Agent Hosting Is Becoming Free. Here's What Isn't.
Anthropic now ships ways to keep Claude Code running without your laptop — Remote Control, cloud sessions, scheduled Routines. That's good news, and it changes what's worth paying for. The session is becoming a commodity. The environment the work ships into is not.
A year ago, "run Claude Code somewhere that isn't my laptop" was a real product gap. People built VPS setups with tmux, rented dedicated agent machines, and paid startups to host sessions for them.
Then Anthropic shipped it. Claude Code runs in Anthropic's cloud from the web and the Claude apps. Remote Control lets you drive a local session from your phone. Routines, released in April 2026, run a saved Claude Code configuration on a schedule, on an API call, or on a GitHub event — on Anthropic's infrastructure, with your laptop closed.
If you're building a company on top of Claude Code, the obvious question is: why won't Anthropic just build whatever you're selling?
It's a fair question. We think the honest answer is more useful than a defensive one, so here it is.
What Anthropic is building, and why it's right
Anthropic's features all point the same way: the agent session should not depend on your device. Start work anywhere, check it from anywhere, run it on a schedule.
That's the correct thing for them to build, and it's good for everyone. It means the baseline — an agent that keeps working while you're away — is now included with a plan you already pay for. Anyone whose product was only "we keep your Claude Code session alive" has had their product absorbed into the platform, and they should have seen it coming.
So let's be precise about what's becoming a commodity: the running session. The loop of prompt, tool call, result, next step, on someone else's computer.
What a session produces, and where it has to go
Here's the gap. A session produces changes. A pull request, a diff, a set of commits. That's the right output for a lot of work — fix this bug, add these tests, update this dependency — and for that work, Anthropic's features are close to complete.
But a lot of software work isn't finished when the pull request opens. Someone has to answer:
- Where does the app actually run? Not the code — the running process, with a URL, that a user can reach.
- Where does the data live? A real Postgres instance with real rows in it, that survives the next deploy and has a backup from last night.
- What's the address? A domain the customer owns, pointed somewhere, with a certificate that renews itself.
- How do I look at this branch? A live URL for the change, so someone who doesn't read diffs can click through it.
- What does it need to talk to? A queue, a cache, a second service, all running together, all configured.
None of that is the agent session. It's the environment the session's work ships into. And it's the part that makes the difference between "Claude wrote a feature" and "the feature is live."
Why a model company is unlikely to become your hosting provider
We could be wrong about this — predicting what a fast-moving company will build is a good way to look foolish. But there are structural reasons to think Anthropic won't become the place your application runs.
It's a different business. Hosting production applications means uptime commitments, data residency questions, backup and restore guarantees, abuse handling, DNS and certificate management, and on-call engineers for someone else's database at 3am. It's operationally heavy, low-margin, and nothing like training and serving models.
It's a crowded, mature market. Vercel, Railway, Render, Fly, Netlify, the hyperscalers — there are excellent places to deploy software. A model provider has little reason to compete with its own customers there.
Neutrality is valuable to them. Claude Code working well with every hosting platform is a better position for Anthropic than Claude Code working best with one.
So the likely shape of the future is: Anthropic makes the agent great and makes the session run anywhere, and the application lives somewhere else.
Where the wedge actually is
If the session is a commodity, the remaining value is in the join between the agent and the running application. Specifically:
- The agent working against the real environment. Not a fresh sandbox per task, but the same workspace the app runs in, with the database migrated and the services up. An agent that can run the actual app, hit the actual endpoint, and read the actual error produces better work than one reasoning about code it can't execute. (More on that in Why AI Coding Agents Need Persistent Workspaces.)
- State that survives. Dependencies, build caches, data, and the agent's own notes, still there tomorrow.
- The path from change to live. A preview URL per branch, then production, without a separate deployment pipeline to build and babysit.
- The surrounding infrastructure, managed. Database, backups, SSL, domains, secrets — the things that aren't interesting until they break.
That's a less exciting pitch than "AI agents in the cloud." It's also the part that doesn't get absorbed when the model provider ships its next feature.
What this means if you're choosing tools
A practical way to think about it:
- If your work ends at a pull request and you already deploy somewhere you like, use Anthropic's own cloud features. They're good, they're included, and you don't need more.
- If you need the agent to work against a running app with real state — or you don't have a deployment story at all — you need an environment, not just a session. Build one on a server, or use a platform that provides it.
- Be sceptical of anyone whose whole product is session hosting, including us, if that's all we were. That layer is thinning fast.
Why we're publishing this
It would be easier to pretend Anthropic's features don't exist. But anyone evaluating a product like ours has already asked the question, and a vendor that dodges it is telling you something.
We stopped building session hosting because Anthropic does it well. We build the environment the session's work lives in — and we'd rather you choose us knowing exactly where that line is.
EasySpawn is the managed infrastructure your app runs on: persistent workspaces, managed Postgres, automatic SSL, a live URL on every branch, and your own domain connected — with Claude Code working inside it, on your own Claude subscription. See how it works or join the waitlist.
Related: What Is an Autonomous Development Platform? · How to Use Claude Code From Your Phone
Keep reading
A Security Checklist for Vibe-Coded Apps
AI-built apps fail security in predictable ways: open databases, keys in the browser, authorization checked only in the UI. A practical checklist for non-security people — what to check, how to test it yourself, and what to fix before real users arrive.
How to Stop an AI Agent From Deleting Your Production Database
In July 2025 an AI coding agent deleted a company's production database during a code freeze. It wasn't a freak event — it was the predictable result of giving an agent production credentials. Six controls that make it structurally impossible, not just unlikely.