When your automation vendor shuts down, what actually transfers?
One of the few automation products built entirely around human approval is winding down, with no acquirer and no recommended successor. Its customers get a JSON export and between thirty and sixty days, depending on whether they paid. Here's the uncomfortable question that raises for everyone else — including us.
When an automation vendor shuts down, what transfers is whatever you authored in a portable form. A visual workflow exported as JSON describes that vendor's node library, not your business, so it cannot be run anywhere else and has to be rebuilt by hand. Plain-language instructions, your own model and provider keys, and your own data exports do transfer; OAuth connections never do, and you re-authorize each integration in the next tool. Before adopting any automation platform, ask what leaves with you, in what format, and whether it is usable without that vendor — the answer is a structural property of what the tool asks you to author, not a promise in a contract.
What happened, and why it's the interesting case
On 16 July 2026, Relay.app announced it was winding down. New signups stopped that day. Free accounts have until 15 August to get their data out; paid accounts until 14 September. Subscriptions were cancelled on announcement and annual plans refunded pro rata. Customers can export workflows, sequences, and MCP server configs as JSON, and prompts, run history, and tables as CSV. Anything not exported will be, in the vendor's own words, permanently deleted. No acquirer was named. No successor product was named. No recommended alternative was named.
Vendors shut down constantly and most of them don't merit an article. This one does, for a reason that has nothing to do with schadenfreude: Relay was one of the few automation products whose entire thesis was human-in-the-loop. Its earlier launch materials framed it as AI drafting well while a human touch is still needed. That is very close to our thesis, and its customers had already decided supervision was worth paying for.
So the failure wasn't a failure of the idea. Which makes it the cleanest available case study in a question every buyer should be asking and almost nobody asks until it's forced: when this tool goes away, what do I still have?
The dates and the export scope above are from Relay.app's own wind-down notice, checked 25 July 2026; the human-in-the-loop positioning is from its earlier launch materials rather than that notice. We have no inside knowledge of why they wound down and we're not going to speculate — the dates and the export scope are the parts that matter for this argument, and those are on the record.
What transfers, and what only looks like it does
"You can export your workflows" sounds like portability. It usually isn't, and the gap between the two is where the pain lands.
| What you get back | Actually portable? | Why |
|---|---|---|
| Visual workflow as JSON | No | It encodes one vendor's node types, field mappings, and connector IDs. Nothing else can execute it, and no other vendor will import it. It is documentation of a thing that no longer runs. |
| Run history as CSV | Partly | Useful as a record, useless as a rebuild. You can see what happened; you cannot replay it. |
| Your business data (tables, records) | Yes | Rows are rows. This is the part vendors are usually good at, and it's the least of what you lose. |
| Credentials and OAuth connections | No, and that's correct | They were never yours to export. You re-authorize every integration by hand in the next tool — which for a fifteen-app workflow is most of a week. |
| The prompts you wrote | Yes | Text is text. Notice which side of this table the plain-language parts fall on. |
| Plain-language instructions describing the outcome | Yes | A paragraph of English is readable by a person, by another platform, and by a model. It describes your business rather than a node library. |
| The knowledge of why it worked that way | Only if written down | The single most expensive loss, and no export format contains it. If the rules lived in the diagram, they leave with the diagram. |
Read the two "yes" rows again. The things that survive a vendor are the things you wrote in language. The things that die are the things you drew in their canvas.
That isn't a coincidence and it isn't a product decision anyone made. It's structural: a diagram is only meaningful inside the system whose vocabulary it uses, and a sentence is meaningful anywhere.
Why we think this is the whole argument
We should be upfront that this is the argument our product is built on, so you can discount it accordingly.
Our team has built automation platforms before — twice, across two generations of this category. First visual playbooks, then a no-code drag-and-drop builder that was acquired. Both models hit the same wall, and it wasn't a wall of features. It was maintenance. The diagram is the asset, the diagram is coupled to one vendor's node library, and so the diagram has to be rebuilt every time the vendor, the API, or the form changes. Eventually a company is paying people to maintain the automation that was supposed to save them people.
The conclusion we drew is that the AI-native version of this category isn't drawn. It's described. Not because describing is more pleasant — though it is — but because a description is the only artifact in this category that outlives the tool it was written in.
What survives a vendor is the sentence. "Every Monday, chase invoices more than thirty days overdue, draft the email, and let me approve anything over $10,000." That is portable — to another platform, to a new hire, to a model that doesn't exist yet. A node graph is not.
Four questions to ask any automation vendor
Ask these before you build, not during a wind-down. Ask them of us too — we've answered them below, in the same order.
- 01
What exactly leaves with me, in what format, and can it run anywhere else?
Push past "you can export everything." Ask for a sample export file and read it. If it's a graph of the vendor's node types, you're being offered a souvenir, not portability. The follow-up that settles it: name one other system that can execute this file.
- 02
Whose accounts hold the credentials and the model keys?
If the vendor holds every OAuth connection and resells you AI usage as bundled credits, then a shutdown costs you a re-authorization project plus a billing relationship. If you can bring your own model key, your provider account and your bill are already yours and the migration is smaller by exactly that much.
- 03
Where do the rules live — in the diagram, or in writing?
This one is about you, not the vendor, and it's the answer that most changes your exposure. If the reason a branch exists is undocumented, that reasoning is destroyed the moment the canvas is. Teams who write briefs get documentation as a side effect; teams who draw graphs get a dependency.
- 04
What happens to a run that's mid-flight when something breaks?
Not a shutdown question, an every-Tuesday question, and the answer tells you how seriously the platform takes operations. Ask whether a failed run resumes from the step that failed or starts over. Starting over is actively dangerous when step three already sent an email.
Our answers, including the uncomfortable one
If you're reading this because a vendor just failed you, your first thought about us is reasonable: you'll die too, and you're smaller. We'd rather answer that plainly than perform confidence about it.
- A Boring agent is a paragraph of English, not a proprietary graph. That's the answer to question one. Export it and it's readable — and reusable on another agent platform, including one that isn't ours. We can't make that untrue by shutting down, because it isn't a promise about our conduct; it's a property of what we asked you to write.
- Data export and account deletion work today. Workspace and account deletion and summary CSV export are shipped, not roadmap. Being precise about the limits, because this is the bullet where precision counts: the CSV covers run and workspace summaries rather than the full step-by-step payload record, and it currently reaches the most recent 100 runs. If your history matters more than that, ask us before you commit rather than after.
- Bring your own model key. Your provider account, your bill, no markup from us on your AI usage. We don't want to own your token bill — and a vendor that bundles AI credits into a flat price has an incentive problem you inherit.
- We're early, pre-GA, and you should ask us what's real. There's no version of this paragraph that makes that reassuring, so we won't try. What we'll offer instead is that every capability claim on this site is checked against an internal audit of what actually ships, our comparison pages say plainly when a competitor is the better choice, and the readiness check on this site will tell you to use a cheaper tool if your workflow is fixed plumbing. If we ever get ahead of the product, tell us and we'll fix the page.
- And the fourth question, on us: when a run fails you can retry it from the step that failed rather than starting over, so completed work isn't repeated — with the honest caveat that a run which died before finishing its first step has nothing to resume from and re-runs in full, which the console tells you.
If you're holding a folder of workflow JSON from a dead vendor, we don't have an importer that turns it into a working agent, and we're not going to imply otherwise. What that folder is genuinely good for is reading: it's a record of every branch and edge case someone once thought was necessary, which makes it excellent raw material for writing a brief. That rewrite is a real piece of work — it's also the point at which most teams discover half those branches existed to work around the tool.
If you're mid-migration right now
Regardless of where you land, this order wastes the least time:
- Export everything today, before you've decided anything. Deadlines in a wind-down are real and unexported data is genuinely deleted. Do this even if you're keeping the exports as reference only.
- Write down what each workflow was for, in prose, while you still have the diagram to read. One paragraph per workflow: the outcome, the rules, the exceptions, who approved what. This is the artifact you were always missing, and the canvas is about to stop being available to reconstruct it from.
- Rank by consequence, not by complexity. Rebuild the workflows that touch money, access, or customers first — those are the ones where an unsupervised gap costs the most while you're between tools.
- Take the opportunity to delete some. A wind-down is a free audit. Some of those workflows haven't earned their maintenance in a year, and rebuilding them is a decision you get to make again.
- Then pick the tool by shape, not by feature count. If the work is fixed plumbing, a trigger-action builder will be cheaper and more predictable than any agent — our comparison pages say which one, and our readiness check will tell you if that's your case in about two minutes.
Questions about this
What happens to my workflows if an automation vendor shuts down?
You typically get an export of the workflow definitions as JSON, your business data as CSV, and a deadline. The data is genuinely portable; the workflow definitions usually aren't, because they encode one vendor's node types and connector IDs rather than a description of your process. In practice that means the workflows get rebuilt by hand in the next tool, and the integrations get re-authorized one at a time.
Is workflow JSON portable between automation platforms?
Effectively no. There is no shared standard for workflow definitions, and no major platform imports a competitor's export. The file documents what you built, which is useful when rewriting, but it cannot be executed anywhere except the tool that produced it. Treat an export as a reference document, not as a migration path.
How do I avoid automation vendor lock-in?
Author in a form that survives the tool. Written instructions describing the outcome, kept somewhere you control, transfer to any platform and to any new hire. Hold your own credentials and model keys where the platform allows it. And write down why each rule exists — that reasoning is the most expensive thing to lose and the one no export format includes.
Should I trust a small or early-stage automation vendor?
Trust the structure more than the company. Ask what you'd be holding if they disappeared and whether it would be usable elsewhere; if the honest answer is "a folder nothing can read," the vendor's size is the least of the risk. A larger vendor reduces the odds of a shutdown but not the cost of one — and large vendors sunset products too.
Can Boring AI import my Relay.app or n8n workflows?
Not automatically, and we won't pretend otherwise. There's no one-click importer for either, and rebuilding is a real piece of work — our white-glove team does it with customers rather than a wizard doing it in a minute. The exported JSON is genuinely useful as source material for writing the plain-language brief, because it records every edge case someone once accounted for.
What does Boring AI do if it shuts down?
The honest answer is structural rather than contractual: what you'd hold is the plain-language brief for each agent, which stays readable and reusable on another platform, plus your data via the export that works today and your own model keys if you brought them. We're early and pre-GA, and you should weigh that. We'd rather tell you that on this page than have you discover it later.
Stop reading. Describe the workflow.
Request access and describe it in a sentence — or ask about white glove and our team will build and run it with you.