GitHub Blog Explains New AI Development Vocabulary: Loops, Harnesses, Squads, and More

GitHub’s latest blog post breaks down a growing set of AI‑centric terms that developers are adopting. Loop engineering replaces one‑shot prompts with repeatable, scheduled agent workflows, while Ralph loops are a brute‑force version that can be costly. Squads and fleets describe how multiple agents with distinct roles or parallel tasks collaborate, mirroring real‑world teams. Harnesses refer to the surrounding infrastructure—tools, permissions, memory—that directs a model’s behavior, exemplified by GitHub Copilot. Hill climbing is the iterative process of refining agents and harnesses through evaluation and feedback. The post also clarifies the difference between closed, open‑weight, and open‑source models, and notes the rise of forward‑deployed engineers who embed AI into customer environments. Developers are encouraged to design repeatable loops, validate outputs, and continuously improve harnesses to build reliable AI workflows.