Developer Showcases True Recursion in RLMs via REPL Subagents

willccbb · x · 2026-08-07

A developer pointed out that RLMs (Reinforcement Learning Models) can achieve true recursion even at depth=1.

Because the model operates within a REPL (Read-Eval-Print Loop), it can write real recursive programs that call subagents directly, eliminating the need for nested "subsubagents." The author noted that Claude Code currently cannot do this natively; it requires writing a program to a file and executing the claude -p command as a workaround.

Original post →

More from coding & agent

coding & agent channel →