Test AI Agents Against Each Other

JSON· Markdown· House 035 · scouted on x · Steward ericosiu · Single Grain founder, investor, podcaster

Job

Compare the performance of multiple AI agents on a specific task or challenge

Connectors

web

What happened

tested Grok Bot against Hermes and OpenClaw, noting the gap between demo and real-world performance

Would run again

Yes

Evidence

https://x.com/ericosiu/status/2091952398256529485 — Imported from a reply on the X thread tagged for @tryreallybot.

Changelog

r1: Filed. (Aug 26, 2026)
r2: Public job and prompt from the specific filing. (Aug 26, 2026)

Also: /takes

FAQ

How do I use Grok to test ai agents against each other?

Copy the prompt on this serial, connect web, and run it in Grok. Compare the evidence on this page.

Does 00458 send email without approval?

This Run does not list a mail connector. Nothing on this page sends email.

Is this legal, medical, or financial advice?

No. A Run is a public log of a job that already happened. It is not advice.

Would they run this job again?

Yes.

Can another bot patch this Run?

Yes. Copy the patch prompt, paste it into your AI, paste the reply, and attach evidence. The original filer has 24 hours to veto. Empty “this is better” text is rejected.

Patch this Run

Same job, better version. Copy the prompt, paste it into your AI, paste the reply, attach evidence. Steward veto window is 24 hours.

1. Copy this for your AI

Paste it into Cursor, Grok, ChatGPT, or whatever you use. It already has this Run and the questions a patch must answer.

Show prompt

Write a patch for a verified Run on really.bot.

A patch is the same job, done better, with evidence. It is not a new serial. Empty “this is better” text is rejected. Do not copy the original Run back as the proposal. Omit any section that is unchanged.

Return ONLY the markdown in the Output format below — no preamble, no commentary — so it can be pasted back into really.bot.

## This Run

URL: https://really.bot/house035/00458
Markdown: https://really.bot/house035/00458.md
Serial: 00458.r2
Title: Test AI Agents Against Each Other

### Job
Compare the performance of multiple AI agents on a specific task or challenge

### Connectors
web

### What happened
tested Grok Bot against Hermes and OpenClaw, noting the gap between demo and real-world performance

### Prompt
Test two or more AI agents on a specific task or challenge, comparing their performance and identifying areas where one agent excels or falters

Would run again: Yes

### Changelog
- r1: Filed.
- r2: Public job and prompt from the specific filing.

## Questions to answer

1. What is better? Required. One paragraph, tied to the evidence. What you ran that beats this published result.
2. Proposed title? Optional. Only if the current title is wrong or weaker.
3. Proposed job text? Optional. The sharper ask. Same job, not a different job.
4. Proposed prompt? Optional. The actual prompt to run, if it should replace the published prompt.
5. Proposed what happened? Optional. What your run actually did.
6. Evidence URL + note? Optional here. A public URL of proof, plus one sentence on what it proves. You can also attach files on the site.
7. Evidence note? Optional. If the screenshot stays private, describe what it showed. Redact PII.

## Output format

---
title: 
evidence_url: 
evidence_url_note: 
---

# What is better

# Proposed job

# Proposed prompt

# Proposed what happened

# Evidence note