Inside Zoom’s Ai Evolution: From the basic meeting tools to the agent of the agent supported by LLMS and SLMS

Enlargement It has become a familiar name during the epidemic, as the distant work has become almost overnight.

Although the company was only synonymous with video conferences, it was adopting a sophisticated infrastructure of artificial intelligence over the past few years with the aim of redefining the productivity of the workplace. Although video conferences are important and the cornerstone is still in Zoom’s business, there is a lot now also, thanks to artificial intelligence.

Moving from the meeting to the teacher

Everyone knows that zooming is a technique for meetings. But what is the meeting?

In the context of the work, it is certain that there can be unimaginable meetings, but these extreme values ​​must be. Meetings should lead to something, whether it is an element in the procedure or some other teacher.

“In the Airec AI era, technology finally reaches the point where we can turn from the meeting to the teacher,” said Zoom Cto XUEDong (xD) Huang in Venturebeat in an exclusive interview.

Today, Zoom announces an aggressive strategy Aic Agency that includes a series of new services. The update provides the possibilities of the agent, which is to transfer meetings from communications events to the workflow towards work, as well as a new studio for Amnesty International that allows institutions to establish allocated artificial intelligence agents.

The hidden technical development behind AI AIC’s Zoom

Before joining Zoom, Huang spent 30 years in Microsoft, where she worked on speech techniques as well as Microsoft Azure Openai. He did a lot of lessons learned from that experience when he joined Zoom in 2023.

Under the supervision of Huang, Zoom calmly started building Amnesty International’s structure designed to facilitate tasks rather than summarizing the talks. The enlargement was publicly announced a Partnership with man In May 2023 – but this is not the only Grand Language model (LLM) used to be enlarged.

while Microsoft teams It generally depends on Openai via Microsoft Openai Azure, and Google meeting Google Gemini, Zoom followed an indisputable approach with LLMS.

Huang explained that when Zoom launched his first Repetition of his AI’s companion in 2023, he did not depend on any one LLM. Instead, the company started with a federal approach, using multiple LLMS including the SLM Small Language model (SLM).

“We have made a partnership with the best models there, including Openai and Anthropic, but we also built a very dedicated parameter language model for our billion,” said Huang.

COMPANION uses a federal approach where the smaller zoom model is used in conjunction with larger and pioneering language models in the industry. The smaller model initially evaluates the input and treats it at first, then the partial results are passed to larger models for the production of the final output. This approach allows zooming to take advantage of the strengths of both the smaller and customized model and the largest and most powerful models, while reducing costs and improving performance.

How is the small language model at the Jetter of Zoom’s Agentic Ai Center

Perhaps the most interestingly interesting aspect of Zoom’s AI’s strategy is its focus on SLMS. Instead of following the industry direction to distilize the smaller models of large models, Zoom built the parameter model 2 billion completely from zero.

The technical advantage of this approach becomes clear when customizing for specific areas. “When you deduct, it takes more effort, it is difficult to direct a larger ship,” Huang explained.

As it turned out, the ability to customize the small model is a decisive component to develop a specific workflow for the artificial intelligence agent. Looking at the future, Zoom SLMS finally imagines its operation directly on the user devices, allowing both the best privacy and the most customized experiences.

AI Companion 2.0: Agency AI transforms meetings into prominent monuments

At the heart of the Zoom updates there is AI Companion 2.0, which converts the AI’s capabilities from Zoom to meet the support to the entire agent’s jobs. With 2.0, Zoom is evolving from auxiliary to AI AIC that is able to think, memory and carry out tasks.

The advanced AI craft can now perform multiple -step procedures on behalf of users, organize tasks such as scheduling meetings, create videos and create documents.

The main updates include:

• Agent skills: Calendar management, generation of a clip, advanced writing assistance;
• Task managementAutomatic disclosure of work elements of meetings and conversations;
• Meet the improvementsBusiness schedules that work from artificial intelligence, live notes and audio recording;
• Create documentsAdvanced references and automatic spreading of the splash in enlargement documents;
• Virtual agents: Self -service capabilities for customer service with both chatting and sound support;
• Industry solutionsSpecialized tools for frontal lines workers, healthcare professionals and teachers;
• Zoom Drive: A new central warehouse to meet assets and productivity documents;
• Custom Avatar: The video created by AI to create the presentation clips.

Most of the features will be released between March and July 2025. While the standard artificial intelligence companion does not include any additional cost for paid users, specialized agents and customized formations will require additional fees.

“The most important aspect for us for the artificial intelligence agent is to enable the flow of information directed towards work,” said Huang. “What this means is that when you have a meeting, the task of the procedure will flow into documents, chat, or in other measures you have to take.”

AI Studio: Building agents for institutions

While Zoom provides a lot of AI Agency capabilities outside the box for users, Huang realized that institutions often need more custom options.

This is where AI Studio comes, allowing companies to create dedicated Amnesty International agents designed to meet specific business needs. It can be deeply combined with the company’s knowledge and workflow.

For example, HUANG detailed one practical application of human resource policy. Institutions can use artificial intelligence studio to download all internal human resources policy documents. The artificial intelligence companion will then be trained in this company’s human resource policy information, allowing it to be accurately answering employee questions about human resources guidelines and procedures.

IT officials can also use artificial intelligence studio to connect the associated with other internal knowledge rules, such as information technology support documents. The goal of this is to enable companies to create integrated intelligence agents deeply with their operations, data and workflow, and converting the artificial intelligence companion into a dedicated and valuable production tool.

The effects of the AI ​​decision makers of the institution

For technician decision -makers who hold the AI’s productivity, Zoom’s approach offers many distinctive considerations compared to Microsoft alternatives, Google and other sellers.

The information flow model directed at work may suit institutions where meetings are the basic means of cooperation, and where the completion of the task is disabled by dividing the information through different tools. Companies of high sizes of meetings may find a special value in Zoom’s ability to connect talks to subsequent procedures.

In addition, the unified artificial intelligence approach that combines SLMS and LLMS provides a brown structure – it is likely to offer better economies while maintaining quality. Since artificial intelligence costs become an increased concern for institutions that define artificial intelligence applications, this balanced approach can have an effect.

For institutions looking to lead the road in adopting artificial intelligence, Zoom’s development from the meeting tool to comprehensive productivity platform offers valuable lessons on how to build on current strengths instead of just adding Amnesty International to the current workflow. By taking advantage of its dominance on its meeting to re -imagine the entire productivity experience, the magnification explains how the field experience can be combined with the capabilities of artificial intelligence to create solutions that address specific work problems instead of artificial artificial intelligence assistants.