Steve Yegge posted about Gastown in January 2026. I read it at the time but didn’t really understand it. Some concepts resonated with me- but I found the language and metaphors of gastown obfuscatory. There are parts of the writing that I associate with LLM-psychosis. I’m disappointed in myself for that reaction. My bias kept me from understanding with what he was actually doing. The nano-banana graphics didn’t help. There’s an uncanny valley quality to them. Whimsical, but slightly off. My mind rushed to label the ideas as lower value because the marmots are for weirdos.
Fast-forward to April: I’d been building my own orchestration solutions around Kanban boards and Claude Code / OpenCode. Something prompted me to re-read the Gastown post in April. I discovered I that had duplicated parallel versions of his orchestration (minus beads, which looks amazing). I didn’t realize what he was doing, and ended up with my own version of this thing. Parallel innovation is the theme of 2026.
If you haven’t read it yet, take some time to read the Gastown post. It’s hard to absorb on the first pass. If you lose all patience, the interview links below will help you get oriented on the important parts without the hyperbole and less dependency on mastery of the metaphors.
————————
From TWiT — Full episode
(note: Leo Laporte’s team can wander into tangents that go nowhere. They’re usually easy to spot, but set your own tolerance level going in.)
- 8 stages of vibe coder evolution
- What is Gastown?
- Is vibe coding reasoning-capable?
- What is Gastown? (second summary)
- What is Beads?
- Gastown is just a tool for vibe coding
- Steve’s workflow
- Steve’s workflow (continued)
- The human only works on the hard problems
- The AI vampiric effect
————————
From AI Tinkerers — Full episode
- Exponential growth of AI
- Why do LLMs find bugs in their own code?
- LLMs are better at evaluating outputs than generating them — it’s easier to spot what’s wrong than to build from scratch.
- Technique: ask it to build something, force it to propose an alternative, then have it review both and synthesize a better approach.
- Code review techniques
- Beads
————————
Another interview (good material, but heavy on ads): YouTube
————————
Beads quickstart: gastownhall.github.io