How I AI
What a harness is and how to build one with Claude Agent SDK
· Summarized
Key takeaways
- A harness is simply code that adds structure to an AI agent to improve its effectiveness for specific tasks.
- Harnesses are most useful for repetitive, structured workflows like bug triage, incident handling, or research, where consistency and outcomes matter.
- Building a harness involves defining workflows, choosing tools, encoding permissions, and structuring output artifacts for team use.
- Using models like GPT-5.5 and Claude can assist in generating harness code, but precise prompts and workflows are essential for reliable results.
- A UI—whether CLI, terminal, or web—can significantly enhance the usability and testing of a custom harness.
Summary
Understanding What a Harness Is
The episode clarifies that a harness is simply code that surrounds an AI agent to make it more effective for specific tasks. It can include context, actions, and a goal, and aims to streamline structured workflows. The host emphasizes that harnesses are not inherently complex; they are about adding structure around AI to tailor it for particular use cases, thereby improving efficiency, consistency, and outcomes.
When and Why to Build a Harness
Building a harness makes sense for repetitive tasks that require a predefined workflow, especially when precise outcomes are desired. Examples include bug triage, incident management, PR preparation, support escalations, and research. The host shares how focusing on specific, structured workflows allows the harness to outperform general-purpose AI tools by constraining actions and providing clear instructions, resulting in better reliability and results.
Design and Architecture of the Bug-Triage Harness
The host showcases a custom-built terminal UI for the bug management harness, which interacts with tools like the Claude Agent SDK, Sentry, Linear, GitHub, and Vercel. The harness gathers evidence, performs root cause analysis, and generates artifacts, structured for easy follow-up and team sharing. It runs specific workflows, encodes permissions, and uses multiple models (like Claude Sonnet 4.6 and GPT-5.5) to optimize different tasks. The architecture includes task management, integrations with external tools, artifact storage, and custom prompts to guide AI behavior.
Development Process and Practical Tips
The host describes how he used GPT-5.5 and Claude models to generate the harness code by prompting and iterating, noting resistance from the models for determinism. Emphasis is placed on defining workflows precisely, coding adapters for specific tools, and structuring artifacts for team use. The process involves testing against real data, creating opinionated integrations, and building a UI for easier interaction. He advises that building a harness involves identifying workflows, documenting steps, choosing appropriate tools, and encoding rules and permissions—making AI operations reproducible and controllable.
Main Takeaways and Future Ideas
The episode underscores that harnesses are not mysterious; they are structured code around AI tailored for particular jobs. The host envisions harnesses as a way to upgrade end-to-end workflows, perhaps even suggesting that frameworks like Rappers could themselves be considered harnesses. The key insight is that constraining and structuring AI interactions enables more effective, reliable, and repeatable outcomes. The host invites feedback and future topic suggestions, emphasizing the educational value of demystifying AI components.
Sponsors
- Bolt.new · Bolt is the AI at Builder for people who have ideas and want to ship them. It generates production-ready code from descriptions, allowing users to build startups, MVPs, landing pages, and internal tools in minutes. Founders, product managers, designers, and marketers are using Bolt to rapidly launch and test ideas, with real revenue and usability.
- Customer.io · Customer.io helps you build customer engagement campaigns with AI. Describe your campaign, and the AI creates the audience, messages, and timing. Review, tweak, and launch — saving hours and making your marketing more effective while tying everything back to results.
Show notes
Everybody is saying, “It’s not the model, it’s the harness,” but almost nobody stops to explain what a harness actually is. So I did. I built one live on the show: a Sentry bug-debugging harness for my company ChatPRD, using the Claude Agent SDK, a custom terminal UI built with the Ink library, and opinionated adapters for Sentry, Linear, GitHub, and Vercel. The harness handles evidence gathering, root-cause analysis, and follow-up artifact creation, all without me needing to type “dear agent, please fix this bug” ever again. I also walk through the architecture, share the code structure, and give you the exact process I used so you can build your own harness for any repetitive, structured workflow in your business.
What you’ll learn:
- What a harness actually is
- When to build a harness versus when to stick with a general-purpose tool like Claude Code or Codex
- How to encode specific permissions into a harness
- The three components every harness needs
- How I used GPT-5.5 and Claude Opus to build the harness code itself (and where they both initially resisted)
- How to structure the artifacts your harness produces so the whole team can use the output
—
Brought to you by:
Bolt.new—Turn your idea into a real product
Customer.io—Build customer engagement campaigns from a single prompt
—
In this episode, we cover:
(00:00) What is an AI harness?
(03:19) When to build a harness
(04:33) Why Claire picked bug triage
(06:00) Why not just use Claude Code?
(07:48) Demo: The custom harness interface
(11:04) Architecture: runs, tasks, tools, and artifacts
(13:44) Building it with Codex and Claude
(15:08) Code map and file layout
(16:51) A look at the code
(19:18) The live investigation result
(21:01) How to build your own harness
—
Tools referenced:
• Claude Agent SDK (Anthropic): https://code.claude.com/docs/en/agent-sdk/overview
• Claude Sonnet 4.6 (model used inside the harness): https://www.anthropic.com/news/claude-sonnet-4-6
• Claude Opus (used to build the harness): https://www.anthropic.com/claude/opus
• GPT-5.5 (Codex, used to build the harness): https://openai.com/index/introducing-gpt-5-5/
• Ink (terminal UI library for Node.js): https://github.com/vadimdemedes/ink
• Sentry (error monitoring): https://sentry.io/
• Linear (project management): https://linear.app/
• GitHub: https://github.com/
• Vercel: https://vercel.com/
—
Where to find Claire Vo:
ChatPRD: https://www.chatprd.ai/
Website: https://clairevo.com/
LinkedIn: https://www.linkedin.com/in/clairevo/
—
Production and marketing by https://penname.co/. For inquiries about sponsoring the podcast, email jordan@penname.co.
Follow this show on AudioBrevity and new episodes get summarized for you. Create a free account