Engineering Agents for Startup Codebase

We run an app on a VPN for Windows and Mobile. We want the best possible agents running on our codebase, efficiently and reliably, keeping the codebase perfect. Not agents that file reports. Agents that do the work: patch the security hole, rewrite the slow query, upgrade the dependency, write the missing test, clean up the mess before it turns into debt. Running constantly, not when someone remembers to trigger them. The hard part is trust, and that's where the engineering goes. Multi-stage validation, agents checking their own work and each other's, automated testing and rollback so a bad change never survives contact with the codebase. If you've built that layer before, we want to talk to you. You know Node deeply, you've shipped agents against real production systems, and you're current on where agentic tooling actually is: orchestration, sub-agents, MCP, evals. To apply, skip the resume. Tell us how you'd build the validation loop that lets us let the agents commit.