Intel / video

Recursive Coding Agents - Raymond Weitekamp, OpenProse

Source: youtube.com
Author: AI Engineer
Date: Published Jun 25, 2026

Context

A recursive coding can work with material stored outside its immediate prompt and decide how to divide the investigation. This talk's proposed design applies recursive language-model ideas to coding: keep large inputs in files or variables, let the model inspect them through executable code, and let it call other model instances on selected subproblems.

The returned results feed back into the parent task. The distinguishing feature is that the model chooses how to break down the problem as it investigates, rather than following only a predetermined sequence of parallel calls. This is a way to organize access to large inputs and distribute work; the talk's broad performance comparisons do not establish that it reliably improves every coding task.

Terms in this piece · Glossary

AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
test-time compute — Spending more computation when the model answers — thinking longer, trying multiple attempts — to buy accuracy without training a bigger model.

Key quotes

“the bottleneck to this is not intelligence. The models are intelligent enough.”
Raymond Weitekamp

“The next day, and I swear this actually happened, cloud code empties the entire contents of my Solana wallet.”
Raymond Weitekamp

“The thesis is today's agents are mismanaged geniuses.”
Raymond Weitekamp

“I don't care whether it's latent space or reasoning tokens or code execution. I want results.”
Raymond Weitekamp

“in an RLM, the context itself is the object of computation”
Raymond Weitekamp

Read the source www.youtube.com