Zapier Agents is now AI by Zapier

Zapier Agents' features—prompt, tools, trigger, and permission to act—are now consolidated into a single AI step within the Zap editor, replacing the standalone product. This integration allows AI steps to leverage Zapier's existing triggers, filters, paths, and run history, eliminating the previous limitations of isolated agent workflows. Users can now build more cohesive automations by combining AI reasoning with standard Zapier actions in a single workflow.

The new AI by Zapier step introduces granular controls not available in Agents, including per-tool approval settings and model tier selection across Standard, Advanced, and Premium options. Tasks are billed uniformly under Zapier's task-based pricing, replacing the uncertainty of standalone agent billing. The platform also provides clearer visibility into costs per run through Zap History, with options to cap usage and pause runaway processes after 75 tasks.

AI by Zapier enhances workflow predictability by separating reasoning tasks from deterministic actions, such as letting AI analyze data before a human approves client communications. Outputs are structured as clean, mappable fields like categories or sentiment scores, reducing the need for downstream parsing. The system also auto-detects required tools for agent operations, streamlining setup while allowing manual adjustments if outputs are unsatisfactory.

Migration from Zapier Agents to AI by Zapier is designed to be straightforward, with eligible agents displaying a banner guiding users to a step-by-step guide. While some capabilities like on-demand runs can be mimicked via webhooks, the new system offers improved observability and control. Users can monitor AI step runs in Zap History, including tools used, inputs, outputs, model tier, and task consumption, mirroring the interface for traditional Zaps.