Coming from Zapier Agents?
This page compares the documented Zapier Agents activity model. Zapier now says Agents is moving to AI by Zapier; check its migration notice before choosing a plan. Whenever writes fixed procedures from plain-language rules. Code decides the known steps; optional model steps remain available, can vary and have runtime AI costs.
What you get on Whenever
Run 1,000 behaves like run 1
The same plain-language brief becomes TypeScript you review before it runs. The published code decides the steps, so a run does not ask a model what to do next. This applies to fixed non-AI logic with the same inputs and dependencies; optional AI and changing external state can change outcomes. You can review the summary and results without reading code.
No model deciding each step
AI writes the workflow once, with free building subject to weekly AI usage limits. A published run does not need a model to choose its steps, so the only model call left is one you deliberately put there.
Review each change before relying on it
Review the changed rules, sample outcomes and run history before relying on an update; AI by Zapier also lets admins require approval before AI steps publish. The code is available for you or a developer, but reading it is not required.
A concrete decision
Illustrative scenarios, not customer results or recorded runs.
Greg: no “done” without the invoice receipt
Illustrative decision, not a recorded customer run: Greg wants each paid order checked against Stripe, then an invoice created, its ID saved, and only then a confirmation sent. He specifies that an invoice failure must stop the confirmation, and an already-handled order must not be charged or invoiced again. He checks the summary, sample records, provider receipts and run history—not source code. Test a missing payment, a failed invoice and a duplicate event before publishing. Whenever fits when those prescribed steps matter more than an agent deciding them again; a fixed workflow elsewhere can fit too. No screenshot here claims these actions were actually delivered.
Jaunius: make the edge cases explicit
Illustrative decision, not a benchmark: Jaunius needs all pages of 5,000 contacts, a saved last-seen ID, updates only for changed records and a clear record of which writes failed. He brings the old workflow and sample inputs, reconnects accounts, and tests pagination, an API failure and a repeated event before switching triggers. Choose a single procedure if it makes this logic easier to maintain; the other tools can also support complex logic. An optional model that classifies a note can vary and adds runtime AI usage. Compare the monthly subscription and actual usage, not the number of steps alone.
Agents activities or AI by Zapier tasks?
Zapier is moving Agents to AI by Zapier, where native deterministic steps and model-directed work can coexist. Whenever also supports explicit model steps: fixed logic does not need inference, but optional runtime AI still costs money.
Zapier Agents
Activity billing; migration announced
Triggers, actions, searches, browsing and knowledge answers
Agents publishes activity plans; migrated AI by Zapier work uses Zap task billing. Zapier says it has not set a date to turn off Agents.
Agents counts activities including triggers, actions, searches and knowledge answers. If a run reaches its per-run activity limit with actions remaining, it requests input before continuing; exhausting the monthly allowance instead stops agents from running. Zapier has announced migration to AI by Zapier and says it has not set a date to turn off Agents. This page does not quote Agents list prices or convert activities into runs.
A run is one execution of one workflow, start to finish. AI building and editing are free within weekly usage limits; runs are metered separately.
A closer look
What changes when you describe the job in plain language and code runs it: to build, to run, and to keep.
Repeatable execution logic
Zapier Agents
Zapier’s Agents FAQ describes nondeterministic AI outcomes. AI by Zapier also supports mixing agentic and deterministic native steps, so compare the execution mode, not a blanket claim about every Zapier run.
Whenever
AI writes it, then published code runs the same execution logic. With identical inputs and dependencies, non-AI logic is deterministic. Changing API data, time, randomness or an optional model step can change results; this is not an exactly-once delivery guarantee.
Runs per dollar
Zapier Agents
The published Agents plans count activities. Zapier has announced migration to AI by Zapier, but says it has not set a date to turn off Agents and will email affected users first; the pricing page still offers activity plans. Check which product and meter your account uses.
Whenever
Pro includes 10,000 runs per month; see the pricing page for current US dollar prices. AI building and editing are free within weekly usage limits. Published runs are metered separately: the meter counts the whole run, not individual steps. External services and optional model calls can have their own charges.
What you can express
Zapier Agents
Instructions in prose, plus the tools and data sources you connect. What the agent will actually do with them is decided at run time, by the model.
Whenever
Describe the rules you already know; AI writes the procedure. Review the summary and sample results, not necessarily the code. Keep an explicit model step for judgment. Agent platforms may also offer fixed workflows or scripts; compare the mode you actually use.
Export and portability
Zapier Agents
Zapier documents an account-data export whose Agents files include basic agent information with instructions, plus chats and messages. Exported account data is not a promise of execution on another platform.
Whenever
The TypeScript is yours: read it, edit it, export it, take it with you. Moving it to another runtime can require adapting managed connections, triggers and runtime-specific integrations.
Debugging a failed run
Zapier Agents
Zapier Agents shows how your agents use activities in its Activities view. Revise the instructions or connected tools and test the behavior again.
Whenever
Check the summary, execution log and sample outcomes, then ask for a correction in chat. Inspect TypeScript if useful; use safe inputs when testing external writes again.
Where the AI cost lands
Zapier Agents
Agents uses activity accounting. Migrated AI by Zapier steps use Zap task billing and can be mixed with deterministic native steps. Model-directed work and configured native steps have different usage profiles.
Whenever
AI writes the workflow once, with free building subject to weekly AI usage limits. A published procedure runs its steps without a model choosing them. Any model call you explicitly include still runs and has its own cost.
When the model is the right answer
Zapier Agents
Consider model-directed work for unstructured input and decisions that cannot be fully specified as fixed rules. This is a fit recommendation, not a claim that only a model can process an invoice or CV.
Whenever
Call a model from a workflow when a step genuinely needs judgement. The difference is that it is one step you chose, not the thing running your automation.
Sources and scope
Official sources captured on the date shown below. These describe published features and billing rules, not independent performance tests. Check the linked documentation for current modes, plans and limits. No competitor dollar price is asserted.
Checked
Start with a workflow brief
Pick a starting point, connect your accounts and review what Whenever builds before publishing.
Build this workflowAnnounce closed-won deals in SlackCelebrates each won deal in Slack with the numbers and a short recap of how it was won.
Build this workflowPublish Instagram posts from a calendarPublishes due rows from your Instagram calendar on the cadence you choose and writes each post link back.
Build this workflowLog survey responses and alert on low scoresAdds each survey response to Google Sheets within minutes and emails the owner when the score falls below your threshold.
Build this workflowTrack keyword rankings in a sheet each weekChecks the search position of every keyword on your list, appends the history to a sheet and reports the moves.
Build this workflowRoute large expense claims to a managerLogs every expense submission in a sheet and alerts a manager in Slack when the amount is over the approval limit.Build this workflowStart with one job your agent does.
Describe what the agent is supposed to do. Whenever turns it into tested, running code.
Describe the job
Tell Whenever what the agent should do, and what a good result looks like.
Check the logic
Review the summary and representative outputs. Include a model step where judgment is needed, and account for its variable results and runtime cost.
Test before it runs
Use real inputs to check the result, then publish the workflow on your trigger.