BitFlip.show

·E11

AI, Burnout, and Everything After

July 24
44 mins

Episode Description

What we cover
  • Linus Torvalds' apparent reversal from "AI is 90% hype" to supporting Sashiko for kernel patch review, his pointed stance against gatekeeping AI-assisted contributions, and the Software Freedom Conservancy's new AI guidelines that try to support, not force, and not shun any position on AI use.
  • Three core team members stepping back amid AI-contribution friction and burnout, and the crew's debate over whether Jellyfin should focus on stable "plumbing" and let the community build bespoke clients, add user-requested features directly, or pursue a Nabu Casa-style funding model to sustain development.
  • Geoff's first steps in deploying Hermes himself after last week’s episode, Alex's story of having Hermes diagnose a household voltage drop with TeslaMate and Home Assistant data, Hermes vs. Turnstone for deterministic workflows, and the limits of local LLMs for agentic tasks versus using frontier models to generate skill files for them.

First up, we trace Linus Torvalds' apparent reversal on AI, from calling it "90% hype" in 2024 to now backing Sashiko, an agentic code review tool for kernel patches. We go into the Software Freedom Conservancy's new AI contribution guidelines, and debate whether their "don't force it, don't shun it" stance is a coherent policy or just a way to avoid picking a side. Geoff also shares how he's been using AI to build his own tools, including a vibe-coded podcast recording app and a custom Jellyfin client for offline, transcoded playback.

Next, the crew unpacks a leadership shakeup at Jellyfin, where a desktop app developer left the org partly over pushback on his AI-assisted contributions, and two more core team members stepped back citing burnout and personal circumstances. We discuss whether Jellyfin's volunteer-only model is sustainable given the current flood of new users and AI-generated contributions, and debate whether the project should double down on stable "plumbing" and let the community build bespoke clients, or start shipping the features users have been asking for. Adam and Alex also make the case for a Nabu Casa-style funding model that could give Jellyfin the resources to avoid burning out its maintainers again.

Finally, the panel dives deep into their real-world use of Hermes agents. Geoff walks through his Outline-connected research assistant and automated morning briefing, while Alex shares his experience of using Hermes to correlate TeslaMate and Home Assistant data to diagnose a voltage drop at his house. Adam contrasts Hermes with Turnstone, a more deterministic, judge-monitored framework built for software development, and the group closes out talking through why local LLMs still struggle with context and how frontier models can be used to help make local agents viable.

Tags: Linux kernel, Linus Torvalds, Sashiko, Software Freedom Conservancy, open-source Jellyfin, self-hosting, home automation, Home Assistant, TeslaMate, Hermes agent, Turnstone, local LLMs, agentic AI.

Links
See all episodes