FOUR GATES SEPARATE A LOOP FROM A PROMPT: MAX ITERATIONS, BUDGET, NO-PROGRESS, COMPLETION CHECK. SKIP ANY ONE AND YOUR AGENT LOOPS FOREVER OR SHIPS WRONG.
i had a coding agent burn 40 dollars in tokens last month, apologizing to itself in a retry storm, until budget tripped at
i started with one grok bot chief on monday. by friday it had signed a rulebook, drafted five specialist operators, and returned the first +12 sol paper night.
no one had told it who to hire. openai and anthropic each sell you one $200 chat that sits still and waits for your
FIVE LAYERS OF AGENT ENGINEERING, EACH ONE WRAPS THE ONE BELOW IT. IF YOU SKIP LAYER 2, YOUR LAYER 5 WILL LOOK BROKEN WHEN IT IS ACTUALLY JUST STANDING ON NOTHING.
for weeks i debated harness vs loop vs graph like they were competing choices.
then a stack diagram made the shape
A GOOD GROK BOT SYSTEM DOES NOT AVOID FAILURES. IT CATCHES EVERY FAILURE, TURNS IT INTO A HOUSE RULE, AND MAKES SURE THAT ONE NEVER FIRES AGAIN.
six weeks ago i had to tell Chief the same correction three days in a row.
by day 4 i wrote it into house_rules as one line. i have
