A detailed case study on building a greenfield SaaS product (TextForge, an AI email automation tool) using LLM-driven development practices the author calls 'Software 2.0.' Key techniques covered include: using Claude Code's planning mode to define requirements and architecture before writing code, providing reference architectures (DrawTogether.NET) to guide LLM output style, creating reusable agent skill files for non-obvious patterns, writing PRDs for feature scoping, and building a verification pipeline with static analysis, automated tests, snapshot testing via Verify, and Aspire MCP-powered browser automation. The post also discusses incremental refactoring to prevent code quality decay and the tradeoffs of LLM-authored tests. The project passed a CASA2 security audit despite being nearly 100% LLM-written code.
Table of contents
Planning mode, reference architectures, agent skills, and verification pipelines — the core techniques for building greenfield projects with LLMs.The Project: TextForgePlanning and Designing the ProjectRequirements and RefinementsImplementationSort: