Zapier Next Gen Zaps review: tell your AI the job and it keeps running. Decide what it may fix before December 1
Describe a job to Claude or ChatGPT and Zapier now builds a workflow that keeps running after you close the chat. It's free in early access until December 1. What matters is what happens the first time one of those runs breaks.

Zapier's Next Gen Zaps, in early access and free until December 1, 2026, let an AI assistant build a workflow that then runs on Zapier by itself, with run history and a built-in agent that can diagnose and fix failures. It's worth trying on one low-risk job. Keep your Classic Zaps running, keep fixes behind your approval, and leave financial and legal workflows out until the audit and approval features are confirmed.
What Zapier launched
Most automation breaks in a boring way. A field gets renamed, a login expires, and a Zap that has run quietly for a year starts doing nothing, or doing the wrong thing, until somebody notices.
Zapier's new Next Gen Zaps are built around that moment. According to Zapier's help pages, updated November 13, you describe a job to an AI tool you already use, such as Claude, ChatGPT, Gemini or Cursor, and the tool builds the workflow for you. It picks the app connections, wires the steps and deploys the result to Zapier. You can also do the same from the AI assistant inside Zapier's Workflow Manager.
The detail that matters is where it runs. A Next Gen Zap runs on Zapier, not inside the chat. It starts from a trigger or a schedule with no chat window open, you can also run it on demand, it keeps a full history of its runs, and it follows your Zapier account's permissions. Zapier's examples include looping through a whole list with no 500-item cap, catching an error and carrying on with the rest, and pausing until a person replies.
Next Gen Zaps run alongside your Classic Zaps, which keep running unchanged. You decide whether and when to import one. Zapier points out that importing makes a copy, the original keeps running until you turn it off, and leaving both on can cause duplicate runs.
It's in early access by request, with limited availability, and Zapier describes it as in active development.
Who it's for
Think of an owner or an office manager who already lives in an AI chat and has a list of small jobs that never got automated. A weekly digest of unpaid invoices. A tag that goes on every lead who mentions a deadline. A reminder that waits three days and stops if the customer replies. Zapier's own examples sit in that territory: HubSpot lists, QuickBooks invoices, Slack digests, trial-user emails that stop when someone upgrades.
It's less of a fit if your automation is already a tangle of Classic Zaps that nobody fully understands. An AI that can build a workflow quickly can also build a fourteenth workflow quickly, and nothing in the product makes you less likely to lose track of them. And if the job touches money or contracts, read the security section further down before you try it.
Decision guide
Classic Zap or Next Gen Zap?
| Consideration | Classic Zap | Next Gen Zap (early access) |
|---|---|---|
| How it's built | Step by step in the Zap editor | Described to an AI tool, which wires and deploys it |
| Where it runs | On Zapier | On Zapier, not inside the chat, so no window needs to stay open |
| When it fails | You open the Zap and find the broken step | An agent can diagnose it and prepare a fix, if you allow that |
| Cost | Tasks from your plan, as today | Free in early access until December 1, 2026; tasks after that |
| Maturity | Keeps running unchanged | Early access, by request, still in active development |
A workflow your AI built is still your workflow. Decide who gets the 3 a.m. failure before it happens.
The run that fails at 3 a.m.
Here's the way we'd review it. Forget the building part for a minute. Anyone can be impressed by a workflow appearing in two minutes. Look at what happens on a Tuesday night when the connection to your accounting tool expires.
Zapier's answer is a feature called Agentic Management, and its help page says it's on by default. If a run fails because of a renamed field or an expired connection, an agent can diagnose the problem and prepare a fix. After successful runs it can also suggest improvements, such as turning an AI step that keeps making the same decision into a fixed step that runs faster and more predictably. When it applies a fix, it publishes the change as a new version of the workflow, and you can see exactly what changed in the version history. The help page also says you control what the agent can change without your approval.
That last sentence is the whole review. A fix that restores a broken connection is a very different thing from a fix that changes who gets an email, how a total is calculated or what a reminder says. Both would count as "a fix" to an agent. They don't count the same to the customer who receives the message. A workflow your AI built is still your workflow. Decide who gets the 3 a.m. failure before it happens.
So before the first Next Gen Zap goes live, write three lines. What may the agent change without asking? Our suggestion is nothing, to start. Who gets told when it proposes a change? A named person, not a shared inbox. And how do you roll back? The version history is there, but someone has to know to look at it.
If you've read our note on moving Zapier Agents into Zaps without sending everything twice, the same rule applies. Anything that can run twice will run twice at the worst time.
Where it falls short
It's early access. Zapier says availability is limited and the feature is in active development. Don't make it the only copy of anything you can't redo by hand.
Enterprise controls aren't all there yet. For Enterprise accounts, Zapier's page lists features that aren't supported or aren't confirmed to work the same way in Next Gen Zaps: audit log, log streams, per-asset change history, action restrictions and publish approval requests among them. The page says plainly not to run your most sensitive workflows through Next Gen Zaps during early access. If you're a business that needs an audit trail for a regulator or an insurer, that settles it for now.
The security warning is real. Zapier's own guidance says that an agent with access to private data, exposed to untrusted instructions and allowed to send data elsewhere could take actions you didn't intend, including sharing sensitive information. Its advice is to start with low-risk workflows and connections, avoid accounts holding legal, financial or private company files, and use a separate account or test workspace if you want a stricter boundary. A workflow that reads incoming customer email and has permission to send messages is exactly that combination.
The meter starts December 1. Zapier says Next Gen Zaps use tasks from December 1, 2026, and are free until then during early access on Professional, Team and Enterprise plans. If your AI tool runs app actions through Zapier MCP while it builds, each successful tool call uses tasks at the Zapier MCP rate, and your AI tool may charge for its own tokens too. Zapier's billing page explains what counts as a task, so read it before you run anything big, then run one job on a small list and read the count.
You still own the thinking. The AI picks the connections and wires the steps. It doesn't know that your Tuesday invoice run exists because a customer once got billed twice.
How the work moves
Try one before December 1
- Pick a boring job
Something that pings a person if it goes wrong: a weekly digest, a lead tag, a reminder.
- Use a throwaway connection
A test workspace or low-risk accounts, as Zapier itself suggests for early access.
- Set the fix rule first
Decide what the agent may change without asking. Start with nothing.
- Name the person who gets the failure
One human, one inbox, written down.
- Compare with the Classic version
Run both on test data, then turn one off. Never leave both on.
How it compares
The obvious alternative is the Classic Zap you already have. It stays put, behaves the same and is the right place to keep anything that matters until Next Gen has a track record. If a Classic Zap works, leave it alone. Importing one just to try the new thing is how you end up sending every email twice.
The other alternative is a person who builds and maintains the workflow, either you or a developer. That's slower to start, and it costs more up front. In exchange you get someone who knows why each step exists. For anything that touches payments, contracts or customer records, we'd still lean that way, with the AI-built version as a prototype you compare against.
For a simple, low-stakes job, an AI-built Next Gen Zap on a test workspace is a fair way to find out whether the idea works. The price of finding out is zero until December 1.
Plain verdict
Try it, on one boring job, before December 1. Use low-risk connections. Keep your Classic Zaps running and never leave a copy and its original both switched on. Set the agent's permission to change things to "ask me" and name the person who answers. Leave financial and legal workflows alone until Zapier confirms the audit and approval features you'd need.
If the automations in your business have grown past what one person can keep in their head, with some in Zapier, some in scripts and some in a spreadsheet nobody owns, that's the part we do. Our business automation work starts by mapping what runs, who owns it and what happens when it breaks. Talk to us.
Related work: Spin:Market
We build alongside the people doing the work.
Bring us the problem you’re thinking about.