Build or buy an agent developer workspace?
8 hours ago
- #AI Agents
- #Developer Productivity
- #Cloud Workspaces
- Monaco initially experimented with a 'Million Interns' initiative using tmux and MCP servers to manage multiple AI agents but faced issues like developer interface preferences, port conflicts, and resource contention.
- After exploring external agent orchestration platforms like Ramp's Inspect and Stripe's Minions, Monaco found gaps such as high costs, lack of a populated data plane, and challenges in consolidating local service setups.
- Monaco adopted Coder to create 'Monacoder,' an in-house solution featuring cloud-hosted workspaces with 1:1 app-to-VM cardinality, baked-in dependencies via AMIs, and secure Claude access through Secrets Manager.
- Key technical decisions included using Docker Compose for runtime, seeding the data plane with de-identified sales data, managing databases with Postgres Kubernetes Operator, and integrating MCPs like Linear and Datadog.
- Monacoder has been widely adopted, with many developers using 3-6 workspaces simultaneously, and it supports workflows like automated PR review and CI failure addressing, accelerating code quality and security improvements.
- The system has proven low-maintenance with minimal costs, and extensions have been added for Linear delegation and GitHub webhooks, with plans to further integrate with Slack and other developer tools.
- Despite competition from tools like Cursor and Niteshift, Monaco sees value in building in-house, vendor-agnostic solutions to keep pace with evolving AI technologies, and is hiring for developer experience roles.