Hoplite, a startup backed by Y Combinator, has launched a platform that lets AI write and fix code for you, running entirely in the cloud rather than on your own computer. Each AI agent gets its own separate machine, set up with a copy of your code, with enough power to handle large projects. The founders, Bence and Ryan, announced the product through Y Combinator's launch page and a post on Hacker News. (Hoplite)

Each of these separate machines can be configured with up to 8 processor cores, 32 GB of memory, and 10 TB of storage. The platform works with AI models from several companies, including Anthropic, OpenAI, and providers that share their models for free. You can pick a different AI model for each task. If you already pay for an OpenAI Codex subscription, you can connect it and use it through Hoplite. (Hoplite)

The AI agent works through a set series of steps: it reads your code, makes changes, runs tests to see if things work, checks any failures, tries again if needed, and then waits for a person to approve before submitting the changes. Each machine also runs your app on a live web page that both you and the AI can open in a browser, so the agent can see the results of its work, not just read text from a command line. (Hoplite)

Hoplite connects to several popular tools: GitHub for code storage, Linear for tracking tasks, Slack for messaging, and MCP, which is a standard for connecting AI tools to outside data sources. The platform's command-line tool can import your existing settings from other AI coding tools like Claude Code or Codex, including your saved sessions and configurations. You can start a coding task by sending a message in Slack, Linear, or iMessage. An Automations feature can also start agents automatically based on web alerts or a schedule. If you use Sentry, which tracks software errors, Hoplite can automatically turn those errors into tasks for the AI to fix. (Hoplite, Y Combinator Launches)

You pay for Hoplite using prepaid credits. Each coding run is priced individually, and you can see what each one costs. The Scale plan allows up to 150 AI agents running at the same time per person. At launch, Hoplite offered a free trial with $100 in credits using the code "HACKERNEWS." (Hoplite, Hacker News)

Behind the scenes, Hoplite runs on Amazon's cloud servers and uses several tools to keep everything running: Temporal to manage long-running tasks, Modal to set up the separate machines, and PlanetScale as its database. The founders built their own system for controlling how the AI agent works, rather than using existing tools like Codex or Claude Code. (Hacker News)

The company did not start here. Bence and Ryan originally applied to YC with an AI product for retail investing but abandoned it because, as they wrote, it was not a product they themselves would use. The pivot to cloud coding agents produced Hoplite's current offering. (Hacker News)

The design choices here are worth understanding. Giving each AI agent its own separate machine costs more in computing resources than sharing one, but it avoids problems where one task slows down another or creates security risks. The large 10 TB storage space per machine suggests the platform expects agents to work with very large codebases. And the live web page preview, where the AI can open and interact with the app it is building, solves a real problem: without being able to see the results, an AI working on a website or app interface is essentially guessing whether its changes look and work correctly.

Building their own system instead of using existing tools like Claude Code or Codex has tradeoffs. Ready-made tools are faster to launch and get updated by someone else. A custom system gives Hoplite full control over how the agent works, how it retries failed tasks, and how it connects to other services, but the team has to keep up with rapid changes in AI models on their own. Supporting models from multiple companies makes that harder, since each company's software works differently.

The scale is ambitious for a new company. Allowing 150 agents to run at the same time per person means a small team could have hundreds of AI tasks going in parallel. Whether all those agents produce useful work depends on how well the system can coordinate them and avoid conflicts when multiple agents try to change the same code at the same time.

The pricing approach, where each run shows its cost, addresses a genuine problem. People using AI coding tools have struggled to predict how much they will spend, since some tasks take the agent many attempts and others just a few. Seeing the cost of each run helps teams learn what different types of tasks typically cost. The prepaid credit system, though, means teams have to guess how much they will use overall before they start.

The bigger picture here is whether AI-assisted coding moves from running inside a developer's own editor, like GitHub Copilot, to running as cloud-based agents that can work independently. Cloud-based agents can use more computing power than a laptop provides, run for longer without tying up your machine, and connect more easily to other tools in a company's workflow. Hoplite's connections to task trackers, chat platforms, error monitors, and code storage position it as a system for continuous, automated development rather than a tool that helps you type code. Whether that approach catches on depends on whether the AI's work has become good enough that having a person review its changes is faster than writing the code from scratch, a bar that depends heavily on how complex the task is.