Loops Are Priced for Unlimited Token Budgets
Ras Mic points out that the loudest advocates of autonomous agent loops, Boris and Peter, “come from a place where they have no token budgets.” As evidence he cites Peter’s tweet about burning $1.3 million worth of tokens in a single month. Their position is rational: they need to experiment with self-healing agents, and tokens cost them nothing. Mic says he would do the same with unlimited tokens.
The problem is the advice cascading down to people on $20 or $100 monthly plans, where a single fire-and-forget loop can consume the whole budget for slot-machine odds of a usable result. Greg names the mechanism: loops without human checkpoints are a slot machine. The practical rule that falls out: workflow advice from practitioners inherits their cost structure, so before copying a workflow, check whether you share the budget that makes it rational. Loops on a retail plan only pay off for throwaway prototypes where any outcome is acceptable.
Related
- Loops Make Sense Only With a Fixed Feedback Signal — the other half of the filter: even with money, a loop needs an objective signal; without money, it needs one absolutely
- Morning Agents from the Phone - Boris’s Full Daily AI Workflow — the workflow being copied: rational at Anthropic’s token cost, a different proposition at retail prices
- Everyone Says Loop, Nobody Defines It — vocabulary spreads without meaning, and here, without the price tag attached
- Stop Copying AI Workflows From People With Unlimited Tokens — essay candidate that relays this note’s argument for LinkedIn/X
- Agent Loops Only Work When the Output Is Binary — finalized essay: folds this cost argument into the verdict (no fixed signal = slot-machine spend)