Overview
- This month leading AI figures and companies have urged designers to stop hand‑writing prompts and instead craft persistent 'loops' that spawn, monitor, and re‑prompt agents until goals are met.
- A loop is a recurring workflow that keeps agents working toward a goal without a human typing each instruction; common building blocks include automations, worktrees, skills, plugins or connectors, sub‑agents, and external memory.
- Practical examples already in use include /goal commands that tell coding agents to keep working and OpenClaw/Codex loops that wake periodically to maintain repositories and assign tasks into threads.
- Running fleets of agents increases token and compute use, so developers advise scheduling, task‑level model routing, budget caps, and separate verifier models to prevent runaway costs and biased self‑reviews.
- The shift recasts developers as job designers and managers who compose automations and verification pipelines, which is creating demand for new tooling for scheduling, monitoring, testing, and governance.