Simon Willison — Independent developer and writer documenting practical experiments, tools, and deep analysis around large language models, generative AI, web development, security, and emerging programming workflows through detailed posts and daily TILs.
381 articles from this blog
A quote from Tarn Adams, co-creator of Dwarf Fortress, about dwarf behavior and AI, shared on Simon Willison's blog.
Python 3.15.0 release candidate 2 announced, final RC before October release, urging third-party maintainers to test and publish wheels.
Introducing wrapture, a Python library for tracing and testing via monkeypatching, with OpenTelemetry support and config-based tracing.
An in-depth analysis of OpenAI's ChatGPT Work, covering its dual products, features, model options, and comparisons to standard ChatGPT.
Tencent's new Hy4 open-weight LLM with 770B params, 49B active, 1M context, and reasoning effort modes.
Security exploits are now found within minutes of bug rumors due to AI agents, challenging open-source embargo practices.
Analysis of a prompt injection attack bypassing Claude Code Opus 5 Auto Mode, highlighting the need for sandboxing coding agents.
26th August 2026 - Link Blog Qwen3.8-Flash-Next (via) Another open weights model from Qwen. This one is "a multimodal MoE model that also serves as an
Paul Dix discusses AI's ability to write and refine 1M lines of code, producing reliable software running on millions of machines, highlighting the role of verification systems.
EVE Online begins its long-awaited migration from Stackless Python 2.7 to Python 3, using futurize on 2.4M lines and manual review of ~20K differences.
Release notes for llm-anthropic 0.27, an LLM plugin for Anthropic models, focusing on compatibility with anthropic v1.0.0.
A Linux pattern where a SQLite database file is made executable using ELF format and binfmt_misc.
Anthropic's top AI model Fable 5 sees low adoption due to cost, while cheaper models and OpenAI's GPT 5.6 drive revenue growth.
A reflection on how the high cost of Fable AI model changes coding workflow strategies, quoting Drew Breunig.
Release notes for llm 0.33, a command-line tool for accessing large language models, with updates to OpenAI library, embedding keys, and template combining.
Release of llm-openrouter 0.7, an LLM plugin for OpenRouter models, now compatible with LLM 0.32 and supporting the Responses API.
Thomas Ptacek argues for building native UIs instead of TUIs, citing coding agents lowering the cost of GUI development.
Matt Webb discusses using ChatGPT as a tutor to learn quaternions for his app's AR mode, highlighting AI's role in facilitating learning.
Analysis of ChatGPT search's increased use of the site: operator, based on Promptwatch tracking data, and its implications for GEO.
A zero-dependency TypeScript service using Bun 1.4's Bun.WebView to provide a shot-scraper-style JSON API for JavaScript evaluation and screenshots without Puppeteer or Playwright.
