Codex Camp is set for December 12, where Every subscribers can learn about building with OpenAI’s coding agent. The article explores the transition in software engineering where coding is now predominantly handled by AI agents. This shift has made traditional coding practices feel outdated, as the process of manually writing code and tests becomes increasingly unnecessary. At Every, they’ve developed a new approach called compound engineering, which transforms how features are developed. Instead of each new feature complicating future development, compound engineering allows each addition to simplify the next by creating a learning loop for AI agents and engineers. In this model, planning takes precedence; developers spend most of their time researching and crafting detailed implementation plans. Agents analyze the existing code and gather best practices before writing code based on the plan. The actual coding phase is straightforward, as agents convert plans into actionable tasks. They also utilize context protocols to simulate user interactions during development, which enhances the quality of the output. After coding, the assessment phase involves reviewing the agent’s work using various tools, from traditional linters to advanced AI review agents. This multi-faceted review process helps catch errors and improve code quality. The final step, compounding, is where previous learnings are documented for future use, ensuring that solutions to past problems are integrated into the system. For instance, when developing features for Every's AI email assistant, the agent must consider the existing architecture and where new functionalities fit within it. This systematic approach not only accelerates the development process but also enhances the overall efficiency and effectiveness of software engineering at Every, demonstrating a significant shift in how software can be built and maintained.