The factory analyzes failed agent runs in aggregate. An "observer agent" identifies common failure modes and then automatically generates code changes to the factory's underlying definitions and prompts, creating a continuous self-improvement loop that prevents future errors.

OpenAI structures its repositories to be a complete, self-contained knowledge base for AI agents. All project artifacts—design docs, historical implementation plans, and even text versions of external library documentation—are checked in, allowing the agent to find any needed context via simple search.

Instead of codebases becoming harder to manage over time, use an AI agent to create a "compounding engineering" system. Codify learnings from each feature build—successful plans, bug fixes, tests—back into the agent's prompts and tools, making future development faster and easier.

Move beyond manual agent improvement by creating an automated loop. In this process, an agent runs, its performance is evaluated, failures are identified, and another process suggests and implements code fixes. This creates a foundation for self-improving systems.

Inspired by fully automated manufacturing, this approach mandates that no human ever writes or reviews code. AI agents handle the entire development lifecycle from spec to deployment, driven by the declining cost of tokens and increasingly capable models.

Replit uses an internal agent that analyzes user interaction traces, identifies errors, generates prompt changes to fix them, submits them as pull requests, and initiates A/B tests. This creates an autonomous, self-improving loop for the platform's AI capabilities.

A more effective way to increase developer velocity with AI is to have champion engineers embed knowledge directly into the systems. This includes creating context engineering techniques, `agents.md` files, and agent skills within the repo itself. This way, any agent pointed at the repo benefits, rather than relying on every individual developer's expertise.

The current model of a developer using an AI assistant is like a craftsman with a power tool. The next evolution is "factory farming" code, where orchestrated multi-agent systems manage the entire development lifecycle—planning, implementation, review, and testing—moving it from a craft to an industrial process.

Instead of a standard package install, providing a manual installation from a Git repository allows an AI agent to access and modify its own source code. This unique setup empowers the agent to reconfigure its functionality, restart, and gain new capabilities dynamically.