The author critiques coding agents as the bottleneck in AI-assisted development, arguing that while LLMs have improved dramatically, agent frameworks like Claude Code and Codex remain primitive in task management and error handling. After initial excitement in early 2025, practical usage revealed persistent issues: agents stall, mismanage workflows, and often declare tasks complete prematurely. The core argument is that the model (the 'brain') is doing the heavy lifting while the agent (the 'body') fails to effectively connect outputs to codebases.
Background
Coding agents like Claude Code, OpenAI Codex, and Devin have emerged as popular tools for AI-driven software development in 2024-2025, generating significant hype around autonomous coding capabilities.
- Source
- Lobsters
- Published
- Oct 9, 2026 at 10:22 PM
- Score
- 6.0 / 10