Shift from Intent-Based Bots to Proactive AI Agents

Zendesk has partnered with OpenAI to unveil a new class of AI agents that go beyond existing bots based on simple intent classification, instead planning and executing solutions to problems on their own. Existing models were limited to predicting the customer's intent and following a fixed scenario, but the new agents lead the conversation with the goal of achieving actual resolution.

These agents are built on models such as GPT-4o, and perform complex tasks through the following multi-agent architecture:

• Task identification agent: Identifies the user's requirements through conversation and asks questions to clarify ambiguous points.
• Conversational RAG agent: Maintains multi-turn conversation context while retrieving specific information such as region-specific policies.
• Procedure compilation agent: Converts business rules written in natural language into structured flows to secure control.
• Procedure execution agent: Executes actual system tasks through API calls and workflow triggers.

These technical advances make it possible to reduce setup time from days to minutes and push automation rates up to around 80%. Through a hybrid model where businesses define procedures in natural language, companies can strike a balance between AI autonomy and adherence to business guidelines.