The short answer
If you used Zapier Agents, the standalone agent builder is going away in favor of a unified AI by Zapier step. In practice, that means the agent's tool calling, reasoning, and autonomous action now live inside a normal Zap — so it can finally use the triggers, filters, paths, and run history that the rest of Zapier already has[1].
What changed
Zapier Agents was built from four ingredients: a prompt, a set of tools, a trigger, and permission to act. The problem was that the bundle sat outside the platform, so an agent couldn't tap into Zapier's triggers, filters, paths, or its run history[1].
The same four ingredients now live in a step you drop into any Zap, running on the infrastructure that already executes millions of tasks a day. That unlocks access to the rest of the Zapier platform for the first time[1].
New controls you get
Moving the agent into the editor also added capabilities the standalone product didn't have[1]:
- Mix agentic and deterministic steps. Let the AI decide which record to update, then use a normal Zapier action to send a confirmation email exactly the way it should read.
- Gate risk per tool. "Require approval before running" toggles let you force a human check before an irreversible action, while still letting the agent read data autonomously.
- Structured output. Define the fields you want back — a category, a sentiment score, a subject line — and they arrive as clean, mappable outputs.
- Auto-detected tools. Zapier suggests tools the agent might need, removing the guesswork.
- Full visibility. Every run lands in Zap History with the tools called, inputs and outputs, the model tier used, and the tasks consumed.
Task-based pricing
AI steps now bill as tasks, on the same task-based pricing as the rest of Zapier. You can use default models for free, while other model tiers carry different multipliers — so a single AI step can cost 1, 3, or 5 tasks[1].
You can also set a cap. If a single run crosses 75 tasks (adjustable later), it pauses for a human instead of looping quietly — Zapier's answer to the "runaway agent" concern that followed the standalone product[1].
Model choice
Agents previously didn't let you pick a model. AI by Zapier adds Standard, Advanced, and Premium tiers — or you can name a specific model — so you control cost and capability per step[1].
Migrating from Zapier Agents
You don't have to rebuild from scratch. Migration recreates your agent as an AI by Zapier step — the same goal and tools, now inside a Zap where you can wrap deterministic guardrails around it. Most settings carry over. If your agent is eligible, you'll see a banner leading to a step-by-step guide[1][2].
A few capabilities are still rolling out. Notably, some knowledge sources aren't yet supported in AI by Zapier — HubSpot, Asana, Zendesk, Zoom, and Jira — though there are usually clean workarounds[1].
Frequently asked questions
What happened to Zapier Agents?
It's now "AI by Zapier." The agent's tool calling, reasoning, and autonomous action live in a single AI step inside the Zap editor, with access to the rest of the platform[1].
Does AI by Zapier cost extra?
AI steps bill as tasks. Default models are free; other model tiers carry multipliers so an AI step can cost 1, 3, or 5 tasks. You can cap tasks per run to avoid surprise spend[1].
Can I still choose a model?
Yes — a change from Zapier Agents, which didn't let you pick. You can choose Standard, Advanced, or Premium tiers, or name a specific model[1].
Sources
- Zapier — "Zapier Agents is now AI by Zapier" (Ruchira Kulkarni, Oct 6, 2026) — zapier.com
- Zapier Help — "Migrating from Agents to AI by Zapier" — help.zapier.com
- Zapier — "AI by Zapier guide" — zapier.com