SimonWillison

Simon Willison does not run a startup. He does not have a venture capital firm behind him. He does not have a PR team, a growth strategy, or a head of marketing. What he has is a blog that has published without interruption since 2002, a GitHub profile with 954 public repositories, and an uncanny habit of naming things that matter.

He named prompt injection - the attack vector that makes every AI agent a potential security liability - in 2022, by analogy to SQL injection, when he saw a tweet from Riley Goodside describing the behavior. Now it's in every AI security framework on the planet. He wrote a blog post in May 2024 calling unwanted AI-generated content "slop" - and Merriam-Webster named it their Word of the Year for 2025. He runs a benchmark called "pelican riding a bicycle" where he asks every LLM to draw exactly that, tracks the results on GitHub, and calls it his "dastardly multi-year plan to trick AI labs into making genuinely good SVG art." Google used it to demo Gemini 3.1 Pro.

He is, in other words, a man who keeps accidentally defining the conversation.

As of 2026, Willison lives in Half Moon Bay, California - about 30 miles south of San Francisco - where on a clear day he can see the Pacific Ocean. He wants to build a system that alerts him when whales are visible offshore. This is, more or less, what Simon Willison is: someone who sees a problem, sees a tool missing, and builds it. Then publishes it. Then blogs about it. Then moves on to the next one.

He reports writing 95% of his code from his phone now. He says he is mentally exhausted by 11am from AI-native development. He is also more productive than at any point in his 25-year career.

The story of Django is the story of a side effect. Willison wasn't trying to create a web framework that would eventually power Instagram. He was a university student on a year-long industrial placement, helping a regional American newspaper move its editorial workflow online. The newspaper's deadline-driven culture demanded fast results. Fast results demanded reusable tools. Reusable tools became a framework.

He returned to the University of Bath, finished his Computer Science degree in 2005, and by the time he graduated, Django had already been open-sourced and was being adopted by developers around the world. He never really returned to academia. He didn't need to. He had built something that would shape the next two decades of web development.

Django's core philosophy - "batteries included," convention over configuration, the admin interface that generates itself from your models - bears the marks of a newspaper developer who needed to ship things before print deadline. It is opinionated in the right places and flexible in others. It has remained the dominant Python web framework for 20 years, which is nearly eternal in software terms.

After a stint at Yahoo, a role as Software Architect at The Guardian, and co-founding Lanyrd (a social conference directory acquired by Eventbrite in 2013), Willison spent several years as Engineering Director at Eventbrite. By 2017, he had started building Datasette on the side. By 2019 he was on a Stanford John S. Knight Journalism Fellowship, full-time, exploring how open-source tools could serve data journalism. By 2023 he had become genuinely, fully independent - funded entirely by GitHub Sponsors, supported by Mozilla's MIECO program and GitHub Accelerator.

Datasette is his most consequential independent work. It takes a SQLite database file and turns it into a browseable, queryable, shareable web application with zero configuration. Investigative journalists use it to publish document troves. Scientists use it to share research data. Governments use it to expose public records. It has 11,000+ GitHub stars and an ecosystem of plugins that Willison and community contributors have built over seven years.

Then came the LLM wave - and Willison rode it differently than most.

Rather than building a startup around AI, he built a tool called llm - a command-line interface and Python library that gives you access to OpenAI, Anthropic, Google Gemini, Mistral, Ollama (for local models), and many others through a single, consistent interface. It has 11,700+ GitHub stars. He keeps adding plugins as new models emerge - which in 2024 and 2025 meant essentially constantly.

In 2025 alone, he shipped more than 110 tools. He blogs about each one. He runs Datasette Public Office Hours for the community. He maintains a TIL (Today I Learned) site at til.simonwillison.net with 575+ entries cataloguing technical discoveries. He is a one-person engineering organization producing the output of a small team - and he does it on GitHub Sponsors, by the goodwill of 645+ people who believe his work is worth paying for.

His productivity secret is less mysterious than people assume. He practices what he calls "conference-driven development" - committing to give a talk at a conference, then using the accountability of that commitment to force himself to ship the feature he wants to present. He spent years doing weekly "weeknotes" - public accountability posts documenting what he shipped. He is now on a daily blogging streak started in January 2025, taking 10-15 minutes per post, benefiting from 22 years of practice and an internal publishing pipeline that partially automates the newsletter from blog content.

Willison occupies an unusual position in AI commentary. He is not a researcher. He is not employed by an AI lab. He is not funded by one. He gets early access to new models from OpenAI, Anthropic, Google, and Mistral - and he has never accepted payment from any of them. This is not a detail. In an era when almost every prominent AI voice has a financial relationship with a lab, his independence is architecturally unusual.

The result is commentary you can actually trust. When he says a model is good at something, it is because he tested it. When he raises an alarm about an AI security risk, it is because he saw it happen. His 2023 post about Microsoft Bing's early AI behavior got 1.4 million page views and was amplified by Elon Musk on X. Not because he was trying to go viral - because he had something precise and accurate to say and said it without hedging.

The newsletter - Simon Willison's Newsletter on Substack - has grown to 54,000+ subscribers without advertising or paid growth. He generates it semi-automatically from a Datasette instance that tracks his blog content, running it through a pipeline he built himself. The paid tier ($10/month via GitHub Sponsors) provides a curated monthly summary of the most important LLM trends of the past 30 days. He covers new model releases, security research, open-source tooling, and the changing patterns of how developers actually use these systems.

In November 2025 he declared an inflection point: AI coding agents had crossed from "mostly works" to "actually works." He wrote about it carefully - not as hype, but as a behavioral observation from someone who had been tracking the space methodically for three years. By early 2026, he reported writing 95% of his code from his phone, and being mentally exhausted by 11am. He called it "agentic engineering" and began writing a serialized book on the patterns involved.

There is something almost perversely consistent about Willison as a person. He started his blog in 2002. He has never stopped. He was an early OpenID evangelist. He built one of the first pingback implementations. He contributed to the Incutio XML-RPC Library for PHP that ended up in both WordPress and Drupal. He is the kind of engineer who shows up early, builds the foundation, and then moves on before it becomes fashionable to care about the foundation.

He married Natalie Downe, with whom he co-founded Lanyrd - a social conference directory that they pitched to Y Combinator, got funded (YC W11), and eventually sold to Eventbrite in 2013. That is, by any measure, a successful startup story. For Willison, it was apparently an interesting episode on the way to doing what he actually wanted: building open-source tools and writing about them.

He uses GitHub Issues as personal to-do lists across 250+ active projects. He has 954+ public repositories. He describes himself as "a jack of all trades, master of none" - which is accurate in the same way that calling the ocean "some water" is accurate.

When the CIA discontinued the World Factbook ZIP file - a comprehensive, freely available database of country information that had been available for years - Willison noticed, archived the final copy himself, and made it available on GitHub before anyone else had thought to ask whether it should be preserved. This is another signature move: the infrastructure that seems permanent is not, and someone needs to care.

He is funded by 645+ GitHub Sponsors. This is the architecture of his independence. No VC term sheet. No acquisition conversation. No performance review. The people who find his work valuable pay what they think it's worth, and that's enough for him to keep building. He is a Python Software Foundation board member. He has been recognized as a GitHub Star. He speaks at PyCon regularly. He does all of this from Half Moon Bay, with the ocean visible when the fog clears.

The whale-spotting notification system he wants to build - for when whales are visible from his window - has not yet been built. He has 109 other things to build first. The whale will wait. So will the perfect pelican SVG. The blog will publish tomorrow regardless.

simonwillison.net has been running since 2002. That is longer than Twitter, longer than GitHub, longer than the iPhone. It has outlasted del.icio.us, Google Reader, Lanyrd, most of the companies whose tools it has covered, and two complete cycles of "blogging is dead." It is not a personal brand exercise. It is not a content marketing funnel. It is a working engineer's notes, published in public, for 22 years, because Willison finds writing clarifying and thinks other people might find it useful too.

His stated goal since January 2025 is one post per day. He spends 10-15 minutes per post. He says the speed comes from 22 years of practice, a writing environment he has optimized obsessively, and a mental model of his audience (other engineers who want the bottom line fast) that he has developed over those same 22 years.

The TIL (Today I Learned) site - til.simonwillison.net - is a companion publication: 575+ short technical discoveries, catalogued by date and topic. Want to know what Simon Willison learned about PostgreSQL in March 2023? It's there. It's indexed. It's searchable. Of course it is - it runs on Datasette.

In 2026, Willison is writing a serialized book on agentic engineering - the patterns, principles, and failure modes of building reliably with AI coding agents. He has more material than most people who would attempt this: three years of intensive, documented experimentation, a blog full of case studies, and the hard-won experience of having shipped 110 tools in a single year using AI-assisted development.

He is speaking at PyCon US 2026. He ran a live demonstration in February 2026 where he vibe-coded a custom macOS SwiftUI app during a presentation on "The State of LLMs" - building the app to display his slide data rather than using traditional presentation software. The app worked. The presentation shipped on time. This is what agentic engineering actually looks like from the inside.

Willison's aspirations are not corporate. He is not building toward an acquisition or an IPO. He wants to help people explore, analyze, and publish their data. He wants to document what agentic engineering actually is, before the hype cycle defines it incorrectly. He wants the whale-spotting notification system. He wants, eventually, a genuinely good SVG of a pelican on a bicycle.

He is, in a tech world full of people who use the word "ecosystem" to mean "moat," one of the few engineers who actually means it when he says the work is the point. The output - 954 repos, a 22-year blog, a vocabulary word in the Merriam-Webster dictionary, a framework that runs the web - is just what happens when someone does that kind of work for long enough.