เกี่ยวกับตอนนี้
Jordan Urbs, an AI educator who works full time with Venice AI, joins The Daily Stack to discuss reports of a possible Iran deal, the China summit, and how he went from Bitcoin and the DAO to building the AI Captains Academy. He explains context files, subagents and skills, and makes the case for building with frontier models like Claude Opus 4.7 while running workflows on cheaper open source models like Kimi K2.6, Qwen 3.6 and Gemma 4. Sam Knight later joins to talk about enterprise AI agents, private inference, Micron ($MU) crossing a trillion dollars, Codex computer use and using Obsidian as a second brain.
This podcast is powered by Scanley, AI visibility for banks and credit unions. Get your 1st month free with code DAILYSTACK at https://meetscanley.com
Taking a BTC loan? Our friends at Ledn are offering .25% off their lowest rates with our link: https://www.ledn.io/partner/dailystack
Chapters:
0:00 Intro: Jordan Urbs joins, yerba mate and alcohol tolerance
7:19 Iran and Trump deal talks, Iranian protest in London
9:13 Multipolar order, China summit, Boeing and Nvidia
14:13 Jordan Urbs's path from Bitcoin and the DAO to AI
22:16 Inside AI Captains Academy: models, tokens, knowledge base
31:23 Screen share: CLAUDE.md, subagents and skills
38:50 Build with frontier models, execute with open source
49:17 Sam Knight joins: open models inside enterprises
1:04:29 Trump drops 90 day model approval, distillation claims
1:12:03 Venice vs OpenRouter and private inference
1:21:37 Micron hits a trillion, buying AI hardware
1:27:34 Sam Knight on Codex computer use and Obsidian
Key takeaways:
- Jordan Urbs recommends using frontier models like Opus 4.7 to build harnesses and workflows, then running them on cheaper open source models such as Kimi K2.6, Qwen 3.6, GLM 5.1 or Gemma 4.
- Jordan avoids security risks from downloaded skills by having Claude recreate any skill he likes rather than copying a random repo into his workspace.
- Sam Knight says that in enterprise testing, open models like Kimi K2.6 come close to GPT 5.5 output for most use cases, and fine tuning an open model beat building a model from scratch.
- Both guests warn that subscription token pricing from labs like Anthropic and OpenAI may not stay subsidized, especially if those companies go public.
- Sam Knight stores his context in an Obsidian vault that Codex updates daily, so he can switch between Codex, Claude Code or other agents without losing context.
Listen and follow The Daily Stack:
Spotify: https://open.spotify.com/show/0JQzqfVmkKiXuwSA6hyoDs
Apple Podcasts: https://podcasts.apple.com/us/podcast/the-daily-stack/id1831884653
Website: https://dailystackhq.com
X: https://x.com/dailystackHQ
Not financial advice.
