Claude SDK: 24-Hour Coding Agent | Cole Medin

Description

Cole Medin walks us through Anthropic's harness for long-running agents and shows what happens when you let Claude code for 24 hours straight. We built on top of the official Claude Agent SDK quickstart and added Linear for task management and observability. The core problem this solves: agents lose all memory between context windows. The harness uses two specialized agents - an initializer that sets up the environment and creates tasks, and a coding agent that makes incremental progress while leaving clean artifacts for the next session. Cole showcases a Claude.ai clone that was built entirely by this system running autonomously. Timestamps: 00:00:00 Introduction to the live stream and guest Ray. 00:00:59 Explaining the main topic: Anthropic's harness for long-running agents. 00:03:32 Ray introduces himself and his AI coding background. 00:08:13 Explaining the core problem: agents and long context windows. 00:10:30 Why Cole integrated Linear for task management and observability. 00:13:42 Discussing Archon, Cole's open-source AI command center. 00:21:55 Architectural deep dive: How the agent harness works. 00:25:50 Showing the detailed `app_spec.md` file for building a Claude.ai clone. 00:30:52 Pro-tip: How to use a Claude Max subscription instead of API credits. 00:39:20 Showcase: A Claude.ai clone built by the harness in 24 hours. 00:46:57 Detailing the prompt changes for the Linear integration. 00:49:02 Ray's 'AGI' moment seeing the fully integrated system. 00:50:58 Explaining the Test-Driven Development (TDD) approach used. 01:04:18 Kicking off the live demo. 01:09:21 Live demo: The initializer agent creates tasks in Linear. 01:14:35 Live demo: Agent uses Puppeteer to validate the app. 01:22:30 How to adapt the harness for other models (Codex, OpenCode). 01:27:36 Live demo: The agent builds the initial UI layout. 01:34:50 A fun detour into cooking and favorite pizza. 01:41:09 Will AI stop the creation of new programming languages? 01:50:08 The importance of automated testing in AI workflows. 01:56:48 Ray provides a complete summary of the harness architecture. 01:59:04 Cole discusses his Dynamous community and coding course. 02:04:50 Final thoughts and wrap-up. Resources: Anthropic Engineering Article: https://www.anthropic.com/engineering/effective-harnesses-for-long-running-agents Claude Agent SDK Quickstart: https://github.com/anthropics/claude-quickstarts/tree/main/autonomous-coding Cole's Channel: https://www.youtube.com/@ColeMedin Dynamis Community: [Cole's link] The harness and principles here are tool-agnostic. You could swap Claude for Codex, Linear for JIRA, and the architecture still applies.