Critic Finds Prime Agent Lacks True RLM Recursion
Code review of agent project prompts replies on recursion limits and RLM definitions.
A code analysis posted on X challenges hype around Prime Agent by showing RLM_MAX_DEPTH set to 1 and no recursive decomposition. Chenhao Tan comments that human verification may involve an AI assistant. Alex Zhang agrees Prime Agent is not an RLM and recalls that the RLM paper addresses abstractions for systems where language models call other language models through programmatic tools and sub-agents.
whenever something is crazy hype and is followed up by an armada of PR, you gotta bring out the bullshit detectors so i cloned the repo and did a little digging and found a lot of sound and fury. please enlighten me if any of these facts are untrue, would love to be educated…
Introducing Prime Agent: A self-improving RLM harness for coding and long-running autonomous tasks. Designed to be both token-efficient and expressive through programmatic tool calling, context as a variable, multi-agent messaging, and a self-modifiable harness state.
Reactions
2 postsCT 
Chenhao Tan@ChenhaoTanHuman verification at play 👀potentially with AI assistant?
- likes: 5
- replies: 1
- bookmarks: 1
- reposts: 0
AZ alex zhang@a1zhangThat answers it I guess Prime Agent isn’t an RLM thank you BrainsAndTennis Jokes aside, I have answered this question many times in the past. The purpose of the RLM paper is to explain: “what type of abstraction lends itself well to a system around LMs calling LMs calling…?”…
- likes: 287
- replies: 17
- bookmarks: 194
- reposts: 5