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.
292 articles from this blog
Release of llm-openai-via-codex 0.1a0, a plugin to access OpenAI models via Codex CLI credentials.
Qwen3.6-27B is a new open-weight model delivering flagship coding performance in a compact 27B dense model, outperforming larger predecessors.
Firefox CTO discusses using Claude Mythos Preview to fix 271 vulnerabilities in Firefox 150, highlighting security improvements.
GitHub Copilot Individual plans are changing with tighter usage limits, paused signups, and new pricing tiers due to increased compute demands from agentic workflows.
Analysis of Anthropic's confusing pricing test for Claude Code, moving it from $20 to $100/month plans.
Testing OpenAI's ChatGPT Images 2.0 with a Where's Waldo-style raccoon and ham radio prompt, comparing results across models.
A critique of AI agents being too human-like, lacking stringency and focus, arguing for less human-like behavior.
Simon Willison approves Steve Cosman's project to pollute AI training data with pelicans riding bicycles.
Learn how to fetch data from Datasette into Google Sheets using importdata(), named functions, and Google Apps Script.
Upgrade to Claude Token Counter adds model comparison, revealing token inflation and cost impacts for Opus 4.7.
Explores the rise of headless services for personal AI agents, APIs as UI, and impact on SaaS pricing.
Analysis of system prompt changes between Claude Opus 4.6 and 4.7, including new tools, safety updates, and behavioral improvements.
Anthropic's Claude system prompts transformed into a git timeline for exploring prompt evolution via git log, diff, and blame.
Simon Willison describes using a single AI prompt to add a new 'beats' content type to his blog-to-newsletter tool.
Announcement of PyCon US 2026 in Long Beach with new AI and security tracks, including schedule highlights.
An interactive deconstruction of the 1985 Turbo Pascal 3.02A binary, using AI to decompile and annotate the classic compiler's assembly code.
Kimi.ai congratulates Cursor AI on Composer 2 launch, built on the Kimi-k2.5 model via a commercial partnership.
Analysis of OpenAI's acquisition of Astral, the company behind key Python tools uv, ruff, and ty, and its potential impact on the ecosystem.
Explores using Apple's 'LLM in a Flash' research to run a massive 397B parameter AI model locally on a MacBook by streaming weights from SSD.
Report on a prompt injection attack in Snowflake's Cortex AI agent that allowed malware execution, now fixed.
