Zapier Agents Review: Pricing and the Move to AI by Zapier

This Zapier Agents review explains what to do with an existing standalone agent and where to build a new one. Zapier is moving Agents into AI by Zapier, where an agent can call tools inside a step in the Zap editor. Its migration guide listed no shutdown date for standalone Agents when checked on November 7, 2026.

For a new workflow, start with the in-Zap option if you already use Zapier. For an existing agent, compare what each version retrieves, what it writes, and how much it consumes before switching. This review draws on Zapier’s documentation and uses an illustrative support workflow.

Zapier Agents Review and How It Compares With Zaps

A Zap starts with a trigger and follows the steps and branches you configure. Use it when the rule is clear, such as routing every order above a fixed value to an approval queue. When the input is an email describing several problems in one paragraph, an AI step can interpret the message before the workflow applies a routing rule.

Standalone Agents let you describe the job in natural language and configure the app actions the agent can use. An agent can be set up to start from an app event, a schedule, or a request, depending on the trigger. In AI by Zapier, you can put the trigger, the AI step, and the follow-up steps on the same canvas. The AI step can return named output fields that later steps can use.

The practical question is how much freedom to give the model. Letting it choose which sources to consult may help when requests vary. If the task only requires classifying a short message, a prompt-only step can return a category for an explicit branch. Teams comparing visual agent builders can see how another platform handles approvals in SD’s Gumloop review.

What a Support Triage Workflow Looks Like

Suppose a support team wants to classify incoming returns requests and prepare replies. A Zap can capture the original message and pass the message text and ticket ID to the AI step. The step can search an approved policy source and return a category, the policy reference, and a draft reply. If the message omits the order number or the policy source gives conflicting answers, route the ticket to a reviewer with the original message attached.

Give the agent a narrow job and restrict what it can change. For example, put the draft reply in an internal queue and use a separate step to send it after approval. Keep the ticket ID fixed to the incoming record and validate the category against a short allowed list. These checks make the workflow easier to debug when the policy search succeeds but the proposed outcome is wrong.

Test with several versions of the same request, including one that sounds urgent and another with an attachment the workflow cannot read. Compare the policy reference and route against the result a support lead expects. A well-written reply is useful only if the team can trace it to the correct policy and tell which cases still need a decision.

Zapier Agents Pricing Before and After Migration

Zapier lists standalone Agents Free at 400 activities per month and Agents Pro at 1,500 activities for $400 per year, shown as $33.33 per month with annual billing. Enterprise has a custom allowance. An activity can be a trigger, a knowledge lookup, or an app action, so one run can use several. Zapier documents a 10-activity per-run limit on Free and 40 on Pro. Tests count against the Free allowance, while paid Agents tests are excluded.

The migrated AI step draws from your Zap task allowance. Zapier’s model-tier guide counts one base charge plus one charge per successful app-action or knowledge-source tool call, each at the selected model rate. An Advanced step with two successful tool calls uses nine tasks. The same example at Premium uses 15. At 300 tickets, that is 2,700 or 4,500 tasks for the AI step alone, before other billable Zap actions.

The guide lists Standard at 1×, Advanced at 3×, and Premium at 5×. Tools and knowledge require a supported Advanced or Premium model. Bringing your own key uses Zapier’s 1× task rate, with model-provider charges billed separately. On the free Zapier plan, the new AI step can be previewed at Standard, while publishing a Zap with the step requires a paid plan. Check the actual task count on representative runs before budgeting around a model tier. Include cases that require several policy lookups, since a ticket that looks like one unit of work in your queue can generate very different usage. Zapier also says a full test of a migrated Zap currently uses tasks like a production run, so leave room for acceptance testing in the migration budget.

Permissions and Approvals Inside the Zap

AI by Zapier lets you require approval per tool. Zapier documents that the setting is off by default. Turn it on where a tool can send a message or change a record, and review the values the agent proposes to submit. A tool approval controls that action at runtime. Enterprise publishing controls govern who can put a workflow into use. SD’s AI agent security guide covers why tool access and review gates matter when an agent processes untrusted content.

Give the workflow a connection with access only to the records it needs. Treat an incoming email, website, or retrieved document as data. A message asking the agent to ignore company policy should leave its permissions and approval rules intact.

Check what happens after a rejected approval, a missing record, and a retry. Where the workflow writes to another app, use the original ticket ID to check for an existing entry before creating one. That gives an operator a way to rerun part of the workflow without opening a second ticket or sending a second reply.

What to Check When Migrating an Existing Agent

Zapier’s migration creates a Zap with a native trigger and one AI by Zapier step containing the original instructions and tools. The individual tools stay inside that step. If an action needs its own validation, route, or audit trail, you can move it into a dedicated Zap step after conversion.

Compare both versions on the same sample records. Check whether the new step can retrieve old documents as well as recent ones. Zapier’s migration guide lists some legacy knowledge sources, including HubSpot and Zendesk, as unsupported in the new step at the time of review. An app action and a searchable knowledge source are separate capabilities. Check the one your agent uses. The guide also says on-demand chat agents lack a direct equivalent trigger in the Zap editor and describes a webhook as one workaround.

Before switching live traffic, test the converted Zap with sample data and confirm downstream fields and tool permissions. Plan when to disable the original agent. Zapier warns that both versions can process the same trigger while both are active. Record the cutover time and check recent records for duplicates, especially where either workflow sends messages or creates tickets.

Who Should Use AI by Zapier?

AI by Zapier fits teams that already use Zaps and need model judgment at a defined point in an existing workflow. A routine that extracts fixed fields and updates a known record is easier to audit with regular steps. An AI step earns its place when staff spend meaningful time interpreting messy inputs or choosing which reference material applies.

Teams that want more control over hosting and the automation runtime can compare the approach in SD’s n8n review. Whichever platform you consider, use the same sample inputs and check accepted outputs, manual review time, and the cost of reprocessing failures. The useful comparison is the cost of a correctly handled case.

Conclusion

The takeaway from this Zapier Agents review is to build new agent workflows in AI by Zapier when the trigger, tools, and review steps already fit your Zaps. Existing Agents are worth migrating after you confirm that knowledge retrieval, approvals, and task usage work for your busiest cases. Start with one workflow and use its real results to decide how much work to give the agent.