Article

Fix Code by Zapier steps in imported Next Gen Zaps

Automation owners can rebuild imported Code steps, check downstream data or restore a Classic Zap, with live-test and webhook cautions.

Editorial illustration for Fix Code by Zapier steps in imported Next Gen Zaps: a controlled task flows from input to output. Not documentary evidence.

Since November 17, 2026, imported Next Gen Zaps that retain a Code by Zapier step stop with an error at that step, according to Zapier. Rebuild the logic and verify its outputs, or turn off the Next Gen copy and reactivate the original Classic Zap.

The import leaves a nonworking placeholder for each Code step; its position in the diagram does not preserve its behavior. This is not a retirement of Classic Code steps. Moving to Next Gen is optional, and importing leaves the original Zap unchanged.

Confirm which step is failing

Open Automations, select Next Gen Zaps, then open the affected workflow. Under Runs, select a failed run and click Explain error in the Error section. Confirm the retained Code step is the cause before changing anything.

Shared access is viewer-only. If you cannot edit, involve the workflow owner, a super admin or the account owner: these are the roles with editing rights.

Choose a rebuild or return to Classic

A practical decision rule: rebuild when you understand the original logic and can check the values later steps need. If you cannot yet verify that behavior, returning to the existing Classic workflow gives you a way to keep using the original logic while investigating.

Zapier’s documented fallback is to turn off the Next Gen copy and turn the original Classic Zap back on. Confirm the original still exists and its app connections and trigger work before relying on that fallback.

For a webhook trigger—an address another app sends events to—the import receives a different URL. The trigger panel shows both addresses. If you redirected the sending app during migration, point it back to the Classic address when returning to Classic. The practical implication is that switching the Zap toggles alone is not enough: the sender must reach the version you intend to run.

Rebuild the logic and check downstream fields

In Workflow Manager, open the intended repair draft, or use Edit Zap to create one from the published version. Ask Zapier to recreate the unsupported step’s behavior in code. The editor supports plain-language changes and TypeScript editing; this is a rebuild, not a promise that existing code can be pasted unchanged.

Before asking for that change, make a short acceptance checklist from the original Classic step:

  • Which input fields does it read, and what should happen when a value is missing?

  • Which output fields and data types must it produce?

  • Which later actions use those outputs, and what exact values should those actions receive?

Check every dependent field, not just the replacement step. Values controlled by code cannot be edited directly on the visual canvas; change them through the Chat panel or code editor. Also select an accessible app connection if the import removed one you could not access, and check for unsupported Global, Zap or System variables.

Before testing, open Settings and inspect Agentic management. Newly imported workflows start with Healing set to Automatically fix. If you want to review agent-led repairs, choose Suggest fixes: it requires approval before the agent tests and publishes a fix. Automatic fixes can perform live tests without asking, except when Zapier flags the change as higher risk or uncertain.

If you use MCP to connect an AI assistant to Zapier, you can repair in the website editor without changing that connection’s restrictions. RohitAI’s guide to limiting Claude to chosen Zapier actions covers managed mode. That mode does not include Next Gen workflow-building tools; it is not an alternative repair interface.

Test the resulting data before publishing

Open the repair draft, click Test and use Edit Inputs to choose the starting data. Tests execute against live app connections, so a workflow that sends messages or creates records can do so during the test. Use deliberately chosen, low-impact records and recipients.

Treat a successful run as only the first check. Inspect the destination app and compare its actual field values with your acceptance checklist. A populated field containing the wrong value should fail that check. Where empty inputs are possible, check the intended empty-input behavior too.

A draft test runs the draft rather than the published version. Confirm which draft is open before testing, especially if more than one repair is in progress.

Activate only the intended workflow

Before publishing the repaired Next Gen draft, ensure the original Classic Zap is off. Zapier identifies leaving both copies active as a cause of duplicate processing.

In Workflow Manager, open the intended draft, click Publish, review the version name and confirm. Publishing there turns the workflow on even if it was off. Check the active version and trigger state afterward. Publishing through MCP differs: it preserves the existing on/off state unless activation is requested separately.

If you are switching a webhook sender to Next Gen, use the new address shown in the trigger panel. Check the next intended event and its destination record. Publishing a repair does not replace the version used by an already-running workflow.

Reconcile failed runs and missed events separately

Before retrying an old event, inspect any records or messages already produced. Re-run uses the failed run’s inputs with the currently live version; it is not a promise to resume only at the failed step. Release is different: it continues a held run from its holding point. Only the workflow owner or a super admin can use these recovery controls.

Events arriving while a Next Gen Zap is off are not automatically recovered when it is turned back on. Reconcile the interruption against the source app’s records. Whether missing events can be replayed depends on that app; do not assume reactivation clears a backlog.

A search for an existing record can help identify prior work, but a search-before-write check does not guarantee a single execution: overlapping runs may both search before either writes.

Keep December billing separate from the repair

As of November 20, Next Gen remains limited early access by request on Professional, Team and Enterprise, not Free. The November 17 Code-step rule is already in effect; it is separate from task billing scheduled for December 1, 2026, at 17:00 UTC.

Zapier’s current billing instructions say that Next Gen code which only transforms data uses no tasks. Successful ordinary app actions and external calls through Zapier generally use one each, with free-app and specialized-action exceptions. Rebuilding a Code step is therefore not, by itself, a one-task charge. The dedicated billing page sets out these distinctions.

Next Gen usage is free until the cutover, but the paid subscription remains necessary. Build-time app calls through MCP can already consume tasks, and an external AI client may charge for tokens. For account-level planning, see budgeting for Zapier Next Gen and SDK billing. That earlier guide notes a logic-only pricing disagreement; the dedicated billing page updated November 18 explicitly allows zero-task logic-only runs.