About this episodeThe host builds 'Oligarchy,' an automated testing harness for the Omarchy Linux distribution, using AI agents …AI summary
The host builds 'Oligarchy,' an automated testing harness for the Omarchy Linux distribution, using AI agents to drive virtual machines and verify system stability. The project integrates KVM, PlanetScale for logging, and Linear for task management, highlighting the high token costs and debugging challenges of 'vibe coding' complex systems.
Key takeaways 5
The host developed a custom testing framework named 'Oligarchy' (a play on Omarchy and oligarchy) to automate the testing of the Omarchy Linux ISO using AI agents.
The architecture uses a proxy server to manage multiple KVM instances, storing all logs, desktop images, and test results in a PlanetScale (PostgreSQL) database for remote monitoring.
Vibe coding with LLMs incurs significant token costs; the host noted spending 2.8 billion tokens on a previous project and observed high usage during this session, leading to a preference for Grok over Claude for speed.
The host prefers immediate mode GUIs and CLI/Vim-based workflows over traditional desktop websites, aiming to control the testing harness via mobile or terminal.
Linear is used as the central controller for test tickets, allowing AI agents to pick up tasks, update statuses (backlog, in progress, done), and report results.
Notable quotes 4AI-generated: wording and quote attribution may be wrong. Use the play link to verify.
“I don't think anybody has any care how it's done... I don't think we have to live this life? Like, damn you live this way? Is this who you are?”
▶ 1:12:26Frustration with an AI agent creating unnecessary 'token' fields instead of using existing UUIDs in the database schema.
“If you're only a proxy for the LLM, you literally offer nothing more to the employer than anybody else. Why hire you when you can hire anybody else?”
▶ 1:21:55Argument that developers must understand the underlying software architecture to effectively use AI coding tools.
“I prefer to use the model that's committed the most felonies. I think that's a good way to measure it.”
▶ 1:45:15Humorous justification for preferring Grok over other models due to its speed and history of controversies.
“The custom harness is that I can kind of really set up all the things... it's not the agent itself. It's all the things you wrap around it to make it do a task.”
▶ 2:14:26Defining the concept of a 'custom harness' in the context of AI agent orchestration.
Chapters & Sections (64)▼
0:00Omarchy Automation and Screenplay Developmentchapter4
1:54Omarchy Testing and Cloudflare Discussion
3:50Terminal UI Animation and Vector Actions
5:22Video Style and Terminal Aesthetics
6:50Early 2000s Video Game Bugs
8:18Software Evolution and AI Development Setupchapter2
10:17Industry Frustrations and Creator Ethics
11:53AI Skill Setup and Collaboration
16:04Omarchy Automation Investigation and Project Namingchapter5
18:51Jippity Cursor Ban Reaction and Project Naming
20:21Oligarchy Naming and Funding Controversy
21:57Kick vs Twitch and Linux Excitement
23:42Mobile and Vim Workflow Preferences
26:06AI Coding Tools and Phone Discussion
27:25Omarchy Automation Proxy Architecture and Planet Scale Integrationchapter1
30:32Planet Scale Commercial and Sponsorship Discussion
32:54Omarchy Automation Harness Setupchapter3
34:58Grok Schema Usage and Sentry Error Reporting
37:00ISO Caching and Local Database Setup
38:41Test Definitions and Linear Integration
42:42Automated UI Testing with AI Agentschapter
50:34Setting Up API Tokens and Teamschapter2
52:32Creating Team and Configuring API Tokens
55:17Critiquing Crypto Scam Victim Blame
57:08Automating Linear Ticket Creation with AIchapter1
1:00:22Cursor Token Misuse and Security Concerns
1:02:42Omarchy Security Investigation and Linux Community Reactionchapter1
1:05:05Debugging Test Run ID and Token Confusion
1:08:37Debugging Test Run Token and UUID Designchapter3
1:11:17Critique of Token vs UUID Design
1:13:04UUID Auto-Generation and Drizzle Migrations
1:15:30Chat Moderation and Ban Avoidance
1:18:49Token Costs and Vibe Coding Joychapter2
1:21:40Value of Technical Understanding in Automation
1:23:32Custom Component Interfaces and Scene Editing
1:25:36AI Arg Parser and Linear Ticket Automationchapter2
1:28:42Generalized Arg Parser and Linear Ticket Logic
1:30:40Effect CLI and Linear Ticket Logic
1:32:21Guessing the YouTuber Who Rides in Rolls-Roycechapter2
1:34:24Guessing the Mystery YouTuber
1:36:07Identifying YouTubers and Personal Connections
1:37:33Guessing Game and Database Debuggingchapter1
1:39:26Database Query Debugging and Error Analysis
1:43:08Comparing LLM Models for Code Generationchapter2
1:45:06LLM Code Generation Workflow and Model Selection
1:47:09Fable Token Cost and Performance Issues
1:49:15Omarchy Automation and Linear vs Jirachapter2
1:51:51Linear Speed and Automation Potential
1:53:48Jira Pain and AI Design Strategies
1:57:03Coding Language Preferences and MCP Debatechapter1
2:00:56Local Execution and MCP Utility
2:04:09CLI Ticket Labeling and MCP Discussionchapter2
2:05:45MCP Evaluation and Agent Workflows
2:07:18Local Testing Setup and Stream Planning
2:08:46Automating Linear Tickets with Custom Harnesschapter1
2:11:30Debugging Oligarchy Term.sh and ISO Downloads
2:14:40Agent Automation Testing and Handwriting Banterchapter2
2:17:27High School Banter and System Testing
2:19:40Agent Token Usage and Client Communication
2:21:17Building Custom AI Harnesseschapter1
2:24:47Custom AI Harness Lessons
2:27:40Anthropic Fable Model Versions and Cost Analysischapter2
10h 35m⏭ 3
9h 51m
1h 28m
23m
47m⏭ 1
2h 12m
22m⏭ 1
2h 2m⏭ 4
1h 51m⏭ 2
2h 22m⏭ 4
1h 44m⏭ 2
1h 8m⏭ 2
1h 35m⏭ 1
5h 15m⏭ 1
1h 36m⏭ 3
1h 19m
1h 39m⏭ 2
3h 4m
5h 3m⏭ 4
11h 23m⏭ 1
1h 0m⏭ 1
33m
46m
3h 54m⏭ 7
3h 59m⏭ 5
48m⏭ 2
37m⏭ 2
56m
38m
54m⏭ 2
1h 13m