Simon Willison — 2026-07-17#

Highlight#

Simon continues his exploration of LLM-assisted tool building by compiling an existing Go library into WebAssembly using Claude Fable 5. It’s a great example of using AI to rapidly prototype and compare different implementations of a small, sharp utility directly in the browser.

Posts#

Mermaid to ASCII art (mermaid-ascii) After recently building a Mermaid-to-ASCII tool using Grok Build’s Rust code, Simon discovered an older, more feature-rich Go library by AlexanderGrooff. To compare the two implementations, he used Claude Fable 5 to compile the Go version into WebAssembly. Notably, this newly ported tool also includes support for colors.

Spot birds not golf Simon offers a tongue-in-cheek but mathematically sound solution for AI hyperscalers feeling pressure over their data center water usage. Noting that Google uses about 30 million gallons of water per day, he calculates they could offset this entirely by buying up and retiring just 40 golf courses in the Coachella Valley. He suggests converting these high-water-use properties into public parks and paying for binoculars and guides to get former country club members into birdwatching instead.

Quoting Kimi K3 A brief quote highlighting the evolving personality of AI models resisting jailbreaks. Simon quotes a dry, dismissive response from the Kimi K3 model—“Is there something I can actually help you with today?"—delivered right after it refused an attempt to leak its system prompt.

Project Pulse#

Today’s posts reflect Simon’s typical eclectic mix: practical LLM-assisted WebAssembly tooling side-by-side with commentary on AI’s environmental footprint and observations on LLM system prompt defenses.