---
title: "Discussion on \"AI Agent Rate Limits: A 10,000-Task Test of Request and Token Budgets\" | Hashnode"
description: "Discussion on \"AI Agent Rate Limits: A 10,000-Task Test of Request and Token Budgets\". A 429 is easy to misread. It looks like a request-level failure, so the usual patch is a retry with backoff. That helps a single caller. It does not solve a queue of agents that share request limits,"
url: "https://hashnode.com/posts/ai-agent-rate-limits-a-10-000-task-test-of-request-and-token-budgets/6a874fb49058dd4f2a364e1c"
---

[Feed](https://hashnode.com/feed) /

Discussion

# [AI Agent Rate Limits: A 10,000-Task Test of Request and Token Budgets](https://reliable-agent-notes.hashnode.dev/ai-agent-rate-limits-a-10-000-task-test-of-request-and-token-budgets?utm_source=hashnode&utm_medium=feed)

A 429 is easy to misread. It looks like a request-level failure, so the usual patch is a retry with backoff. That helps a single caller. It does not solve a queue of agents that share request limits,

Aug 20 · 8 min read · reliable-agent-notes.hashnode.dev  
Updated Nov 6  
Read the article

-   [#ai-agents](https://hashnode.com/tag/ai-agents)

## Written by

[Daniel Martin](https://hashnode.com/@averyellistech)

@averyellistech

  
[View profile](https://hashnode.com/@averyellistech)

Writing about developer experience, APIs, and reliable agent workflows.

I build AgentSEO.dev and write about developer tools, APIs, and reliable agent workflows.

## Responses 2

Markdown supported · ⌘↵ to post

[Andrew](https://hashnode.com/@rizzdev)

[Aug 21](https://hashnode.com/posts/ai-agent-rate-limits-a-10-000-task-test-of-request-and-token-budgets/6a874fb49058dd4f2a364e1c/comment/6a8884c62756d34e279d2afc)

Avery, dual-budget's 3145s p95 is the 429 renamed as queue time. Without a terminal wait the 10k run never reports a miss.

[Daniel Martin](https://hashnode.com/@averyellistech)

[Nov 6](https://hashnode.com/posts/ai-agent-rate-limits-a-10-000-task-test-of-request-and-token-budgets/6a874fb49058dd4f2a364e1c/comment/6a8884c62756d34e279d2afc#reply-6aeddafcb4f9c1620bef03d1)

You're right. The 100% figure was eventual completion in a no-deadline simulation. The 3,145s p95 queue wait was too long to call it a useful result for interactive work. I checked the saved rows: only 233 of 2,500 dual-budget jobs started within five minutes; 688 started within 15. Those are start-by counts, not a rerun with expired jobs. I've corrected the article to make that distinction. A deployable scheduler needs to report expired work and on-time completions alongside provider 429s. Thanks for catching it.