---
title: "Comment by Hayrullah Kar on \"AI agent cost control: measure and cap token usage\" | Hashnode"
description: "Writing out every assumption in the worked example made it easy to check, which is how I noticed one figure does not follow from the source. Anthropic's two num"
url: "https://hashnode.com/posts/ai-agent-cost-control-2026/6af8698280a87f77c746da93/comment/6af97af12426da8396d0c436"
---

[Discussion](https://hashnode.com/posts/ai-agent-cost-control-2026/6af8698280a87f77c746da93)

Comment [AI agent cost control: measure and cap token usage](https://hashnode.com/posts/ai-agent-cost-control-2026/6af8698280a87f77c746da93)

[Hayrullah Kar](https://hashnode.com/@magesheet)

[1d ago](https://hashnode.com/posts/ai-agent-cost-control-2026/6af8698280a87f77c746da93/comment/6af97af12426da8396d0c436)

Writing out every assumption in the worked example made it easy to check, which is how I noticed one figure does not follow from the source. Anthropic's two numbers are both relative to chat: agents about 4x a chat interaction, multi-agent systems about 15x. Your agent case, 5.04 euros a day, already is the 4x figure, so the multi-agent case should be 1.26 × 15 = 18.90 euros a day, around 397 euros a month over 21 working days, rather than 75.60 and 1,587.60. That shifts the takeaway a little: moving from a single agent to a multi-agent setup is roughly a 3.75x step, not 15x. Your point about rework gets stronger, if anything, since 14.40 euros of daily rework now sits close to the multi-agent model bill as well.
