Launch HN: Hoplite (YC S26) – Effortlessly deploy cloud coding agents
hoplite.sh BenceRed ● Covered by 2 sources
A YC startup called Hoplite launched a tool that runs coding agents in the cloud instead of on your laptop. It ports your dev setup over so you spend time reviewing what the agent built, not babysitting the code.
Bence and Ryan didn't start out building a cloud agent platform. Their original YC pitch was an AI tool for retail investors, but they scrapped it once they realized neither of them would actually use the thing, nor did they feel any pull toward the customers it served. What they did feel strongly about was talking to developers and founders, and they had opinions — strong ones — about the mess that is running coding agents at scale. So they pivoted, and Hoplite is the result.
The pitch is straightforward: deploy coding agents in the cloud, and make it painless to QA whatever those agents produce. During onboarding, Hoplite copies over your local setup — sessions, memories, MCP servers, whatever context your agent needs — so it can run projects remotely without you rebuilding your whole environment from scratch. The founders looked at existing cloud-agent tools first and came away unconvinced; none of them, in Bence and Ryan's view, actually exploited what running in the cloud makes possible, and most felt sluggish to use day to day.
Rather than wrap Codex or Claude Code, Hoplite built its own agent harness from the ground up. That's a deliberate bet: owning the harness means they're not stuck waiting on Anthropic or OpenAI to ship a feature they need, and they can experiment freely. Their infrastructure choices lean the same direction — AWS for the bulk of it, but Temporal for durable workflows, Modal for sandboxing, and Planetscale for the database, all picked because the founders think agents are turning into tier-0 infrastructure that needs to behave like it.
The bigger thesis here is about where coding is headed. Bence and Ryan expect that within six to twelve months, developers will spend less time reading diffs line by line and more time judging the actual output — does the new flow make sense, does the feature look right, does the CLI still work on Windows — often while dozens or hundreds of agents run in parallel. Hoplite is betting its product on that shift being real and imminent.
Right now anyone can try it free with $100 in credits using the code HACKERNEWS, and it supports plugging in an existing Codex subscription to run OpenAI models through the platform. The founders say they're mostly heads-down on two things at the moment: onboarding and preview environments, and they're actively asking users to flag any task where the agent underperforms.
My take
Betting the whole product on a custom agent harness instead of wrapping Codex or Claude Code is either a smart differentiation play or a slow-motion tax on engineering time that the big labs will erase with their next release. History leans toward the latter — plenty of startups have burned a year building around a capability gap that OpenAI or Anthropic closed in a single update. Still, the framing that developers will soon review outcomes instead of diffs is the right read on where this is all going, and it's the part of the pitch worth paying attention to, not the infra stack.
Read more about this at: hoplite.sh
