be broadcast between March-July 2025. Standard AI companion does not have additional costs for paid users, specialized agents and special configuration will require additional payment.

“The most important aspect of the agent of AI is indeed providing movement-oriented data flow,” Huang said. “What does this mean that when you have a meeting, the action task will flow in documents or other actions you have to take.”

AI Studio: Establishment of special agents for the enterprise

Zoom, while many different agents for users provide a lot of AI capabilities, Huang, who are outside the box for users, has often known that businesses need more specific choices.

The AI ​​studio allows companies to create special AI agents in accordance with specific work needs. These can be deep integrated with the company’s special knowledge and workflow processes.

As an example, a practical application for Huang Human Resources Policy details. Enterprises can use the AI ​​studio to download all internal HR policy documents. The AI ​​spouse will be taught to special HR policy information, which allows you to accurately respond to the questions of employees on the HR rules and procedures.

As dog managers, as well as companion support documents, they can use AI studio to connect to other domestic knowledge bases. Goals, companies are to allow them to create deep combined AI agents with their processes, data and work flows, transforming their processes, information and workflows, a special and valuable productivity tool.

Effects for the company’s decision makers

AI solutions for technical decisions that evaluate productivity offer several different considerations compared to alternatives of ZOVROSOFT, Google and other vendors.

The action-oriented information stream model may be better in the fields of better costume institutions where meetings are an environment of primary cooperation and deterring information in different means of completion. Companies with a high meeting can find a certain value in the ability to combine talks on subsequent actions of the zoom.

In addition, the federal AI approach offers an architecture connecting SLMS and LLMs – offers a better economy while maintaining quality. When AI costs become a growing concern for enhanced enterprises that measure AI applications, this balanced approach may be reputable.

For enterprises wishing to lead an AI adoptive way, the evolution of the meeting in the extensive productivity platform, instead of adding to existing work flows, offers valuable lessons in the establishment of existing power. Using a topic meeting to redistribute the entire productivity experience, the magnification demonstrates how to combine the AI ​​capabilities to create more specific work problems than extensive AI assistants.