Organisations have to fully embrace artificial intelligence (AI) if they want to reap the benefits the technology has to offer.
The returns will be limited if they hold back due to concerns they may have about AI, according to Zoom CTO Huang Xuedong.
For example, these companies might refrain from sharing their data with AI, Huang said in a video interview with FutureCIO.
As a result, AI cannot "participate” and is unable to generate relevant insights that can help improve their operations, he said.
He likened it to patients who do not share information with their doctor, who then may not have the data to properly diagnose the medical problem.
“You have to be open. That’s the number one challenge,” Huang said, responding to a question on pitfalls organisations should avoid as they roll out AI initiatives.
“Enterprises are so worried about their data that they’re adding unnecessary concerns. Then there’s no way [for AI] to help employees be more productive,” he said.
As it is, 48% of IT and business leaders have pointed to data security as their top concern in AI deployments, revealed a study by Hitachi Vatara released in April 2025.
A separate report from Salesforce in October 2024 also found that just 11% of enterprise CIOs had fully implemented AI due to technical and organisational challenges led by security and data infrastructure.
Some 67% took a more calculated approach to rolling out AI than other technologies, citing security and data as the most common hurdles to implementation, the study highlighted.
In urging organisations to embrace AI, though, Huang underscored that this did not mean they should overlook the importance of safety and security.
Companies must take the necessary steps to ensure their data is secured and managed with all essential privacy and safety polices, he said.
This is what Zoom itself adheres to, he added.
“[But], if you’re not embracing AI and not willing to share your data, then there’s no way you can amplify your own capabilities,” he said.
Huang shared that he maintains a daily diary via Zoom’s AI Companion, logging in observations from work meetings and reflecting on how his day went.
He then uses the generative AI-powered tool to identify ways to improve, for example, by getting feedback and steps to deliver a more effective presentation.
It offers recommendations on how he can be a more effective leader, said Huang, who noted that this is possible because he chooses to open up to the AI tool.
“Embrace AI, share your data,” he said. “Whether you’re an engineer, manager, or leader, AI can help you if you open up. The most important thing is, there’s a human in the loop [and in control].”
This is essential because AI currently still does have hallucinations, he noted.
Companies need their employees to have critical thinking and be able to make the judgement call to differentiate good AI-generated information from gibberish, he said.
Moving workspaces towards agentic AI
As CTO, Huang leads Zoom’s AI development strategy and services, including its Zoom AI Companion and federated AI architecture. Before joining the company in 2023, he spent three decades at Microsoft where he last served as CTO of Azure AI and was its Technical Fellow.
Like others in the industry, Huang, too, believes agentic AI will be integral in enterprise environments as AI continues to advance.
“It’s about task completion,” he said, pointing to AI Companion as a key component in how Zoom aims to drive the next phase of the GenAI era.
Beyond writing summaries of meetings and email messages, he added that AI Companion can support tasks and identify what is important, and deliver these in a format users want.
He further believes that customised small language models (SMLs) -- running on Zoom’s federated AI architecture -- will be the “backbone” of AI agents, because they run more efficiently and are comparable to the more expensive LLMs (large language models), which currently are commonly used.
In a February 2025 post, Huang said this strategy will enable the vendor to deliver AI agents as “cost-effective, high-performing solutions” for its customers.
Zoom’s SLM is trained on 6 trillion tokens of multilingual data and 256 Nvidia H100 GPUs, and can be customised for specialised tasks, acting as specialised agents to carry out tasks, he said.
He further touted the vendor’s federated AI approach as integral, enabling it to tap multiple customised models instead of depending on a single large model.
“In support of our training efforts to refine task completion quality, our federated approach to AI takes advantage of many closed- and open-source advanced LLMs working together for better results,” he said. “This is in contrast to other providers who are tied to specific LLMs. For example, Microsoft Copilot has relied on GPT-4 and Google has relied on Gemini.”
Three cobblers working together can be as intelligent as Zhuge Liang, Huang quipped, as he explained the merits of Zoom's federated AI approach in the video call with FutureCIO.
He was referring to a popular Chinese proverb, which illustrates how the collective knowledge of three cobblers can achieve a similar level of intelligence as Zhuge Liang, a Chinese stateman who was a renowned strategist and inventor during Three Kingdoms period.
Huang believes Zoom’s move to distribute workloads across customised SLMs, whilst also leveraging major LLMs, will enable AI agents to be optimised to perform specific tasks and to operate more cost efficiently, at scale.
Apart from its own LLM, Zoom taps various other foundational models including Meta Llama, OpenAI, and Anthropic.
With customised AI models the way forward, Huang said Zoom’s enterprise customers also can train AI Companion with their own company’s data, including HR and IT policies.
This option is available as a paid upgrade, he said, adding that companies are willing to fork out the extra for more accurate queries and value add.
The data knowledge and customisation will be specific to the customer, and will not leave their network, he noted.
