# My AI Software Factory Ships Hundreds of PRs a Week (Here's How I Manage It)

## Executive summary

The video details the speaker's 'AI software factory,' an autonomous system designed to take high-level requirements (like a PRD) and build, verify, and validate software in an automated loop. To manage the complexity and maintain reliability, the speaker built a 'Mission Control dashboard.' This dashboard provides centralized observability, tracking key decisions, managing dependency mapping, and providing full audit trails using tools like Archon and Neon Postgres.

## Key takeaways

- The Necessity of a Control Plane: While the goal is a fully autonomous 'dark factory' (no human in the loop), the speaker found that adding a human-in-the-loop control plane is necessary for reliability. This dashboard manages the complexity by surfacing key decisions that require human attention, ensuring the system remains manageable and auditable.
- Centralized Observability and Auditing: The dashboard centralizes all factory activity, including individual runs, key decisions, and version comparisons (diffs). All agent actions, tool calls, and decisions are stored in a Postgres database, providing a full audit trail essential for debugging and system evolution.
- Leveraging Neon for Backend Infrastructure: The entire system is backed by Neon Postgres, which is used for storing all Archon run data and GitHub webhook triggers. The use of Neon's functions and real-time capabilities allows the dashboard to display live data updates (e.g., new PRs or issues) as they occur in the factory.

## Technical details

- AI Software Factory Workflow: The factory accepts a PRD (Product Requirements Document) and autonomously splits it into issues, running each through verification steps and code reviews before merging to production. The process is managed by the open-source harness builder, Archon.
- Archon Harness Builder: Archon is the open-source backbone that converts any AI coding process into a reusable workflow. It manages complex agentic engineering tasks, such as running multiple steps in parallel and providing full observability down to the agent's tool calls and decisions.
- Data Management and Realtime Updates: The system uses Neon Postgres to store all data. GitHub webhooks trigger Neon functions upon new issues or PRs, creating a single entry point for all factory activity. Neon's Realtime feature ensures the dashboard displays live data as the underlying Archon workflows execute.
- Dependency Mapping and Scaling: The dashboard tracks key metrics like the number of runs, merges, and the status of ongoing PRs/issues. It manages dependency mapping and allows for running validation and regression testing scenarios for every feature built.

## Practical implications

- Implement a dedicated observability layer (Mission Control) to manage the complexity of highly autonomous, multi-step CI/CD pipelines.
- Use a robust backend (like Neon Postgres) to centralize all agentic workflow data, ensuring a single source of truth for audit trails and decision logging.
- Adopt a modular workflow harness (like Archon) to manage parallel execution and abstract the core logic of the factory, making the system reusable and observable.
- Integrate real-time data streaming (e.g., Neon Realtime) to provide immediate visibility into the status of long-running, complex build processes.

## Topics

AI Engineering, Build Automation, DevOps, Agentic Systems, Observability, Database Architecture, AI Software Factory, Mission Control dashboard, Archon, Neon

Source: https://www.youtube.com/watch?v=YHKuUjicB3M