The Five Questions That Turn a Messy Task Into an AI Loop (+ the prompts to map yours)

Source: Nate’s Newsletter Date: 2026-06-24 URL: https://natesnewsletter.substack.com/p/ai-loop-managers

Summary

Nate’s Newsletter piece arguing that agents should manage the recurring coordination work that lives between applications — the invisible labour of remembering, checking, following up, and connecting information across email, calendar, Slack, and other tools. The proposed unit is a “loop”: a narrow, recurring workflow with defined memory, information sources, safe action boundaries, and cross-awareness (one loop triggering relevant work in another). The full five-question framework for scoping a loop is paywalled; the preview surfaces questions about existing hidden loops, information dependencies, and safe action boundaries.

Implications

• Agent scope theory. The “loops of loops” framing is a usable mental model for scoping agent work: not autonomous life-management, but narrow recurring workflows with explicit boundaries. This is the practitioner-friendly articulation of the “corrigible agent” design pattern.
• Inter-app coordination as the frontier. The piece correctly identifies that the hard problem isn’t within-app automation (already saturated with integrations) but the glue layer between apps. This is where MCP’s multi-server composition model is supposed to win — worth tracking whether practitioners adopt MCP for this or build bespoke connectors.
• Content category. High-quality practitioner meta-content; the signal value is the framing, not any product or model news. Useful for calibrating where the practitioner discourse is relative to the infrastructure being built.