Developer builds 'agentic factory' to deploy full-stack apps from one prompt

Developer Jake Saunders has built a self-hosted, sandboxed system capable of automating the entire software development lifecycle from a single user prompt. This 'software factory' independently creates a repository, writes application code and tests, passes continuous integration (CI), provisions a Postgres database, and deploys the final application behind HTTPS.

According to Saunders, the system executes these steps without further human intervention. "From one prompt it created a repo, wrote the application and tests, got CI green, provisioned Postgres and deployed the finished app behind HTTPS without another message from me," Saunders stated in a blog post detailing the project.

The Double Diamond Pipeline

To manage the complexity of autonomous development, the system implements an 'agentic double diamond' workflow. This structure splits the process into two primary stages: Diamond 1, which focuses on the Problem Space, and Diamond 2, which handles the Solution Space.

Execution is managed through a series of gated phases to ensure quality and alignment. The pipeline moves through /discover, /challenge, /spec, /plan, /challenge-plan, /build, and /review. The process concludes with a /ship command, which remains human-gated to ensure a developer provides final approval before the code goes live.

Adversarial Personas and Boundaries

To prevent the LLM from hallucinating or bypassing quality checks, the architecture employs specialized personas: @Analyst, @Lead, @Dev, and @Critic. Each persona operates under strict boundaries enforced by tool allowlists.

For example, the @Critic persona is designed for adversarial review and is explicitly restricted from editing or writing code. This separation of concerns ensures that the agent responsible for auditing the work cannot unilaterally change the implementation to hide errors, forcing the @Dev persona to address the critique.

The Shift Toward Disposable Infrastructure

This project reflects a broader industry trend toward 'agentic' coding tools that move beyond simple autocomplete toward autonomous task execution. By combining these capabilities with sandboxed environments, the system automates the repetitive boilerplate of CRUD application development, such as CI/CD setup and basic testing.

This represents a move toward 'on-demand' software infrastructure where the cost of creating a specialized tool drops significantly. As agents become more reliable at handling the full stack from specification to deployment, the role of the human developer shifts from writing line-by-line code to defining constraints, reviewing specifications, and auditing the factory's output.

What's Next

While the system is described as '(almost) fully self-hosted,' it relies on external LLM APIs to power its reasoning. Future iterations of such factories may look toward fully local models to remove external dependencies. For now, the project serves as a blueprint for how structured agentic pipelines can reduce the friction between a conceptual idea and a deployed production asset.