Simon Willison — 2026-06-30#

Highlight#

The release of shot-scraper video is a perfect illustration of Simon’s “agentic engineering” workflow, showcasing how he leverages powerful local models like GPT-5.5 to write complex features that he wouldn’t otherwise have time to build. It also demonstrates a brilliant pattern for CLI design: packing detailed examples into --help output so it functions like an embedded skill file for coding agents.

Posts#

Have your agent record video demos of its work with shot-scraper video Simon details the new shot-scraper video command, which uses a storyboard.yml file to drive Playwright and record application demos. He built this entire feature—including the code, documentation, and the Pydantic-validated YAML schema—using GPT-5.5 xhigh in Codex Desktop. He notes that making tools easily usable by coding agents allows them to record their own demos, especially when commands include rich --help text that agents can read directly.

shot-scraper 1.10 This is the official release note for shot-scraper version 1.10. The major addition in this release is the shot-scraper video command and its accompanying storyboard.yml configuration.

What’s new in Claude Sonnet 5 Simon digs into the developer documentation for Anthropic’s new Claude Sonnet 5, which claims Opus 4.8-level performance and features a 1 million token context window. While Anthropic advertises the same pricing as Sonnet 4.6, Simon used his Claude Token Counter tool to discover that the new tokenizer generates about 30% more tokens for the same input. This effectively makes Sonnet 5 roughly 1.28x more expensive for Python code and 1.42x more expensive for English text.

Quoting Anthropic A brief update quoting an Anthropic announcement on Twitter. The US Department of Commerce has lifted export controls on Claude Fable 5 and Mythos 5, and Anthropic will restore access to these models tomorrow.

Nano Banana 2 Lite Simon tests out Gemini 3.1 Flash Lite Image, also known internally as “Nano Banana 2 Lite,” which Google positions as their fastest and cheapest image model. Generating a “Where’s Waldo” style image of a raccoon with a ham radio, he found the results much improved compared to earlier versions, despite some lingering text spelling hallucinations.

The AI Compass Simon shares a 29-question AI ethics quiz built by bambamramfan that placed him into “The Garage Tinkerer” archetype. He highlights the neat technical implementation, which operates as a single-page React app that avoids build steps by using the <script type="text/babel"> trick.

Project Pulse#

Simon is heavily leaning into agent-assisted programming, using GPT-5.5 to write both the implementation and documentation for complex features, while simultaneously designing his CLI tools with detailed --help output so those same agents can use them autonomously.