Replicas brings sandboxed coding agents

Replicas is trying to be the operating layer for agentic engineering work, not just another chat wrapper with a code editor attached. Its strongest angle is workflow-first automation: assign the task, let the agent work in a real sandbox, then review a pull request.

• –The sandboxed VM story matters because it makes agent execution feel operationally safer than local prompt-and-pray tooling
• –Slack, Linear, GitHub, and dashboard triggers cover the places engineers already work, which lowers adoption friction
• –Claude Code and Codex support makes this feel model-agnostic instead of betting on one assistant
• –The “ship over 30% of pull requests” claim is the right kind of proof point if it holds up in production
• –This sits in a crowded category, but the integration layer and reviewable PR output are what could make it stick