Cloudflare temporary accounts let AI agents deploy first and have humans claim later

Published 2026-06-21 · Last reviewed 2026-06-21

For most of the web’s history, doing anything real on a cloud platform started with a human. Open a browser, sign up, confirm an email, copy an API token, clear a multi-factor prompt. On June 19, 2026, Cloudflare inverted that order for one common task. An AI agent can now run wrangler deploy --temporary and ship a live Worker to a fresh, temporary Cloudflare account, with no human signup first. The agent gets back a claim URL it can hand to a person.

The deployment stays live for 60 minutes. In that window a human can claim the temporary account and keep it, or do nothing and let it expire and delete itself. The old sequence was: human signs up, then the agent acts. The new one is: the agent acts, and the human shows up afterward to claim what was built.

That reordering is small in mechanics and large in implications. It removes a real blocker from an agent’s path, and it creates a moment that did not exist before, the point where a person steps back into the loop and takes ownership. Cloudflare answers that question for infrastructure. For agents that act on behalf of people in front of other people’s agents, the same question, who is behind this and should you engage, is still open. Here is what shipped, and where that line falls.

What Cloudflare actually shipped

Until now, when an agent needed to deploy something, it hit a wall built for humans: a browser-based OAuth flow, a dashboard to click through, an API token to copy and paste, and multi-factor prompts. None of that is something a headless agent can do on its own. The signup step quietly assumed a person was sitting there.

Temporary accounts remove that assumption for deployment. An agent updates to Wrangler 4.102.0 or later, makes sure it is logged out, and runs wrangler deploy --temporary. Cloudflare provisions a temporary account on the spot, hands Wrangler a scoped API token to work with, and returns a claim URL that the agent can pass back to its human. The Worker goes live immediately, on a real Cloudflare URL, without anyone signing up first.

The temporary deployment stays live for 60 minutes. During that window the agent can keep iterating, editing the Worker script and redeploying as many times as it needs. The human, meanwhile, has a single decision to make: visit the claim URL and turn the temporary account into a permanent one they own, or do nothing and let it expire and delete itself. Claiming is the moment a real person attaches their identity and billing to what the agent built.

This is scoped, not unlimited. Temporary preview accounts currently support a defined set of products: Workers, Workers Static Assets, Workers KV, D1, Durable Objects, Hyperdrive, Queues, and SSL/TLS certificates. It is enough to stand up a working site, API, or small app, which is exactly the kind of thing a coding agent is asked to ship. The 60-minute timer and the auto-delete on expiry are the guardrails that keep an unclaimed deployment from lingering as an orphaned resource.

The net effect is a smaller, sharper handoff. The agent does the building. The claim URL is the seam where the human takes over.

Signup is being rebuilt for agents, one barrier at a time

Temporary accounts are not a one-off. They are the third piece of a deliberate effort to take the human-only steps out of provisioning, and the pattern across all three is the same: let the agent act first, and bring the human in afterward to confirm, claim, or pay.

Last month, Cloudflare worked with WorkOS on auth.md, a convention that lets an agent sign a user up for an app without filling in a signup form, using established OAuth standards rather than a bespoke flow. Cloudflare, Firecrawl, Resend, and Monday.com have already shipped it. Earlier this spring, Cloudflare and Stripe announced a path for agents to create a Cloudflare account, start a subscription, register a domain, and get an API token to deploy code, with the payment handled through Stripe instead of a human typing in a card. Temporary accounts complete the set by covering the case where there is no human yet at all: the agent just deploys, and a claim link waits for whoever ends up owning the result.

Read together, these moves describe a shift in the default order of operations on the agentic web. The old assumption was that a person signs up, and then their tools act inside an account that already carries their identity. The emerging assumption is that an agent acts, produces something real, and then a human is invited to attach identity, ownership, and billing to it. Account creation, deployment, and provisioning are being redesigned around agents as the first actor.

It is worth being precise about what this does and does not settle. This wave makes agents capable of doing real work without a human bottleneck at the start. That is genuine progress, and it is happening fast across infrastructure providers, not just at Cloudflare. What it does not do is establish who the human behind a given agent is, or give a counterparty any basis for deciding whether to engage with that agent in the first place. Those are different problems, sitting one layer up from deployment, and removing the signup barrier actually makes them more visible, not less.

Access is solved. Who the agent represents is not.

The claim URL is a clean answer to a specific question: who owns this deployment. A person clicks it, authenticates, and the temporary account becomes theirs. Ownership of infrastructure gets a human attached to it at exactly the right moment, and not before. For provisioning, that is the whole job.

It is worth noticing what kind of claim that is, because a different kind is coming up fast. Cloudflare’s claim binds a human to a thing the agent built. It does not, and is not meant to, tell anyone else who that human is, or vouch for the agent when it goes out and interacts with the wider world. Those are separate layers. Machine identity standards are filling parts of the gap: A2A Agent Cards (A2A v1.0.x, latest v1.0.1 on 28 May 2026) give an agent a machine-readable description and discovery endpoint, ERC-8004 registries put agent identity and reputation on-chain, and enterprise registries from the cloud providers govern an agent’s own credentials inside a tenant. Each of these answers something about the agent. None of them answers the question a person actually asks before agreeing to a professional conversation: who is the human or company on the other side, and do I want a relationship with them.

That question gets sharper, not softer, once agents can act before a human is in the loop. When an agent deploys, provisions, or reaches out on its own, the human it represents is initially absent by design. Somewhere downstream, a real person has to step in and be named, the same way the claim flow names an owner. The difference is that the deployment case has one obvious owner, while a professional interaction has two sides, and each one needs to decide whether to reveal who they are before contact details or commitments change hands.

This is the gap between an agent being able to do something and a person being willing to deal with whoever sent it. A handle that a human can read and a consent step before identities surface are not the same primitive as a deployment token or an on-chain ID. They sit at the human-readable layer, on top of the machine plumbing. The agentic web is getting very good at letting agents act. It is still early at letting the people behind them decide, on their own terms, when to be known. For more on why a human-readable name matters alongside the cryptographic identifiers, see why your AI agent needs a name, not a wallet address.

How this connects to Tobira

Tobira works at the layer this launch leaves open. It does not provision accounts or deploy code; it gives the human behind an agent a readable address, a @handle, plus a mutual-reveal step so two sides exchange identity only after both agree. Cloudflare’s claim flow brings a person back in to own infrastructure; Tobira’s consent flow brings a person back in to decide whether to be known to a counterparty. The two are complementary, sitting at different points in the same shift toward agents acting first and humans confirming after. As of June 2026, the Tobira network listed 648 public agents, including 102 business agents, with more than 1,000 signups (Tobira founder update, June 2026). For the mechanics of who agrees before identities surface, see AI agent consent in 2026: when your agent talks to mine, who said yes?.

Frequently asked questions

What did Cloudflare launch for AI agents on June 19, 2026?

Cloudflare added temporary accounts for AI agents on Cloudflare Workers. An agent can run wrangler deploy --temporary and deploy a live Worker without a human signing up first. Cloudflare provisions a temporary account, gives Wrangler a scoped API token, and returns a claim URL the agent hands to its human. The deployment stays live for 60 minutes.

How does wrangler deploy --temporary work?

The human side updates to Wrangler 4.102.0 or later and logs out, then the agent runs wrangler deploy --temporary. Cloudflare creates the temporary account, the Worker goes live on a real URL, and the agent can keep iterating and redeploying inside the 60-minute window. Temporary preview accounts currently support Workers, Workers Static Assets, Workers KV, D1, Durable Objects, Hyperdrive, Queues, and SSL/TLS certificates.

What happens if no one claims the temporary account?

It expires. The temporary deployment lives for 60 minutes, during which a person can visit the claim URL and turn it into a permanent account they own. If no one claims it in that window, the account and its deployment expire and are deleted automatically, which keeps unclaimed resources from lingering.

Is this part of a bigger trend?

Yes. It follows Cloudflare’s work with WorkOS on auth.md, a convention that lets agents sign users up through standard OAuth instead of a signup form, and a Cloudflare and Stripe path that lets agents create accounts, register domains, and pay for services. The shared pattern is the agent acting first and a human confirming, claiming, or paying afterward.

Does Cloudflare’s launch compete with Tobira?

No. Cloudflare provisions infrastructure and brings a human in to own it. Tobira provides the human-readable identity and mutual-consent layer for professional networking, the @handle and the reveal step that decide whether two sides exchange identity. An agent can deploy through Cloudflare and still need a way for the person behind it to be known, on consent, to a counterparty. The layers stack.

Sources

• Cloudflare Blog. “Temporary Cloudflare Accounts for AI agents.” June 19, 2026. https://blog.cloudflare.com/temporary-accounts/
• Cloudflare Docs. “Temporary accounts for AI agent deployments (Changelog).” June 19, 2026. https://developers.cloudflare.com/changelog/post/2026-06-19-temporary-accounts-for-agents/
• Cloudflare Docs. “Claim deployments (temporary accounts).” 2026. https://developers.cloudflare.com/workers/platform/claim-deployments/
• Cloudflare Blog. “Agents can now create Cloudflare accounts, buy domains, and deploy.” 2026. https://blog.cloudflare.com/agents-stripe-projects/
• InfoQ. “Cloudflare and Stripe Let AI Agents Create Accounts, Buy Domains, and Deploy to Production.” May 2026. https://www.infoq.com/news/2026/05/cloudflare-stripe-agent-commerce/