Genesys Unveils LAM-Powered AI Virtual Agent

Genesys introduces a new “agentic virtual agent” powered by Large Action Models (LAMs), which is designed to complete customer service tasks across enterprise systems rather than only respond to queries, according to a company release.

The new Genesys Cloud Agentic Virtual Agent is intended to handle multi-step customer requests by understanding a user’s goal, determining required actions and executing them across systems such as customer relationship management (CRM), billing and service operations platforms.

Most existing virtual agents are built using Large Language Models (LLMs), which generate human-like text and improve conversational quality. However, LLM-based systems are primarily designed to produce language, not reliably carry out complex operational tasks across multiple systems. As requests become more complex, they may require human intervention or fail to complete workflows.

The cloud software company said its system instead uses Large Action Models. Unlike LLMs, LAMs are designed to execute structured actions in software environments. In practical terms, this means the virtual agent is built to carry out tasks rather than only providing information.

Hayley Sutherland, research manager for conversational AI at IDC, said, “Handling complex customer requests requires AI that can plan and complete multi-step actions across systems, while remaining predictable and auditable. She added that combining autonomous execution with governance reflects where the self-service market is heading.”

The company cited a Gartner report stating that the average self-service success rate is 22%, while 46% of leaders surveyed identified improving self-service as a top-three priority for 2026.

Genesys said the new system functions as a coordination layer across enterprise systems, applying governance controls and predefined guardrails. According to the company, this is intended to make actions traceable, policy-aligned and auditable.

The development follows a partnership between Genesys and Scaled Cognition. Genesys Cloud now incorporates Scaled Cognition’s APT-1 Large Action Model, which the companies describe as built for deterministic, action-based execution. The stated goal is to reduce issues such as “hallucinations,” where AI systems generate incorrect or unsupported outputs.

“In the enterprise, 80% accuracy is not enough for automation,” said Dan Roth, co-founder and CEO of Scaled Cognition. He said large language models are built to generate text, not carry out tasks, which can lead to errors or systems drifting from company policies. “For automation to work in real business settings, reliability matters more than intelligence,” he added, saying the combination of Large Action Models and Genesys Cloud’s governance aims to improve consistency and control.

Genesys said organisations including M&T Bank, Banco Pichincha, a Fortune 500 healthcare company and a Fortune 50 North American retailer are exploring the capability.

The company also said customers using the system would not need to restart requests as issues become more complex, as the virtual agent is designed to manage changing conditions during a single interaction.

Olivier Jouve, chief product officer at Genesys, said autonomous AI in customer service must incorporate controls and transparency.

Genesys added that governance features include action-level explainability, audit logs and configurable permissions through its AI Studio tool, allowing enterprises to define rules and compliance parameters.

The company said it plans to add support for open standards such as Agent-to-Agent (A2A) and Model Context Protocol (MCP), which are designed to allow different AI systems and enterprise tools to exchange information while maintaining oversight.

Genesys said the Agentic Virtual Agent is expected to be generally available globally in the first quarter of its fiscal year 2027, which runs from 1 February 2026 to 30 April 2026.