ConeheadAI · leftover lesson

SNK3-080 · Curriculum How-to

ReAct loops vs marketing 'loops'

L2–L5 · Capable–Builder (bootstrap — system rebalances) · ~3m

generalagentsreactloopsAUT · J8 · S2C3.2C5.2

Look for this insight

When someone says 'agent loops,' check whether they mean the plain ReAct cycle (act → observe → act) or a product buzzword.

Lesson hook

Simon Willison points out that the basic agent loop from the 2022 ReAct paper may be exactly what people mean when they say 'loops' again. That's genuinely useful — you can test a loop without buying a whole platform. Act, observe a checkable result, then act again. No observation step? Then it's not really a loop — it's just a longer prompt.

Do this now

Open any AI chat tool and run a mini ReAct cycle: (1) goal in 8 words (2) one action (3) observe: did a checkable output appear? (4) a second action based on that observation. You're done when you've written the observation line in 12 words or less.

Completion check: writes_observation_line

Open snack link →

Staff review Sign in

Related by competency

All snacks · Take Question 1 · Team training