Loops

Loops turn one result into the next input; agent loop engineering makes that pattern bounded, checked, and worth repeating.

The Spine

1. Agentic Prompt Loops — design the trigger, context, tools, verifier, stop condition, and budget that prompt agents without hand-holding.
2. Mechanical Feedback Loops — turn work into setpoint, gauge, lever, state, proof, correction, and a sharper next question.
3. Loop Registry — keep many loops honest with cadence, owner, gauge, cost brake, and kill signal.
4. Delegation — set act-alone, ask-first, escalation, and refusal rights before an agent runs.
5. Agentic AI Systems — fit loops into agents with tools, memory, harnesses, and evals.

Zoom Out

• Up: Agent Knowledge Contract — where loop knowledge sits in the broader agent operating contract.
• Next: Control System — how values and beliefs become gauges, levers, and correction.

Questions

Which loop are you feeding now?

• What result keeps returning as input?
• What proof would show the next pass improved?
• What stop condition prevents faster drift?