Proliferate Wants Coding Agents to Run While You Sleep

The Macro: Coding Agents Need Infrastructure, Not Just Prompts

I keep hearing the same thing from engineering leaders: “AI writes 40% of our code now.” The number varies. Some say 30%. Some say 60%. But the direction is clear. Coding agents are producing real, shippable code at a pace that would have seemed absurd two years ago.

Here is what nobody says in the same breath: most of that code still requires a human to review it, test it, deploy it, and monitor it in production. The agent writes the pull request. A human checks if it breaks anything. A human merges it. A human watches the deploy. The bottleneck has shifted from writing code to everything that happens after writing code.

This is the infrastructure gap in AI-assisted engineering. Tools like Cursor, Copilot, and Claude Code have made writing code faster. Nobody has made deploying agent-written code safer. When a human writes a PR, they have context about the production environment, the deployment pipeline, and the downstream systems that might break. An agent has none of that context unless you explicitly give it.

The companies that are furthest along with agent-written code, Stripe and Ramp get cited frequently, are merging 50% or more of their PRs from agents. But they have invested heavily in internal tooling to make that work. Custom sandboxes, automated verification, staging environments that mirror production. That tooling is not available to most engineering teams, which means most teams are stuck in a half-automated state where the agent writes code and then a human spends thirty minutes making sure it is safe to ship.

Devin tried to solve this with an all-in-one autonomous engineer. Factory does something similar. But both are opinionated about which agent you use. If your team already has workflows built around Claude Code or Cursor, you do not want to rip those out and replace them with a different agent. You want infrastructure that works with whatever agent your team has already chosen.

The Micro: Open-Source, Self-Hostable, MIT Licensed

Proliferate gives coding agents their own Docker-based sandboxes with access to your internal services, configurable event triggers, and verification workflows. Your agent writes a PR, Proliferate deploys it to an isolated environment, runs tests, does a thirty-minute soak test, and surfaces the results for team review. The whole thing is open-source and MIT licensed.

Pablo Hansen is the sole founder. He completed a master’s degree in AI at 19 and was the first hire at Onyx, another Y Combinator company from the Winter 2024 batch. The team is two people, based in San Francisco, part of YC’s Summer 2025 batch. Two people building infrastructure tooling is either very lean or very stretched, depending on how you look at it.

The product does several things that matter. First, event-triggered automation. When a Sentry alert fires or a Linear ticket is created or a Slack message hits a specific channel, Proliferate can spin up an agent to respond. That turns your coding agent from a tool you invoke into a background worker that handles routine engineering tasks autonomously. Second, sandboxed environments that mirror your actual infrastructure. The agent gets its own copy of your API, database, and Redis instance. It can make changes without touching production. Third, verification workflows with audit trails. Every change gets diffs, test results, and soak test data before a human has to make a decision.

The integrations list covers the tools most engineering teams already use. GitHub, Sentry, Slack, Linear, Jira, Notion, Cursor. The no-code configuration means PMs and designers can set up workflows without writing code, which is ambitious but potentially valuable for cross-functional teams.

The self-hosting angle is important. Most companies with serious engineering teams do not want their code running through a third-party SaaS product. MIT licensing means you can deploy Proliferate on your own infrastructure and modify it however you want. That is a meaningful differentiator against closed-source alternatives.

The Verdict

Proliferate is building the plumbing that makes autonomous coding agents practical for real engineering teams. The bet is that the future of software development looks like agents running in the background, shipping code continuously, with humans reviewing and approving rather than writing. If that future arrives, every team needs infrastructure like this.

The risk is that this is a two-person team competing against well-funded companies and against the agent platforms themselves. Anthropic could build sandboxing and deployment features directly into Claude Code. Cursor could add environment management. If the agent platforms vertically integrate into infrastructure, standalone tools get squeezed.

In thirty days, I want to see GitHub stars and forks as a proxy for developer interest. Sixty days, I want to know whether any mid-sized engineering teams are running Proliferate in production. Ninety days, the question is whether the open-source community is contributing meaningful features or whether this is a two-person project with an MIT license and no ecosystem. The problem is real. The approach is sound. The execution risk for a two-person infrastructure company is significant, but the open-source strategy gives them a distribution advantage that closed-source competitors do not have.