Move the execution boundary
Run the long-lived agent outside the main application server, keeping application-server requests short.
04 / OPERATIONAL ARCHITECTURE
Giving long-running agents their own runtime
I moved long-running application-building agent execution into an isolated VM runtime to release the main application server from holding the entire execution.
The problem
An agent can spend a long time planning, calling tools, waiting for model output, and continuing after user input. Keeping that work on an application-server thread ties product-serving capacity to agent duration.
What I owned
I worked on the transition from JSON-based tool operations to file-based application authoring, and moved the agent’s execution into its own runtime.
Engineering decisions
Keep authoring artifacts explicit
The application-building agent writes the low-code files that represent application components.
Preserve continuity
Execution state and live browser updates keep the interface connected to work happening in a separate process.
The result
The architecture isolates agent workloads and frees application-server threads while preserving the interactive build experience.