The open loop audit prompt that separates real delegation from simulated work + 3 more prompts for overnight execution

Source: Nate’s Newsletter Date: 2026-03-28 URL: https://natesnewsletter.substack.com/p/90-of-what-you-build-on-your-ai-agent

Summary

Nate draws a line between work that leaves your desk (real delegation) and work that produces documents rearranging what was already known (simulated delegation). The open-loop audit prompt is a diagnostic tool for identifying which side a given agent workflow falls on; it’s paired with three additional prompts designed for overnight, unattended execution. The concrete test: does the agent’s output produce fulfilled commitments, pre-gathered decision information, or shipped work — or does it produce summaries that require a human to still do the actual task?

Implications

• The real-vs-simulated delegation distinction is a sharp criterion for evaluating agent workflow proposals: if the agent’s output is a report that a human then acts on, the loop hasn’t closed — the human is still doing the work.
• “Overnight execution” as a design constraint surfaces a useful set of requirements for agent infrastructure: persistence, structured output handoffs, failure recovery, and human-readable audit trails.
• Feeds the agent patterns thread — the audit prompt framing is a practitioner-level complement to academic agent capability benchmarks, grounding the question in what changes after the agent runs.