The hard part of AI coding isn’t generating code — it’s controlling quality, safety, and drift. Drawing from OpenAI’s Codex case study, Mitchell Hashimoto’s post on AI workflows, Stripe’s Minions project, and real-world experience, Kaushik and Iury break down harness engineering: the five pillars for shaping an agent’s environment, and what it looks like when teams build custom harnesses from scratch.
Full shownotes at fragmentedpodcast.com.
Show Notes #
Why it matters #
- Mitchell Hashmito first coined this term in his post My AI Adoption Journey
- Harness Engineering - OpenAI’s post on building their Codex codebase (~1M lines of code, 1,500 PRs merged, zero manually written)
Shaping the harness #
- The Feed’s Lost and Found - Iury’s newsletter consolidating harness engineering themes
- Agent legibility
- Closed feedback loops
- Persistent memory
- Entropy control
- Blast radius controls
Building the harness #
- Minions: Stripe’s one-shot, end-to-end coding agents - Stripe forked Goose to build custom agents for their codebase
- Goose - open-source coding agent from Block
- Superpowers by Jesse Vincent - skills that enforce a proper software engineering process
- Open Code - open-source coding agent you can fork and customize
Other resources #
- Agent Harness Glossary - Latent Patterns
- Towards self-driving codebases - Cursor
- Agentic Workflows - GitHub Next
- Future of Software Development - ThoughtWorks
Get in touch #
We’d love to hear from you. Email is the best way to reach us or you can check our contact page for other ways.
We want to hear all the feedback: what’s working, what’s not, topics you’d like to hear more on.
Co-hosts: #
We transitioned from Android development to AI starting with
Ep. #300. Listen to that episode for the full story behind our new direction.