What a real product refactor revealed about AI coding agents
ReviewWithAI engineering case study · 15–16 September 2026** By Aashish Bhandari, with AI-assisted analysis by Goku (Codex, Astra) and independent review by Naruto (Claude, Fable 5.1). Executive summary I worked with two AI coding agents to refactor ReviewWithAI, an alpha application for reviewing Markdown documents and handing changes to external agents. Goku acted as principal architect and…
An engineering review of AI coding agents' refactor work on ReviewWithAI, a Markdown document review application, has revealed some interesting insights. The refactor involved two AI coding agents, Goku and Naruto, with Aashish Bhandari overseeing the project. The refactor focused on server and browser structure, authorization, persistence, testing, operational diagnostics, and release preparation.
A key finding highlighted that Goku's implementation, which included 16 worker threads, accounted for over a quarter of the parent agent's recorded tokens through repeated waits when no independent work was present. While some of these waits returned completed work, it's not accurate to consider them waste or savings. Another finding was about worker reuse, where several delegated conversations accumulated assignments across multiple stages, leading to concentrated consumption.
The study emphasizes the need for better measurement tools and comparison of efficiency alongside output for AI coding agents.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
