Loop Engineering
Playbook
Build AI loops you can walk away from because the check is real, not because the agent said it finished.
Get the playbook without leaving the flow
Full guide, verifier ladder, real eval examples, and the 7-day build track.
What You Get
The Verifier Ladder
From prompt to closed loop
Who This Is For
The Parts Most People Skip
Loops are easy. These decisions decide whether they are safe.
Closed before open
Start with loops that have clear steps, a limit, and a pass/fail check. Open exploration comes later.
Cost caps first
Every loop needs a stop condition before it starts: attempts, time, spend, or human review.
Understanding debt
If output grows faster than your grasp of it, your system is productive and dangerous.
Built From Real Loops
Content validator
A small script scans for phrases, punctuation tells, and patterns that make writing sound machine-made.
Scorer plus threshold
A score matters only when the pass line is fixed before the agent starts working.
Binary proof contract
yalla asks a reviewer one yes/no question per check. One fail blocks shipping.
Human approval gate
Outreach and judgment-heavy work can be drafted by AI, but the final decision stays with you.
Iwo Szapar
- *Co-created AI Maturity Index with Harvard
- *Trained 25,000+ professionals on AI
- *Built yalla, an open-source closed-loop coding pipeline
Get the Playbook
Free guide for building loops that can prove they worked.
Loop Engineering Fits Inside a Second Brain
The playbook teaches the loop. Second Brain 2.0 gives that loop memory, agents, tools, and a place to store the lessons each run produces.
The complete Claude Code system. 30+ agents, memory that compounds, pre-configured MCPs.
DIY $197 / Kickstart $597 / DWY $2,497
See PackagesA companion skill for Plan, Work, Review, Triage, and Learn. Use it with this playbook.
Get the Skill