A senior developer and author introduces 'AI-driven development' (AIDD), a structured approach to agentic engineering built around orchestrating multiple LLMs for real software projects. The piece documents a real experiment: building Octobatch, a 21,000-line Python batch orchestration system for LLM pipelines, where AI wrote all the code. Key lessons include: experienced developers get more from AI tools because they can evaluate output quality; LLMs overestimate complexity and are biased toward adding code rather than deleting it; architecture often emerges from failure rather than upfront design; and your AI chat history is a valuable project dataset. The author also highlights LLM batch APIs (from OpenAI, Anthropic, and Google) as an underappreciated shift that enables treating LLMs as processing infrastructure rather than conversational tools.
Table of contents
The experimentAgentic engineering and AI-driven developmentThe orchestration mindsetThe path to batchWhy orchestration?Lessons from the experimentWhat’s nextSort: