How to Build a Coding Agent (github.com)

A new workshop has been announced that guides participants in building a coding agent using plugins for the pi and Amp systems. The workshop material includes detailed documentation with ASCII diagrams and runnable demos, focusing on how context engineering has evolved, shifting control from users to the model itself. Attendees will learn about the agent's core loop, harness engineering, and the concept of "compaction," which allows for lossy data recovery — a significant technical advancement attributed to improved model capabilities. This workshop is especially noteworthy for the AI/ML community as it highlights a growing trend towards integrating user inputs more effectively within models, thereby enhancing the efficiency of coding agents. Key details include various stages of compaction, including file-based methods and the introduction of Recursive Language Models (RLM). Overall, this initiative reflects the advancements in AI model performance and the need for practitioners to adapt to new methodologies in agent design, which could reshape how coding tasks are executed in the future.