Moving from intent-based bots to proactive AI agents

Source: OpenAI Date: 2025-03-27 URL: https://openai.com/index/zendesk

Summary

Title-only: A March 2025 case study (URL points to Zendesk) documenting the transition from rule-based, intent-classification customer service bots to proactive AI agents powered by LLMs. The intent-based bot paradigm — where bots match user inputs to predefined intents and scripted responses — is being replaced by agents that understand context, take multi-step actions, and proactively surface information without waiting for explicit user requests. Zendesk’s AI agent products are at the center of this shift in enterprise customer support automation.

Implications

The customer service AI transformation thread. The shift from intent-based to agent-based customer service is not incremental — it’s a different product category. Intent bots have fixed capability ceilings determined by the intents you define; LLM agents can handle novel requests, reason over policies, and take actions (issue refunds, update accounts, escalate appropriately) without scripting. The agent model’s ceiling is much higher, but so is its failure surface: agents can misunderstand context, take wrong actions, and hallucinate policy details in ways that scripted bots cannot. The March 2025 transition is a capability step change with attendant reliability risks.

The support automation economics. Zendesk’s positioning around proactive agents is also an economic story: an agent that resolves tickets rather than just routing them has a different ROI profile than traditional support software. If agents can handle 70%+ of support volume autonomously, the cost per ticket drops dramatically and headcount growth decouples from support volume. This changes the economics of customer support organizations and the competitive dynamics of support software — platforms that enable true agent automation gain at the expense of those offering better routing and analytics.