The core idea in 30 seconds
A loop = 3 things
Trigger, action, and stop condition. Underneath every complicated design, it’s always reason → act → observe, repeating until "done".
Verification is everything
A loop is only as good as the way it checks "done." Without verification, it confidently delivers the wrong work — and calls it done.
You become the architect
Instead of typing the next step, you design the system that decides the next step. That’s the “v2” leap: Loop Engineering.
The 4 paths
Fundamentals: what an agent loop is
The reason→act→observe skeleton, the 6 invariant elements, and the map of the 11 schools that confuse everyone. Each term explained as it comes up.
When (and when NOT) to use a loop
The decision ladder, the 3 setups (Solo, Maker→Checker, Manager→Helpers), and the 4 ways to check whether something is "done." Most tasks don't even need a loop.
Building your first loop
The 7-step sequence, ready-to-copy loops you can run today, the 8 ingredients of a solid loop, and the 8 mistakes that trip up beginners.
Loop Engineering (the "v2" level)
“Stop prompting, design loops”: Ralph loop, scheduled loops, externalized state, adversarial verification—and the real risks of cost and cognitive surrender.
How this course was made (and why it matters)
All the content originated from a verification loop: agents scanned 45 sources (Anthropic/OpenAI/Google docs, the ReAct paper, practitioner blogs, Nate Herk's video), a "maker" agent copied the exact phrase from each source, and a separate "checker" agent searched each URL again to confirm. It's the same maker→checker pattern you'll learn in Track 2 — the course itself is the output of a loop.
Sources independently confirmed · no hallucination presented as fact.