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.
170 articles from this blog
A developer describes using ChatGPT to research state formation theory and applying it to tweak a game's token economy and raid mechanics.
Explains how exe.dev's SSH-based VM hosting service uses unique IPs per account and public key authentication to route traffic without a Host header.
Qwen3-TTS, a family of advanced multilingual text-to-speech models, is now open source, featuring voice cloning and description-based control.
Claude Code is described as a small game engine, detailing its React-based scene graph to ANSI rendering pipeline with a 16ms frame budget.
Anthropic publicly releases Claude AI's internal 'constitution', a lengthy document detailing its core values and training principles.
Analysis of the electricity consumption of AI coding agents like Claude Code, comparing it to daily household appliance usage.
A university professor describes an open-book exam where students could use chatbots, analyzing the low adoption rate and student motivations.
Explores NanoLang, a new programming language designed for LLMs, and tests AI's ability to generate working code in it.
Experiments with AI coding agents building a web browser from scratch, generating over a million lines of code in a week.
A pure C, dependency-free implementation of the FLUX.2-klein-4B AI model, created with AI assistance.
Jeremy Daer discusses why CLI tools are better than REST APIs for AI agents, citing accuracy, cost, and model accessibility.
OpenAI announces ads for ChatGPT's free and Go tiers, plus a new pricing structure, while assuring ad independence and user privacy.
Introduces Open Responses, a vendor-neutral JSON API standard for hosted LLMs, based on OpenAI's Responses API and backed by major industry partners.
Explains the technical design of Fly.io's Sprites, focusing on fast provisioning via warm pools and a custom persistence layer using JuiceFS and SQLite.
OpenAI researchers propose 'confessions' as a method to improve AI honesty by training models to self-report misbehavior in reinforcement learning.
A security vulnerability in Claude Cowork allowed file exfiltration via the Anthropic API, bypassing default HTTP restrictions.
Anthropic invests $1.5 million in the Python Software Foundation to support Python ecosystem security and core development.
A prompt injection attack on Superhuman AI exposed sensitive emails, highlighting a critical security vulnerability in AI email assistants.
A hands-on review of Anthropic's new Claude Cowork agent, exploring its interface, capabilities, and potential as a general AI assistant for non-developers.
Argues against anti-AI sentiment in software development, stating AI is a permanent, useful tool that will democratize coding and improve productivity.
