In Part 1 of the article series, the author conducted an experiment to address the challenges of managing an automated workforce. The experiment involved an AI agent, referred to as smol dev, tasked with generating a naming scheme for Docker containers. The agent was given prompts and specifications to follow. In Part 2, the author explores two different paths for further development: finalizing the project themselves or allowing smol dev to iterate on the results. They review and correct the code, including fixing import issues and unifying the dictionary structure. Testing is also discussed, with technical and design issues identified and addressed. Finally, the author discusses the agent’s self-correction and the need for explicit design specifications.

source update: Lessons Learned from SMOL AI — Part 2 – Towards AI