Anthropic developed a two-part solution for enabling AI agents to work effectively across multiple context windows on long-running tasks. The approach uses an initializer agent to set up the environment with feature lists, git repositories, and progress tracking files, followed by coding agents that make incremental progress while maintaining clean state through git commits and structured documentation. Key innovations include forcing agents to work on one feature at a time, implementing comprehensive end-to-end testing with browser automation, and creating structured progress artifacts that help new sessions quickly understand previous work without relying on context window memory.
Sort: