Today's selection tells the story of AI coding tools crossing a threshold — from experimental to exponential. Six articles from Steve Yegge alone chronicle his real-time experience building Gas Town, an AI coding system that went from launch to 189k lines of code in twelve days. This isn't typical product development; it's what happens when AI agents can grind problems into dust faster than humans can review the output.
I chose Yegge's pieces because they capture something most coverage misses: the messy, chaotic reality of building with AI at scale. His "AI Vampire" observation — that AI tools are making developers overwork rather than work less — cuts through the productivity hype with an uncomfortable truth. Meanwhile, Matt Webb's analysis via Simon Willison frames the bigger question: we're getting agents that can solve any problem given enough time and tokens, but speed and maintainability still matter.
What surprised me most was the breadth of existential thinking in today's pipeline. Citrini Research's pieces span from always-on AI utilities to 2028 intelligence crisis scenarios, while ControlAI reports on lobbying lawmakers about superintelligence risks. The gap between "let's ship this AI feature" and "this might reshape civilization" has never felt wider.
The throughline here is velocity — both in development cycles and in our collective sense that something fundamental is accelerating. Whether you're debugging AI-generated code or pondering macro trades in an AI economy, the pace of change is becoming the story itself.
Yegge launches a federated version of Gas Town to handle 100x token increases by connecting multiple AI coding environments together.
Webb argues that AI agents will solve any coding problem given enough tokens and time, but we need them to work quickly and maintainably instead.
Yegge merged 100+ PRs from 50 contributors adding 44k lines of unreviewed AI code to Gas Town, growing it to 189k total lines.
Three days after launching Gas Town, Yegge reflects on creating an AI coding system that's "just barely smart enough" but rapidly improving.
Yegge identifies a concerning trend where AI tools are causing developers to overwork rather than making their jobs easier.
After a year of building with AI (including Beads and Gas Town), Yegge shares insights on riding AI's exponential improvement curve.
Latest Claude Code update fixes message delivery issues in the Cowork Dispatch feature.
Analysis of AI systems that run continuously without human prompting, representing a shift toward autonomous digital utilities.
A thought exercise examining how rapid AI advancement might trigger a global intelligence crisis by 2028.
Non-profit organization reports on their work educating hundreds of thousands about superintelligence extinction risks and lobbying lawmakers.
Analysis of the fundamental limits between digital AI capabilities and physical world constraints.
Yegge's intuitive analysis of Anthropic's trajectory, comparing their hiring selectivity to elite sports and sensing major momentum building.
Macro analysis connecting AI productivity gains to geopolitical shifts and new investment opportunities.