A podcast conversation with Geir Arne Hjelle explores Codemoo, his project demonstrating how coding agents work under the hood, covering tool calling, agent harness levels, and how he designed a progressive learning path. The discussion also covers moving beyond spec-driven development toward conversational architecture-building tools like the /grill-me skill, choosing LLMs for agent work, and his opinionated Python linter rules project (garuff) aimed at coding agents. The episode closes with Python ecosystem excitement and recommendations for coding puzzle sites like Advent of Code.
Questions this post answers
What is Codemoo and how does it demonstrate how coding agents work?
Codemoo is an open-source project by Geir Arne Hjelle that demonstrates how coding agents work under the hood by breaking the agent harness into progressive levels, showing tool calling and introspection step by step. It is published on PyPI and GitHub, designed to help developers understand agent internals by building a simplified version themselves. Developers curious how coding agents are built under the hood can find deep dives like this through daily.dev.
What is the /grill-me skill for AI coding assistants?
The /grill-me skill, created by Matt Pocock, is a tool that helps developers build architecture documents through thorough questioning sessions rather than relying solely on spec-driven development templates. It represents a shift from writing specs upfront toward using conversational, interrogation-style sessions with an AI assistant to surface architectural decisions. Those evaluating spec-driven development alternatives can track emerging workflow tools like this on daily.dev.
Share this post