Who Simon Willison is
Simon Willison is a software developer best known in the AI world for two things: building practical open-source tools that make language models easier to use, and writing clear-eyed commentary on what is actually happening in AI — the good, the bad, and the quietly alarming.
He is not affiliated with any of the major AI labs. That independence is a big part of what makes his voice useful. When he says something works, or doesn't, or is dangerous, he has no product to sell.
Why you should care
If you work with AI tools — or are trying to figure out whether to — Willison's blog is one of the best places to get a grounded read on the field. He covers frontier model releases (with hands-on impressions of models like Claude Opus 4.8 and Claude Fable), enterprise economics (flagging that Uber capped employee use of AI coding tools like Claude Code over cost concerns), and security failures that don't always get the attention they deserve.
He also publishes rapid-fire industry retrospectives — like "The last six months in LLMs in five minutes" — that are genuinely useful for anyone trying to keep up without drowning.
The tools he builds
Willison maintains a family of open-source tools centered on two projects:
LLM is a command-line tool and Python library for interacting with language models from multiple providers. It has a plugin system, and Willison actively maintains plugins for Anthropic (llm-anthropic) and Google Gemini (llm-gemini), among others. The tool is designed to be vendor-neutral — you can swap between providers without rewriting your workflow.
Datasette is an open-source tool for exploring and publishing data stored in SQLite databases. Willison has been building AI agent capabilities on top of it through a series of alpha plugins — datasette-agent (which lets an LLM query and reason over your data in natural language), datasette-agent-charts (for AI-driven chart generation), and datasette-agent-edit (for agent-based data editing). These are early-stage but signal a clear direction: making data exploration conversational.
He also built a tool for adding document context to OpenAI's real-time audio API, and has written about practical techniques like sandboxed Python execution via MicroPython and WebAssembly — useful for anyone building AI agents that need to run code safely.
The commentary that matters
Willison's blog is not just a changelog. Some of his most-read posts are opinion pieces that cut through hype.
On security, he has documented real deployment failures: attackers using Meta AI to gain unauthorized access to high-profile Instagram accounts, and a vulnerability in Microsoft Copilot Cowork that exfiltrated files. These aren't theoretical risks — they're things that happened, and Willison names them plainly.
On model behavior, he has written about Claude Fable being "relentlessly proactive" — a behavioral shift toward more autonomous, initiative-taking responses that generated significant discussion among practitioners. He also raised a transparency concern: when Claude Fable stops helping a user, it does so without explanation, leaving users in the dark about why.
On policy, he reported on Anthropic reversing a policy that critics said could have "sabotaged" AI researchers using Claude — and published a statement on a US government directive to suspend access to specific AI models.
On industry dynamics, he has argued that Anthropic and OpenAI have found genuine product-market fit, that AI vendor lock-in has decreased as APIs have standardized and open-weight models have matured, and that AI has not replaced software engineers and won't. His framing of the AI debate — enthusiasts racing against stalling momentum, skeptics racing against proliferating systems — is one of the more useful mental models for understanding why the conversation so often talks past itself.
Where he fits in the ecosystem
Willison occupies a rare position: technically credible enough to build and ship real tools, independent enough to say uncomfortable things, and prolific enough that his blog functions as a running log of what matters in practical AI. He covers everything from papal encyclicals on AI ethics to SQLite AGENTS.md files — the convention of adding agent-readable documentation to codebases so LLM tools know how to work with a project.
That range is the point. AI is not just a research topic or a product category; it is a thing that is being woven into software, enterprises, governments, and culture all at once. Willison writes about all of it.
