Zoom is advancing its AI Companion from a simple assistant to an active agent capable of handling tasks independently. This evolution, centered on reasoning and memory, aims to automate complex workflows, enhancing productivity within Zoom’s platform.
Zoom’s AI Companion Takes Charge with Autonomous Capabilities
Zoom’s AI Companion is stepping beyond basic assistance. The latest update introduces agentic skills that enable it to manage tasks, make decisions, and solve problems without constant user intervention.
These capabilities include independently handling multi-step processes, scheduling meetings, generating content, and adapting over time through learned experiences.
According to Zoom’s announcement, the AI Companion will help users get more done by executing tasks on their behalf, managing multi-step actions using reasoning and memory to make decisions, solve complex problems, and learn over time.
This reflects Zoom’s broader strategy to shift the AI from a reactive tool to an autonomous assistant that anticipates and initiates actions.
The new features will be integrated into Zoom Workplace, with a dedicated Tasks tab allowing users to assign and track activities. Additionally, a voice recorder function, set to launch on Zoom’s mobile app, will provide transcription and summarization for in-person meetings.
Another upcoming feature, Live Notes, is expected to roll out in May, offering real-time meeting summaries to reduce manual note-taking.
Custom AI Agents and Business-Centric Integrations
To further tailor the capabilities of tis AI Companion, Zoom will introduce a Custom AI Companion add-on, available from April for $12 per user per month.
This feature will allow organizations to craft specialized AI agents and skills, enabling the automation of unique workflows across departments. Whether streamlining sales processes or handling HR service requests, these custom agents aim to address specific operational challenges.
Zoom’s strategy also includes specialized agents within its Zoom Business Services. These agents are designed to resolve complex service issues by leveraging memory and reasoning capabilities.
They will handle multi-step tasks, deliver contextual conversations, and follow through on resolving issues without human intervention. As Zoom described, these agents will “leverage memory and reasoning skills to deliver empathetic and contextual conversations, resolving complex issues from start to finish.”
The platform’s flexibility is extended further through AI Studio, enabling organizations to build and deploy custom virtual agents. Additionally, Zoom’s architecture supports integration with third-party systems, such as ServiceNow AI Agents, broadening its utility across industries.
Privacy Safeguards for Responsible AI Use
Zoom is also emphasizing its commitment to responsible AI practices. Concerns about data privacy have become more pronounced with the advancement of AI technologies.
In response, Zoom has reaffirmed its privacy commitment for AI development stating that it does not use customer audio, video, chat, screen sharing, attachments, or other communication content to train its or third-party AI models.
This stance is a direct response to growing scrutiny over data usage in AI training. By reinforcing these privacy measures, Zoom aims to position itself as a provider that prioritizes user trust and ethical data handling.
While this approach may limit some avenues for AI optimization, it aligns with rising enterprise demands for stringent data security standards.
Zoom’s In-House AI Model Research
Driving these enhancements is Zoom’s own research on AI models and their efficient use, such as its recently unveiled novel Chain of Draft (CoD) prompting technique.
CoD is designed to tackle one of the primary challenges of AI operations: efficiency. Traditional methods like Chain of Thought (CoT) prompting, while improving AI reasoning accuracy, often result in lengthy, resource-intensive outputs.
CoD flips that approach. By emphasizing concise, structured responses, CoD reduces token usage by up to 92% and slashes operational costs by 90%. Rather than articulating every step of reasoning, the AI focuses on essential intermediate results, mimicking how humans jot down brief notes when solving complex problems.
Benchmarking tests using models like OpenAI’s GPT-4o and Anthropic’s Claude 3.5 Sonnet have demonstrated that CoD can achieve similar levels of accuracy as CoT while significantly lowering computational demands.
For instance, in arithmetic reasoning tasks conducted with the GSM8K dataset, CoD achieved near-equivalent correctness while reducing token consumption to just 7.6% of what CoT required.
However, CoD isn’t universally effective. Testing revealed that smaller models—those with fewer than 3 billion parameters—struggled to match the efficiency gains seen in larger models.
For example, tests on Zoom’s own small language model, Zoom-SLM, highlighted that “small models require fine-tuning with CoD-style data to fully leverage its efficiency benefits.” Without this fine-tuning, smaller models showed more significant accuracy gaps, which may limit CoD’s immediate applicability in lightweight AI systems.
What’s Next for Zoom’s AI Companion
Zoom’s roadmap for the AI Companion doesn’t stop here. Alongside the Custom AI Companion add-on, the platform will introduce new personalization features.
This includes Custom Avatars for Zoom Clips, allowing users to generate AI-powered avatars for video content. This feature is part of the broader customization effort launching in April.
Additionally, Live Notes will provide real-time, automated summaries during Zoom meetings starting in May, aiming to minimize manual note-taking. Meanwhile, the voice recorder feature for mobile will assist in documenting and summarizing offline meetings, ensuring comprehensive coverage across work environments.
Zoom’s broader AI vision was recently outlined in its blog post, which emphasized the company’s ambition to develop AI tools that “radically transform your workday for the better.”
The company’s approach underscores the balance between efficiency, flexibility, and privacy in driving its AI strategy.
Zoom’s shift towards agentic AI reflects broader industry trends focused on automating complex processes and reducing operational friction. While the efficiency gains from CoD prompting appear promising, the technology’s long-term value will depend on its adaptability across varied business scenarios.
