How to use Codex to compare chess coach apps
An Orca experiment compared Codex and Pi teams building chess coaches. Codex finished faster, while Pi’s engine won 18 of 20 games and drew two.
news.ycombinator.com
Prompt to try
Editorial adaptation, not the source author’s exact prompt. Replace bracketed inputs and review tool permissions first.
Use Orca to build [chess coach requirements] with specification, design, and build roles plus a shared reviewer. Run the same task with [agent/model A] and [agent/model B]. Preserve prompts, tests, and logs, then compare completion time and a [match format] engine match.
Reach people exploring AI agents
See placements →What happened
Both teams built chess coach apps. Codex finished faster, while Pi’s engine won 18 games and drew two in the 20-game match.
Reported by piscaries on Hacker NewsHow the agent did it
- Ran a specification, design, and build team on Codex in Orca.
- Repeated the same team structure with Pi in Orca.
- Compared the resulting chess coaches in a 20-game match.
Where a person stepped in
The author ran the comparison and match; a shared reviewer was part of each team. Human code-review details are not supplied.
A person approved each step before the agent continued.
An author used Orca to compare two coding-agent setups on the same kind of application. The experiment had two stated goals: trying Orca as a way to run a team of coding agents and comparing the minimal open-source Pi agent with Codex.
For each run, the author used the same small team structure: specification, design, and build roles, with a shared reviewer. One run used Codex with GPT-5.6. The other used Pi with GLM-5.3. Both teams produced chess coach applications, giving the author two implementations to compare side by side.
The author then let the chess coaches compete. Codex completed its development run faster, but Pi's engine won 18 games and drew the remaining two in a 20-game match. The post presents completion speed and match performance as separate observations. It does not report a numerical development time for either setup.
The linked repository contains prompts, both applications, tests, judging material, and match logs, according to the author. Those materials are described in the supplied post; their contents are not reproduced there.
Who it's for: Developers comparing coding-agent teams who want both a completed application and a task-specific way to compare the resulting behavior.
Prompt
Suggested prompt (written by Agents In Use, not the author's)
Source: Hacker News · piscaries
Frequently asked questions
Which agent teams were compared in Orca?
The author ran specification, design, and build roles with a shared reviewer twice: Codex with GPT-5.6, then Pi with GLM-5.3.
Which chess coach performed better in the match?
Pi’s engine won 18 games and drew two in a 20-game match. Codex finished development faster; the post does not give exact completion times.
More details
About this account
Teams of agents handled specification, design, and application building, with a shared reviewer in the team structure.
Platform: Hacker News. Attribution: https://news.ycombinator.com/item?id=49979728. Prompt: Suggested (written by Agents In Use, not the author's). Sourced by the Agents In Use pipeline.
Credit: piscaries on Hacker News
Source published 6 Oct 2026
Source checked 14 Oct 2026
Tools and topics
Orca · Codex · GPT-5.6 · Pi · GLM-5.3
Agent comparison · Chess · Orca
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.