At ZohoDay26 in Austin, I had the opportunity to sit down with Ramprakash Ramamoorthy (Ram), Head of Zoho Labs, for a conversation on how Zoho is approaching AI, agents, infrastructure, and platform innovation at a moment when the enterprise software market is being reshaped by generative AI.
What stood out most was the discipline behind Zoho’s strategy. In a market full of companies racing to attach AI to everything, Ram described a more measured approach: build around real problems, keep privacy at the center, and deploy the right level of intelligence for the job.
Watch the full interview below to hear Ramprakash Ramamoorthy discuss Zoho’s vision for the future of enterprise automation.
Ram explained that Zoho Labs is not a conventional research lab. Rather than focusing on innovation for its own sake, the team works on common challenges that span Zoho’s broad portfolio of products, co-developing solutions that can be reused and scaled across the company. The goal is not to produce research artifacts alone, but to create technologies that reach production and deliver value across the Zoho ecosystem.
That mindset also defines Zoho’s AI journey. Long before the current LLM (Large Language Model) wave, the company was applying AI to enterprise use cases such as anomaly detection, document processing, translation, grammar assistance, and workflow intelligence. As Ram described it, Zoho’s evolution in AI has been steady and cumulative, moving from traditional machine learning and NLP (Natural Language Processing) toward today’s broader AI stack of LLMs and agents.
A central theme in the discussion was model right-sizing. Zoho’s position is that not every enterprise problem requires a massive language model. In fact, Ram argued that many business use cases are better served by smaller, purpose-built models that are more efficient, more explainable, and less compute-intensive. In Zoho’s architecture, LLMs often serve as a summarization or interaction layer, while much of the real intelligence underneath is driven by specialized models tuned to specific tasks.
That philosophy also helps explain why Zoho chose to build its own LLMs. Rather than pursuing large models for branding value alone, the company sees model development as a way to deepen its technical capabilities and create AI that performs well within business context. Zoho’s current family of smaller models is intended for enterprise tasks such as summarization, paraphrasing, and contextual assistance, where relevance, efficiency, and privacy matter more than general-purpose breadth.
Privacy was another major point of emphasis throughout the interview. Ram made clear that Zoho’s AI systems are designed around strict organizational and user-level boundaries, ensuring that models and agents only access information that a given user is already authorized to see. He also reinforced one of Zoho’s longstanding differentiators: the company does not use customer data to train general AI models. In an era when many enterprises remain cautious about AI adoption, that privacy-first posture continues to be a meaningful part of Zoho’s value proposition.
One of the most compelling parts of the conversation centered on AI agents. Ram shared how Zoho tested agentic capabilities internally, including within its legal help desk workflows. The lesson was revealing: success did not come simply from putting a chatbot in front of users. Instead, real gains emerged when AI was connected to the deeper operational fabric of the business—availability data, customer history, organizational structure, permissions, and workflow logic. In that environment, Zoho was able to improve ticket routing and operational efficiency, illustrating that agentic AI works best when built on top of an integrated system rather than a fragmented stack.
We also discussed AppOS, one of Zoho’s more important platform initiatives. Ram described it as a development framework for business applications that helps organizations move beyond app sprawl and disconnected point solutions. As low-code, AI-assisted development, and “vibe coding” accelerate software creation, Zoho sees growing demand for a common platform layer that provides consistency in permissions, reporting, search, notifications, and business logic. In that sense, AppOS is not just about building applications faster; it is about building them with stronger governance and deeper system-level integration.
Infrastructure, too, plays a significant role in Zoho’s AI strategy. Ramprakash stressed that enterprise AI is not only about models and agents, but also about the underlying database, hardware, and data center architecture that supports them. He pointed to Zoho’s investments in its own infrastructure and its engineering relationships with companies such as Nvidia, Intel, and AMD as examples of how the company is trying to extract value from the full stack, not just the application layer.
Toward the end of the discussion, Ram offered a pragmatic perspective on the broader future of AI. He is clearly optimistic about the technology’s role in helping enterprises operate faster and more intelligently. But he is equally cautious about overstated narratives around AGI (Artificial General Intelligence) and full autonomy. His view was refreshingly grounded: today’s models are powerful tools for augmentation, summarization, and workflow improvement, particularly in bounded enterprise environments, but they are not a substitute for business context, human accountability, or sound system design.
Overall, my conversation with Ram reinforced a key takeaway from ZohoDay26: Zoho is not trying to win the AI moment through noise. It is trying to win it through architecture, discipline, and long-term execution. In a market increasingly defined by questions of trust, context, and operational value, that may prove to be one of the more durable strategies in enterprise software.
To explore how contextual, privacy-first AI can transform enterprise workflows, connect with our experts today.
