Ten months ago, Zoho Corporation announced Zia LLM with general availability targeted for late 2025. At Zoholics USA 2026 in Houston this week, the company showed what availability actually looks like in practice. The Agent Marketplace is live with more than 25 deployable agents. Zia Agent Studio is fully prompt-based, with access to over 700 built-in actions. Agents can now be provisioned as Digital Employees, operating under the same permission structures as human users. The agentic layer Zoho has been building for years is no longer a promise on a slide.

That distinction matters more than it sounds. Most enterprise software vendors have announced agentic AI in 2026. Zoho has deployed it inside a platform that already handles CRM, finance, human resources, collaboration, and IT operations under a single roof, across 55 applications, without routing customer data through third-party AI providers. The infrastructure constraint that limits most agentic deployments is not a constraint for Zoho. They built the stack to avoid it.

I covered the foundational architecture at ZohoDay 2026 in February and the original Zia LLM announcement last July. Zoholics is where the rubber met the road.

From early access to general availability

The July 2025 Zia LLM launch put everything into early access with a late-2025 general availability target. By the Q1 2026 CRM update, Zia was active inside Zoho CRM, powering Smart Prompts, record summarization, and template generation. Zoholics USA confirmed the broader rollout status across the portfolio.

The 32B model Zoho signaled in July 2025 has not been publicly confirmed as shipped. Zoho's stated posture was to scale model sizes through 2025 before general availability. No announcement has confirmed that timeline was met for larger parameter variants.

What Digital Employees actually change

The most operationally significant announcement at Zoholics was not the agent count or the marketplace. It was the Digital Employee model. When an agent is provisioned as a Digital Employee, it inherits the same access controls the organization has already defined for human users. It does not get a separate permission set. It does not bypass existing governance. Admins can run behavioral audits, performance analysis, and impact reviews on every Digital Employee, with the same traceability expected of any human in the workflow.

For IT leaders evaluating agentic AI, this matters because the failure mode in most deployments is not the model. It is accountability. When an agent takes an action, sends a communication, or modifies a record, who approved it? Under what authorization? Most vendors have not answered this directly. Zoho has, by tethering agent behavior to the same governance layer the organization already manages.

The MCP server and what it signals

Zoho's adoption of the Model Context Protocol is worth reading carefully. The MCP server exposes action libraries from more than 15 Zoho applications to any MCP-compliant client, including external tools like GitHub Copilot, under the customer's existing permission framework. Third-party agents can tap into Zoho data and take actions without requiring custom integrations.

Raju Vegesna, Zoho's Global Chief Evangelist, put it plainly: Zoho is not trying to be the only agent in the room. The MCP server makes Zoho a platform other agents can use, not a walled system that requires Zoho agents exclusively. The planned Agent-to-Agent protocol extends this further, enabling Zia Agents to collaborate with agents built on other platforms entirely.

That is a specific bet. Zoho is wagering that openness at the interoperability layer, combined with data sovereignty at the infrastructure layer, is a more durable position than trying to lock customers into a proprietary agent ecosystem. Given that the MCP standard is gaining adoption across the industry, the timing is reasonable.

Three prebuilt agents worth understanding

The Agent Marketplace launched with functional agents built on Zia Agent Studio and available for immediate deployment. Three illustrate the range of the initial use cases. The Revenue Growth Specialist scans existing customer relationships for upsell and cross-sell signals, recommending marketing approaches per customer. The Deal Analyzer runs win probability, next-best-action, and follow-up suggestions on open pipeline. The Candidate Screener ranks job candidates against role requirements, skills, and experience without manual sorting.

These are not experimental. They are operational agents designed for roles that exist in every mid-market and enterprise organization. The deal analysis use case in particular has a direct revenue impact: a small percentage improvement in conversion compounds significantly at scale. The fact that Zoho is not charging separately for these agents, at least through this phase, makes adoption easy to justify.

What has not been answered

Pricing remains an open question. Raju Vegesna said in July 2025 that Zoho was not charging for Zia LLM or agents until it had better visibility into usage patterns and operational costs. That posture has held. At some point, a pricing structure will emerge. Organizations that build workflows around Digital Employees and agent-to-agent handoffs should understand that the current cost model is not the permanent one.

The 32B model and the Reasoning Language Model are on the roadmap without confirmed delivery dates. The RLM in particular is the capability that would unlock logic-heavy workflows: multi-jurisdictional compliance, complex procurement negotiation, multi-step financial decisions. Until it ships, Zoho's agentic layer handles operational automation well but stops short of autonomous decision-making in regulated environments.

And the Agent-to-Agent protocol is still ahead. The MCP server handles external agents connecting to Zoho. A2A handles Zia Agents working with each other and with agents on other platforms. Multi-agent orchestration, the actual frontier of agentic enterprise software, requires both. One is live. One is not.