AI Automation Tools Keep Shutting Down. How to Depend on One Anyway
On July 16, 2026, Relay.app told its users it was shutting down: free accounts deleted August 15, paying customers September 14. No reason was given. It’s a well-liked product — 4.9/5 across 71 G2 reviews, praised for being the automation tool non-technical people could actually learn — backed by $8.1M from Andreessen Horowitz, Khosla Ventures, BoxGroup, Neo, and Entropia Capital, founded in 2021 by Jacob Bank, who previously ran product for Gmail and Google Calendar at Google. It is not an obvious candidate for a wind-down, which is rather the point.
It’s also the fourth or fifth tool in this neighborhood to end inside eighteen months. We build in this category, so the honest disclosure goes at the top rather than buried: everything below applies to us too. Carly is a venture-scale bet in a hard market, and no post we write makes us immune to the thing we’re describing. What follows is not a pitch. It’s the analysis we’d want if we were the ones being migrated away from.
Because the useful observation isn’t “startups fail.” It’s that these endings were structurally different from each other, and the differences — not the sadness — are what determined whether users lost their data.
Five endings that are not the same ending
The acquihire-and-kill: Clockwise
Clockwise raised roughly $76 million and served about 40,000 organizations. On March 19, 2026 it announced the team was joining Salesforce to work on Agentforce. On March 27 — eight days later — the service stopped. Salesforce wanted the engineers, not the calendar app. Accounts stopped working, user data was deleted, and nothing transferred to Salesforce. Prepaid customers got automatic prorated refunds. We wrote the migration guide in Clockwise shut down in March 2026.
This is the worst shape for a user, and it’s worth understanding why: in an acquihire, the acquirer is buying people. The product is a liability on day one — it has support costs, security surface, and no strategic value. Nothing in the deal structure rewards anyone for keeping it alive or moving your data somewhere. The notice period is short because the team’s start date, not your migration, sets the calendar.
The acquihire-and-kill, again: Nerve
Nerve sold an “AI chief of staff” that connected internal tools to agents through Slack and Teams. On February 17, 2026, co-founder Tanooj Luthra posted that he and his small team were joining OpenAI and the product would sunset in about a month. Terms weren’t disclosed — including whether it was a formal acquisition at all, which is increasingly how these are structured.
(Confusingly, there is an unrelated, still-operating product at getnerve.ai that has put a disambiguation notice on its own homepage saying it is not the Nerve that OpenAI took. Names in this category are not unique identifiers; Relay.app shares its name with a London logistics company and a Toronto fintech, and the funding numbers get merged by aggregators constantly.)
The asset license and team hire: Mindy
Mindy was an email-first AI agent. On June 16, 2026, WorkWhile announced it was licensing Mindy’s assets and that the team would integrate into its Product and Engineering divisions. Mindy’s own note thanks the community and describes the team joining WorkWhile.
Here’s the finding, and it’s a finding by absence: neither announcement says anything about what happens to user data, exports, or refunds. When you read a wind-down notice, the things it doesn’t mention are load-bearing. An asset license plus a team hire is an acquihire wearing a slightly different legal costume, and the tell is the same one — nobody has been made responsible for your account.
The strategic pivot: Notion Mail
Notion is shutting the Notion Mail inbox down on September 22, 2026, announced June 25 — about three months of notice. This one is different in every dimension that matters. A reason was given, and it’s a real one: more than half of Notion Mail users were already handling email through agents without opening the inbox, so Notion is going all-in on agents instead.
More importantly, most of the data was never Notion’s to delete. Your mail lives in Gmail and stays there. What you have to export by September 21 is the Notion-only layer: drafts, scheduled sends, snippets, auto-label instructions. Teams relying on HIPAA coverage were told to transition off by June 30.
That’s the difference between a tool that sat on top of your system of record and one that was your system of record. Notion Mail was a client over somebody else’s mailbox, so its death is an inconvenience. Relay held your workflow logic, and Clockwise held your scheduling rules, and neither of those existed anywhere else.
The product acquisition that survives: Reclaim and Superhuman
The contrast cases matter, because “acquired” is not a synonym for “doomed” and treating it that way will make you migrate off perfectly healthy tools.
Dropbox acquired Reclaim in August 2024. Nearly two years on, Reclaim is still open, still shipping, and has grown to more than 320,000 users across 60,000+ companies — it added Microsoft Outlook support in May 2025 and shipped Slack auto-reply and team OOO features in June 2026. Grammarly acquired Superhuman in July 2025 and then, in October 2025, renamed the whole company Superhuman. The product didn’t just survive the acquisition; it ate the acquirer’s brand.
The distinction to watch for in the announcement: does the acquirer describe the product’s roadmap, or only the team’s mission? Clockwise’s announcement talked about Agentforce. Reclaim’s talked about Reclaim. That single tell, available on day one, predicted both outcomes.
The quiet wind-down: Relay.app
Relay is its own category: no acquisition reported, no destination company named, no reason given, no press coverage. Just a notice.
And it deserves credit, because it handled the ending about as well as the format allows. Free users got 30 days; paying customers got 60 days of continued access at no charge, plus a bonus of 25,000 steps and 10,000 AI credits per month during the window. Existing workflows keep running through the wind-down. Annual customers get automatic prorated refunds, promised within five business days. Support continues, with priority for paying customers, and the notice explicitly asks business-critical teams to reach out early. Compare that to eight days and a deletion.
The notice reads, in part:
“We’ve worked closely with many of you for years, and we know this change will be disruptive. We are very sorry for the work this will cause you, and we don’t take it lightly, so our top priority is to make this transition as seamless as possible.”
That’s the right way to do a hard thing. It also doesn’t change the deadline.
The endings, compared
| Tool | Ending | Notice | What happened to data | Refunds |
|---|---|---|---|---|
| Clockwise | Acquihire (Salesforce) | ~8 days | Deleted; nothing transferred | Automatic prorated |
| Nerve | Acquihire (OpenAI) | ~1 month | Not publicly detailed | Not publicly detailed |
| Mindy | Asset license + team hire (WorkWhile) | Not publicly detailed | Not publicly detailed | Not publicly detailed |
| Notion Mail | Strategic pivot to agents | ~3 months | Mail stays in Gmail; export the Notion-only layer | N/A |
| Relay.app | Wind-down, no reason given | 30 days free / 60 days paid | Deleted after window; JSON + CSV export offered | Automatic prorated, annual |
| Reclaim | Product acquisition (Dropbox) | — | Still operating, still shipping | — |
| Superhuman | Product acquisition (Grammarly) | — | Still operating; acquirer took the name | — |
“We have an export button” is close to meaningless
This is the part most buyers get wrong, and Relay is the clean illustration.
Relay’s export is genuinely more than most wind-downs offer: workflows, Sequences, and MCP servers as JSON and AI prompts, plus run history and Tables as CSVs. Read that list again, though. Nothing on earth ingests Relay’s workflow JSON. There is no importer. It is a rebuild aid — a precise description of what you had, which you or an LLM can use to reconstruct it by hand somewhere else. That is worth something. It is not portability.
(The irony is sharp and worth stating without malice: the lack of workflow export/import was a long-running complaint from Relay’s own users. It shipped at the end.)
So the question to ask a vendor is never “can I export my data?” The answer is always yes, and it’s always technically true. The questions that actually discriminate:
- What format, and what else reads that format? A proprietary JSON schema that only your product parses is a souvenir. A CSV of records is data. An open standard is portability.
- Does the export include the logic, or only the rows? Most tools export your records and quietly drop your automation — the triggers, branches, and conditions that were the entire reason you paid.
- Does it include credentials? It cannot, and shouldn’t. Relay is explicit that connected-app tokens are deleted. Which brings us to the expensive part.
Automation is the worst category to migrate out of
Moving a note-taking app is annoying. Moving an automation platform is a project, for a reason specific to this category: every integration has to be re-authorized by hand, by a human, one OAuth consent screen at a time.
Your workflow logic might be recoverable from an export. Your connections never are. If you had Relay wired into Gmail, Slack, HubSpot, Airtable, Stripe, and a dozen more, the new tool needs all of those re-connected, re-scoped, and re-tested — and in a lot of companies, half of those authorizations require an admin who doesn’t report to you and doesn’t consider this urgent. Then every workflow needs to be re-tested against live systems, because an automation that silently fails is worse than one that never existed.
Multiply by the number of workflows. This is why the 60-day window matters more than the export format, and why Relay’s notice tells business-critical teams to reach out early. It’s good advice. Take it literally.
Why this market is brutal, structurally
None of the above implies anyone did anything wrong. It’s worth being precise about the structural problem, because it explains the pattern better than any story about individual companies.
Workflow automation is a business where the moat is integration breadth, and breadth is a treadmill. Zapier has something in the neighborhood of 8,000 integrations. Every one of them is a connector somebody has to build and then maintain forever as the upstream API drifts. A new entrant with 200 excellent connectors — which is what Relay had, alongside 300+ app-based triggers — is genuinely better in most respects and still loses the evaluation to the incumbent whenever a team’s niche CRM isn’t on the list.
And that was, measurably, Relay’s most common user complaint: the catalog felt thin next to Zapier’s, and teams with unusual SaaS hit walls. This is not a criticism of how Relay built its product. It’s a description of a market where the differentiated thing you built — Relay’s human-in-the-loop approvals were genuinely distinctive, free on every plan including the free tier, with no premium-connector gating — competes against a commodity everyone has more of. You can be better and still be structurally disadvantaged. That’s the trap, and it doesn’t have a clever escape.
Buying automation like it might end
Concretely, then. Not “avoid startups” — the incumbents shut things down too, and Notion Mail is a $10B+ company’s product going away with three months’ notice.
Document your workflows outside the vendor. A one-page plain-language description of each workflow — trigger, steps, conditions, what it touches, who owns it — is the single highest-return habit here, and it costs an hour. Every migration in this post is dramatically easier for the team that has this and brutal for the team reconstructing intent from a canvas they can no longer log into. This is the thing to do today, not the tool to switch to.
Know which tier of dependency you’re in. A tool that sits on top of your system of record (Notion Mail over Gmail) is a survivable loss. A tool that is the system of record for its own logic (any automation platform) is not. Price the risk accordingly, and be more conservative the deeper the tool sits.
Keep the crown jewels somewhere neutral. If a workflow’s output is the asset — leads, records, run history for compliance — land it in a database, warehouse, or sheet you control rather than the vendor’s built-in tables. Relay’s Tables exported to CSV, which is the good outcome. Don’t count on that.
Consider self-hostable options as insurance, with the caveat. Activepieces publishes its Community Edition under the MIT license (enterprise features are commercial) and is self-hostable via Docker; it also does human-in-the-loop approvals, which makes it a real conversation for stranded Relay users. n8n is the other obvious one, but describe it accurately: it’s under the Sustainable Use License, not an OSI open-source license. You may self-host and modify it for your own internal business purposes; you may not sell it as a service to others. That’s a meaningful distinction, and it’s still a genuine hedge — code you can run yourself cannot be deleted out from under you on a 30-day notice.
The honest limit on that argument: self-hosting trades vendor risk for operational risk. Somebody now owns upgrades, connector breakage, and the 2am failure. For most small teams that trade is worse, not better, and pretending otherwise would be convenient for nobody. It’s insurance, and insurance has a premium. We’re not going to twist that into a reason to buy from us — if self-hosting is the right answer for your risk profile, it’s the right answer.
Read the announcement for who’s accountable. Product roadmap named, or team mission named? Reason given, or absent? Data destination specified, or unmentioned? Those three questions, applied to any of the notices above, would have told you the outcome on day one.
What Relay users should do in the next two weeks
The window is real and the arithmetic is unforgiving. Free accounts and all their data are deleted August 15, 2026 at 23:59 PT — that’s under a month. Paying customers have until September 14. New signups and free-to-paid upgrades are already off, so a free user cannot buy their way into the longer window.
Export everything now, before you’ve decided where you’re going: workflows, Sequences, and MCP servers as JSON and AI prompts, run history, and Tables as CSVs. The export is a rebuild aid, so its value is highest while the product is still running and you can still look at the thing you’re describing. Then write the plain-language description of each workflow while you can still open it. Then pick a destination — we mapped the options in Relay.app alternatives, the free ones in free Relay.app alternatives, and the deadline detail in Relay.app is shutting down.
Do it in that order. The tool you migrate to is a reversible decision. The data deleted on August 15 is not.
Ready to automate your busywork?
Carly schedules, researches, and briefs you—so you can focus on what matters.
See what people say
"Before Carly, I relied on a Calendly link, but the whole process felt impersonal and not very professional. Carly changed that by handling all the back-and-forth, so I'm no longer stuck in endless email threads trying to line up schedules.
Now Carly reaches out to candidates, shares my real-time availability, lets them pick a slot, then sends a Zoom link and drops it straight into my calendar. She sends reminders to both of us before each call, which has significantly reduced no-shows and last-minute confusion.
On top of scheduling, Carly acts like a full executive assistant, sending me my schedule the night before so I can prepare for each call. It reminds me of the old x.ai assistant, but Carly is noticeably smarter, faster, and better suited to my healthcare recruitment business."
