Every corner of dev tooling has a hot category right now, and cloud coding agents are it. The pitch is irresistible: hand an agent a ticket, let it work in a cloud sandbox while you do something else, get a pull request back. Codex Cloud, Cursor's background agents, Devin, and Jules all sell some version of it. And everyone who has actually tried them has hit the same wall on day one: the cloud box is not your box.

Hoplite, a just-launched YC S26 company, is going straight at that wall. This is an early-stage launch, not a settled product — so treat what follows as "here's what's confirmed and what to ask," not a buy recommendation.

What it is#

Hoplite runs autonomous coding agents in cloud sandboxes. You give it a task; it works asynchronously in its own isolated environment and comes back with changes. That much is the same shape as every other cloud agent.

The wedge is environment parity. Hoplite's stated pitch is that onboarding transfers your local sessions and MCP servers to the cloud and configures the CLIs and dependencies your projects require — so the sandbox behaves like your laptop instead of a bare container you have to reprovision. If that claim holds up on a real, messy dependency tree, it targets the single most common reason cloud agents frustrate people: they fail not because the model is dumb but because the environment is missing the one tool the build needs.

The hard part of a cloud coding agent was never the model. It's that "works on my machine" doesn't survive the trip to someone else's machine. Hoplite is betting the environment, not the agent, is the product.

Who it's for#

The clearest fit is a founder or small team already living inside a coding agent — Claude Code, Codex, or similar — who keeps running into the "I can't reproduce my setup in the cloud" problem, and who wants to dispatch work without opening a terminal.

That second part is Hoplite's other bet: triggers beyond the CLI. You can start a task manually from Slack, Linear, or iMessage — dispatch a fix from your phone — or fire one automatically from a Sentry alert, so an incoming production error can spin up an agent to investigate before a human even looks. The founders (who pivoted into this after an earlier YC idea in AI retail investing) frame the ambition as helping teams "deploy trillions of tokens a day" while keeping the output genuinely good. That's a throughput story: lots of small, low-priority tickets closed continuously, in the background.

If you want a mental model, it sits next to the Claude Code background-agent / draft-PR review-queue pattern — async agents producing reviewable diffs — but aimed at portability of your whole setup rather than one vendor's ecosystem.

What isn't known yet — and why it matters#

Because this is a launch, several things a founder normally needs before depending on a tool are not public:

None of these are red flags — they're just unshipped-documentation gaps normal for a launch. They're the checklist for an access conversation.

How to start#

Hoplite is at the request-access stage: the YC launch page and Launch HN thread are the front doors, and @hoplite_sh is posting updates. The sensible trial: point it at a low-stakes repository with a genuinely annoying dependency setup, confirm the environment-parity claim survives contact with your real toolchain, wire one Sentry or Linear trigger, and judge the resulting diffs the way you'd judge any terminal coding agent — on whether the PR is actually mergeable. If the cloud box really does show up as your box, that's the part nobody else has made effortless yet.