Hi,
I want to add an AI step to a Zap I am building, that will format the text from the body of an email, input from a previous step.
My question is if this AI step costs an agent run, or not (and just a zap run).
Best answer
AI step cost in Zapier: Does it count as an agent run?
- Beginner
Best answer by SamB
Hey there
Are you referring to an AI by Zapier action or an Agents action?
AI by Zapier action steps in a Zap would count towards your task usage. So each time your Zap runs and hits that AI step, it would use one task from your plan's task allowance, just like any other successfully run action step (apart form the ones listed here).
This is different to Agents action steps in a Zap, which would not count towards your task usage when the Zap runs (see our Start an agent from a Zap guide for details). However, any actions the agent takes when the Zap runs would count as an activity. You can learn more about agent activities here: How is Zapier Agents usage measured?
Hope that helps to clarify. Let us know if you have any more questions!
+11- Community Manager
- Answer
Hey there
Are you referring to an AI by Zapier action or an Agents action?
AI by Zapier action steps in a Zap would count towards your task usage. So each time your Zap runs and hits that AI step, it would use one task from your plan's task allowance, just like any other successfully run action step (apart form the ones listed here).
This is different to Agents action steps in a Zap, which would not count towards your task usage when the Zap runs (see our Start an agent from a Zap guide for details). However, any actions the agent takes when the Zap runs would count as an activity. You can learn more about agent activities here: How is Zapier Agents usage measured?
Hope that helps to clarify. Let us know if you have any more questions!
One extra thing I would separate here is Zapier-side usage vs provider-side token cost.
The AI step / Agent activity rules tell you what Zapier counts. If the workflow also calls an external OpenAI-compatible API, I would still track that separately with:
- one scoped API key per Zap or environment- a hard quota on that key- request logs with model, input/output/cache tokens, and request id
That gives you a ceiling before a Zap loop or an agent retry pattern turns into a surprise upstream bill.
I wrote the setup pattern here: https://github.com/XXXXXXXXXXXX/XXXXXX-XXXXX-XXXXX-XXXXXX
Disclosure: I maintain Wappkit, so I am biased toward quota-capped keys and request logs. The same pattern works with an internal gateway too.
This post has been edited by a moderator to remove a link to self-promotional content as per our Community Code of Conduct.
Enter your E-mail address. We'll send you an e-mail with instructions to reset your password.