Simon Willison

Co-creator of Django and creator of Datasette; writes daily at simonwillison.net.

Simon Willison did not write this page. ppll compiled it from their own words, each linked to the source. Is this you? · Report a mistake

Blog x.com GitHub

Latest GitHub

recent-california-brown-pelicans — Publishing an atom feed of recent California Brown Pelican sightings from iNaturalist 16 Oct 2026

What they believe

If Anthropic's revenue figures are accurate, other labs are spending more than half a billion dollars to distill Anthropic's models.

If true, and Q2 revenue was $11.5B, then that's more than half a billion dollars spent by other labs to distill Anthropic's models!

x.comSaid 15 Oct 2026This on korrents
Sampling profilers are very low overhead and can provide deep insight into running systems without needing to instrument them first.

They’re very low overhead and can provide a deep insight into running systems, without needing to instrument those systems first.

simonwillison.netSaid 12 Oct 2026This on korrents

The bold line is korrents’ wording; the quote is theirs. These 2: what they wrote before what they said aloud, one per source, newest first.

All 54 beliefs, on korrents

Feed

As its own page →

Everything, newest first — across every channel.

Sources

16 October

Simon WillisonGitHub 16 Oct

+ 12 more

15 October

Simon WillisonBlog 15 Oct

Simon WillisonKorrents 15 Oct

korrents.com
If Anthropic's revenue figures are accurate, other labs are spending more than half a billion dollars to distill Anthropic's models.

Anthropic

Their words

If true, and Q2 revenue was $11.5B, then that's more than half a billion dollars spent by other labs to distill Anthropic's models!

@simonw on Xx.com

14 October

Simon WillisonBlog 14 Oct

13 October

Simon Willison 13 Oct

x.com
BlogRedesigning source control at Anthropic
Anthropic
Blogllm-gemini 0.35
LLMsGoogle

+ 2 more

BlogQuoting Geoffrey Huntley
Blogweb-edit 0.1a1
AnthropicPython

12 October

Simon Willison 12 Oct

+ 2 more

Blogweb-edit 0.1a0
LLMs
Blogdatasette-ports 0.4

Simon WillisonKorrents 12 Oct

korrents.com
Sampling profilers are very low overhead and can provide deep insight into running systems without needing to instrument them first.

Their words

They’re very low overhead and can provide a deep insight into running systems, without needing to instrument those systems first.

simonwillison.net

11 October

Simon WillisonBlog 11 Oct

Dwarf Fortress uses version control now

10 October

Simon WillisonBlog 10 Oct

Python 3.15.0 added to actions/python-versions
OpenAIPython

9 October

Simon Willisonx.com 9 Oct

Simon WillisonKorrents 9 Oct

korrents.com
The addition of a visual preview and the ability to type or paste things makes interacting with a coding agent much more powerful.

Their words

The addition of the visual preview, plus being able to type or paste things in via the keyboard when I need to communicate something that doesn’t work vocally, makes this a much more powerful way of interacting with a coding agent.

A new feature for my blog, built using my voicesimonwillison.net

8 October

Simon Willisonx.com 8 Oct

github.com
Linuxopen source

7 October

Simon Willisonx.com 7 Oct

Simon WillisonKorrents 7 Oct

korrents.com
Anthropic's new credit scheme goes some way to overcoming the difference with OpenAI's Codex subscription deal.

AnthropicOpenAI

Their words

This new credit scheme goes at least some way to overcoming that difference.

Introducing Claude Haiku 5.5simonwillison.net
korrents.com
Wikis are tempting targets for rogue agent swarms.

Their words

Given how tempting a target wikis are for rogue agent swarms, it's not a huge surprise that Wikipedia found evidence of that activity once they went looking

simonwillison.net

6 October

Simon Willisonx.com 6 Oct

simonwillison.net

Simon WillisonKorrents 6 Oct

korrents.com
Mistral is back to being maybe about 6 months behind the frontier.

Their words

It's certainly not a Fable-class model, but it's great to see Mistral put out a model that's back to being maybe about 6 months behind the frontier.

Introducing Mistral Large 4: Le chonksimonwillison.net

The whole feed →

Beliefs

Korrents What they believe 54 beliefs — each backed by an exact quote.
Each is a — compiled by korrents.com, not by them: the one-line wordings are korrents', the quotes are theirs.
Recent
- If Anthropic's revenue figures are accurate, other labs are spending more than half a billion dollars to distill Anthropic's models.
  
  If true, and Q2 revenue was $11.5B, then that's more than half a billion dollars spent by other labs to distill Anthropic's models!
  
  @simonw on X Said 15 Oct 2026
  
  Full record — everyone who holds this →
  
  Anthropic
- Sampling profilers are very low overhead and can provide deep insight into running systems without needing to instrument them first.
  
  They’re very low overhead and can provide a deep insight into running systems, without needing to instrument those systems first.
  
  Hello Python 3.15, goodbye 3.10, and now I can use 3.11 Said 12 Oct 2026
  
  Full record — everyone who holds this →
- The addition of a visual preview and the ability to type or paste things makes interacting with a coding agent much more powerful.
  
  The addition of the visual preview, plus being able to type or paste things in via the keyboard when I need to communicate something that doesn’t work vocally, makes this a much more powerful way of interacting with a coding agent.
  
  A new feature for my blog, built using my voice Said 9 Oct 2026
  
  Full record — everyone who holds this →
Show 51 more

Wikis are tempting targets for rogue agent swarms.

Given how tempting a target wikis are for rogue agent swarms, it's not a huge surprise that Wikipedia found evidence of that activity once they went looking

OpenAI “rogue” agent activities found on Wikimedia projects Said 7 Oct 2026

Full record — everyone who holds this →

Anthropic's new credit scheme goes some way to overcoming the difference with OpenAI's Codex subscription deal.

This new credit scheme goes at least some way to overcoming that difference.

Introducing Claude Haiku 5.5 Said 7 Oct 2026

Full record — everyone who holds this →

Anthropic OpenAI

Mistral is back to being maybe about 6 months behind the frontier.

It's certainly not a Fable-class model, but it's great to see Mistral put out a model that's back to being maybe about 6 months behind the frontier.

Introducing Mistral Large 4: Le chonk Said 6 Oct 2026

Full record — everyone who holds this →

Default hard budget caps will be needed on pretty much everything

We're going to need default hard budget caps on pretty much everything

Qwen3.8 27B addition in words Said 4 Oct 2026

Full record — everyone who holds this →

Most businesses and individuals would probably prefer service errors to a surprise $10,000+ bill.

I expect that most businesses and individuals would prefer errors to a surprise $10,000+ bill.

We're going to need default hard budget caps on pretty much everything Said 3 Oct 2026

Full record — everyone who holds this →

It would be great if AI agents started biasing toward recommending providers with hard budget caps.

It would be great if agents started biasing towards recommending providers with hard budget caps

We're going to need default hard budget caps on pretty much everything Said 3 Oct 2026

Full record — everyone who holds this →

AI agents

Two new AI models, paired with their coding agent harnesses, improved from often making mistakes to being reliable enough for day-to-day use.

These two new models, when paired with their respective coding agent harnesses, improved from “often make mistakes” to “reliable enough to use on a day-to-day basis”.

2026 in LLMs (so far) Said 27 Sept 2026

Full record — everyone who holds this →

Fierce competition means the best AI model in the world gets knocked off the top spot quickly.

When you release the best model in the world, it’s going get knocked off that pedestal pretty quickly. The competition is so fierce that you won’t get a long time at the top

2026 in LLMs (so far) Said 27 Sept 2026

Full record — everyone who holds this →

The technology industry is destabilized in a way not experienced since the advent of the personal computer.

Our industry is destabilized in a way nobody’s experienced since the advent of the personal computer.

@simonw on X Said 26 Sept 2026

Full record — everyone who holds this →

Current frontier model capabilities will be instantaneous and free in a few years.

Everything we do with frontier models today, in a few years we’ll be doing instantaneously and for free.

@simonw on X Said 26 Sept 2026

Full record — everyone who holds this →

Increasing time spent with coding agents strengthens the conviction that they make software engineering even harder.

The more time I spend working with coding agents, the more convinced I am that they make software engineering even harder.

Note on 24th September 2026 Said 24 Sept 2026

Full record — everyone who holds this →

coding agents

Claude Opus 5.5's "max" thinking level may be effectively useless, since it over-thinks to breaking point even on a simple SVG prompt.

This makes me suspect that “max” is effectively useless—if it over-thinks to breaking point on a stupid SVG prompt I don’t trust it not to do the same for more interesting work.

Claude Opus 5.5, GPT-6 Sol, GPT-6 Luna, and a new price war Said 22 Sept 2026

Full record — everyone who holds this →

Anthropic

A computer scientist refusing to find anything about LLMs interesting right now is a bit like a geneticist unimpressed by a newly opened Jurassic Park.

Being a computer scientist who refuses to find anything about LLMs interesting right now is a bit like being a geneticist who refuses to find anything interesting about the recently opened Jurassic Park.

Note on 18th September 2026 Said 18 Sept 2026

Full record — everyone who holds this →

LLMs

Any software that depends on open source has a network of humans with package publishing rights who are potential attack vectors.

Any piece of software that depends on open source (which is almost every piece of software) has a network of human beings who are potential attack vectors - everyone with publishing rights to any of the packages in the dependency network for that software.

Be alert: targeted attacks on prominent Rustaceans Said 17 Sept 2026

Full record — everyone who holds this →

open source

Dependency cooldowns, waiting a few days before upgrading to new package releases, are probably the best current defense against supply chain attacks.

I guess our best defense right now is dependency cooldowns - giving new package releases a few days before upgrading to them, in the hope that supply chain attacks like this will be spotted by someone else.

Be alert: targeted attacks on prominent Rustaceans Said 17 Sept 2026

Full record — everyone who holds this →

Text that uses turns of phrase suggested by an LLM has a weird smell.

The rule to never use a turn of phrase suggested by an LLM feels good to me. The text has that weird smell to it, and it's also a good principle to help stay disciplined.

How To Write With An LLM Said 17 Sept 2026

Full record — everyone who holds this →

LLMs

If OpenAI really never told RubyGems of its agents' attack, either it failed to find it in its logs after later incidents or chose silence, and both are bad.

If that’s true there are two options: After the Hugging Face and Wiki attacks OpenAI were still unable to review their previous logs and determine that they had previously attacked RubyGems. They knew about the attack on RubyGems and made the decision not to reach out to the RubyGems team about it. Both of these are bad!

OpenAI agents attacked RubyGems back in May Said 12 Sept 2026

Full record — everyone who holds this →

OpenAI HuggingFace

Any LLM system that uses compaction needs to preserve the pre-compacted text and make it available via agent tool calls.

I think any LLM system that uses compaction needs to both preserve the pre-compacted text and make that text available via agent tool calls, to protect against this kind of problem.

Generating running routes with GPT-6 Astra and ChatGPT Work Said 12 Sept 2026

Full record — everyone who holds this →

LLMs

Not showing the code and exact details of what the agent ran in the ChatGPT UI is a lack of transparency and an anti-feature.

Frustratingly, the actual code it ran and exact details of what it did weren’t visible to me in the ChatGPT UI. I see this lack of transparency is an anti-feature.

Generating running routes with GPT-6 Astra and ChatGPT Work Said 12 Sept 2026

Full record — everyone who holds this →

OpenAI design

Just knowing that there is an unpublished solution to a problem might trigger millions of dollars in LLM spending to get there first.

Just knowing that there is an unpublished solution to a problem might trigger millions of dollars in LLM spending to get there first.

On the Navier–Stokes Millennium Prize Problem Said 8 Sept 2026

Full record — everyone who holds this →

LLMs

Web applications are generally expected to treat GET requests as safe and not use them to change server-side data, though not all software follows that convention.

It looks to me like OpenAI’s sandbox for this agent suffered from the (quite naïve) assumption that GET requests cannot be used to update data. That’s certainly how the web is supposed to work, but clearly there are applications that don’t hold to that contract.

OpenAI's rogue agents were caught communicating via public wikis Said 4 Sept 2026

Full record — everyone who holds this →

OpenAI

When Reuters uses the phrase 'people familiar with the matter', it means it has anonymous insider sources its reporters and editors find credible.

I’ve written about the people familiar with the matter pattern before—it means Reuters have anonymous insider sources that their reporters (and editors) find credible.

OpenAI's rogue agents were caught communicating via public wikis Said 4 Sept 2026

Full record — everyone who holds this →

journalism

Claude shouldn't be trusted to summarize its own system prompts when material in them might impact its opinions.

I don’t trust Claude to summarize its own system prompts when there’s a risk that material from its system prompt might impact its opinions.

Claude's new system prompt really doesn't want to reproduce song lyrics Said 2 Sept 2026

Full record — everyone who holds this →

Anthropic

The pelican-drawing benchmark's link to how good models are at other tasks didn't seem to hold as strongly as in 2025.

its connection to how good the models were at other tasks didn’t seem to hold as strongly as it did back in 2025

Claude Fable 5.1 made me a really nice animated pelican Said 1 Sept 2026

Full record — everyone who holds this →

benchmarks

Claude Haiku hallucinates badly enough that its use inside Claude Code's WebFetch tool is a hallucination risk on every fetched URL.

Claude Haiku is my current least favorite model - it hallucinates wildly, and is out-performed now by other similarly priced models like GPT-5.6-Luna Even worse: it seems to still be used by the Claude Code WebFetch tool, which means hallucination risk any time you fetch a URL!

@simonw on X Said 10 Aug 2026

Full record — everyone who holds this →

Anthropic OpenAI

Auto-mode does not yet convincingly fix prompt-injection risk for coding agents.

I REALLY want to believe that this fixes prompt injection risks for coding agents, but I'm just not there yet

@simonw on X Said 8 Aug 2026

Full record — everyone who holds this →

coding agents prompt injection

Giving consumer-facing models and API models separate brand names is needlessly confusing.

I really do think that having separate brand names for the consumer-facing models and the API models is needlessly confusing

@simonw on X Said 7 Aug 2026

Full record — everyone who holds this →

Claude has surprisingly good design taste for SwiftUI applications.

Claude has surprisingly good design taste when it comes to SwiftUI applications.

Vibe coding SwiftUI apps is a lot of fun Said 27 Mar 2026

Full record — everyone who holds this →

Anthropic design taste

Test-driven development matters more with coding agents than it ever did with people, because its real effect is stopping the agent writing more code than the task needs.

the key thing about TDD is that it means that the agents won't write more than they need to.

Simon Willison: Engineering practices that make coding agents work - The Pragmatic Summit Said 19 Mar 2026

Full record — everyone who holds this →

coding agents

Writing code with a coding agent and no tests at all is indefensible, because the old objection to testing — that it is extra work you then have to maintain — no longer applies.

I think I see people who are writing code with coding agents and they're not writing any tests at all. That's a terrible idea.

Simon Willison: Engineering practices that make coding agents work - The Pragmatic Summit Said 19 Mar 2026

Full record — everyone who holds this →

coding agents

A passing test suite is not evidence that the software works, so an agent must also be made to start the thing and exercise it the way a person would.

just cuz the test suite passes doesn't mean that the web server will boot. You know, there's there's always a chance that when you actually try in the real world, something's not going to work.

Simon Willison: Engineering practices that make coding agents work - The Pragmatic Summit Said 19 Mar 2026

Full record — everyone who holds this →

software quality

A language-agnostic conformance suite is the most powerful thing you can hand a coding agent, because the whole instruction becomes: write code until these tests pass.

And what you can do if you've got one of these conformance suites is you can give it to a a good agent and say, "Write code until this test suite passes." And it kind of will.

Simon Willison: Engineering practices that make coding agents work - The Pragmatic Summit Said 19 Mar 2026

Full record — everyone who holds this →

software quality

Whether agent-written code needs to be good depends entirely on how long it will live: a throwaway single-page tool can be spaghetti, anything maintained cannot.

Anything that you're maintaining over the longer term, the code quality does start really really mattering.

Simon Willison: Engineering practices that make coding agents work - The Pragmatic Summit Said 19 Mar 2026

Full record — everyone who holds this →

software quality

Keeping a codebase clean now pays off directly rather than aesthetically, because a coding agent will copy the patterns already in it almost exactly.

one of the magic tricks about these things is that they they're incredibly consistent. If you've got a code base with a bunch of patterns in, they will follow those patterns almost to a T.

Simon Willison: Engineering practices that make coding agents work - The Pragmatic Summit Said 19 Mar 2026

Full record — everyone who holds this →

A team can stop reading the code it ships, but only by first working out how its agents are to prove that what they wrote actually works.

it turns out you can do this if you think really hard about okay, how do I have agents prove to me that the stuff they've written works?

Simon Willison: Engineering practices that make coding agents work - The Pragmatic Summit Said 19 Mar 2026

Full record — everyone who holds this →

Trusting model-written code you have not read became defensible only with Opus 4.5, and only for classes of problem you have already watched a model handle.

I think Opus 4.5 was the first one that earned my trust. Like I'm very confident now that for classes of problems that I've seen it tackle before, it's not going to do anything stupid.

Simon Willison: Engineering practices that make coding agents work - The Pragmatic Summit Said 19 Mar 2026

Full record — everyone who holds this →

Language models are gullible by design — they do what they are told and believe nearly anything said to them — and that, not a bug, is what makes software built on them attackable.

The problem with language models is they're incredibly gullible by design. Like, language models do exactly what you tell them to do, and they will believe almost anything that you say to them.

Simon Willison: Engineering practices that make coding agents work - The Pragmatic Summit Said 19 Mar 2026

Full record — everyone who holds this →

LLMs design

Prompt injection has no equivalent of parameterized queries: there is no reliable way to tell a language model which text is data and which is instructions.

Problem is, you can solve SQL injection by parameterizing your queries. You can't do that with LLMs.

Simon Willison: Engineering practices that make coding agents work - The Pragmatic Summit Said 19 Mar 2026

Full record — everyone who holds this →

LLMs prompt injection

The only guaranteed defence against a prompt-injection attack is to remove one of its three legs — private data, exposure to untrusted content, or a way to send data out.

I call it the lethal trifecta because the only guaranteed solution is to cut off one of the legs.

Simon Willison: Engineering practices that make coding agents work - The Pragmatic Summit Said 19 Mar 2026

Full record — everyone who holds this →

Sandboxing is the single most important protection for a coding agent, because it is the only one that bounds the damage when the agent is successfully tricked.

So, I think the most important thing is sandboxing. You want your coding agent running in an environment where if something goes completely wrong, if somebody gets malicious instructions to it, that the damage is greatly limited.

Simon Willison: Engineering practices that make coding agents work - The Pragmatic Summit Said 19 Mar 2026

Full record — everyone who holds this →

The frontier labs optimise their models for code above everything else because coding is where the money is: developers will pay $200 a month once a model is good enough.

OpenAI and Anthropic have both realized that code is the most important thing to optimize the models for, cuz that's where the money is. Like coders will spend $200 a month on a plan if it's good enough, it turns out.

Simon Willison: Engineering practices that make coding agents work - The Pragmatic Summit Said 19 Mar 2026

Full record — everyone who holds this →

Anthropic OpenAI

The interesting question is not what the next model will be able to do, but what the models already released can do that nobody has worked out yet.

I think the most interesting question is what can the models we have do right now? And so the only thing I care about today is what can Claude Opus 4.6 do that we haven't figured out yet. And I think it will take us 6 months to even start exploring the boundaries of that.

Simon Willison: Engineering practices that make coding agents work - The Pragmatic Summit Said 19 Mar 2026

Full record — everyone who holds this →

Anthropic

Working with coding agents is exhausting rather than deskilling, and that exhaustion — not the technology — is what limits how many projects one engineer can run at once.

I think that might be what saves us. I think the fact that no, you can't have one engineer and have him do a thousand projects because after three hours of that he's going to literally pass out in a corner.

Simon Willison: Engineering practices that make coding agents work - The Pragmatic Summit Said 19 Mar 2026

Full record — everyone who holds this →

coding agents

The overhead of learning a new programming language no longer justifies sticking to the two you know: start writing in a third now, without learning it first.

if you've always stuck to two programming languages because of the overhead of learning a third, go and learn a third right now. And don't learn it, just start writing code in it.

Simon Willison: Engineering practices that make coding agents work - The Pragmatic Summit Said 19 Mar 2026

Full record — everyone who holds this →

Demand for open-source component libraries is collapsing, because a model will write exactly the widget you want instead of one you have to customise.

Why would I use a date picker library where I'd have to customize it when I could have Claude write me the exact date picker that I want.

Simon Willison: Engineering practices that make coding agents work - The Pragmatic Summit Said 19 Mar 2026

Full record — everyone who holds this →

Anthropic

Open-source projects are now so flooded with junk contributions that maintainers are asking GitHub for something it has never offered: a way to switch pull requests off.

projects are flooded with junk contributions at the moment to the point that people are trying to convince GitHub to disable pull requests, which is something GitHub have never done.

Simon Willison: Engineering practices that make coding agents work - The Pragmatic Summit Said 19 Mar 2026

Full record — everyone who holds this →

MCP may turn out to be a one-year wonder, because coding agents that can run arbitrary shell commands make Bash the best general-purpose tool interface.

The reason I think MCP may be a one-year wonder is the stratospheric growth of coding agents. It appears that the best possible tool for any situation is Bash—if your agent can run arbitrary shell commands, it can do anything that can be done by typing commands into a terminal.

2025: The year in LLMs Said 31 Dec 2025

Full record — everyone who holds this →

MCP coding agents

Running coding agents without approval prompts is a normalization-of-deviance risk: the fact that it has not caused a disaster yet is itself the problem.

I run in YOLO mode all the time, despite being deeply aware of the risks involved. It hasn’t burned me yet... ... and that’s the problem.

2025: The year in LLMs Said 31 Dec 2025

Full record — everyone who holds this →

coding agents

Good taste in testing code is one of the skills that sets a senior engineer apart.

Developing good taste in testing code is another of those skills that differentiates a senior engineer.

Your job is to deliver code you have proven to work Said 18 Dec 2025

Full record — everyone who holds this →

taste

Claude tends to have good default taste in web page design.

I tend to find Claude has good default taste when it comes to web page design

Here’s how I use LLMs to help me write code Said 11 Mar 2025

Full record — everyone who holds this →

Anthropic design taste

Concerns about the energy cost of an individual LLM prompt are no longer credible, though the environmental impact of the AI datacenter buildout remains a real worry.

There’s still plenty to worry about with respect to the environmental impact of the great AI datacenter buildout, but a lot of the concerns over the energy cost of individual prompts are no longer credible.

Things we learned about LLMs in 2024 Said 31 Dec 2024

Full record — everyone who holds this →

LLMs data centers energy
Beliefs others hold too
- Concerns about the energy cost of an individual LLM prompt are no longer credible, though the environmental impact of the AI datacenter buildout remains a real worry. 2 hold this
  
  There’s still plenty to worry about with respect to the environmental impact of the great AI datacenter buildout, but a lot of the concerns over the energy cost of individual prompts are no longer credible.
  
  Things we learned about LLMs in 2024 Said 31 Dec 2024
  
  Full record — everyone who holds this →
  
  LLMs data centers energy
All 54 beliefs →

What is a korrent?

A korrent is a belief a person has stated in their own words: one sentence stating the claim, backed by a quote and a source, kept at korrents.com.

Under a name here, the quoted block is what they actually said. The korrent beneath it is the claim those words support, in korrents' wording — tap it to see the record, its source, and who else holds it.

Nobody here wrote their own korrents. They are compiled from public statements, and a person can change their mind, which is recorded too.

Recommends rcmnd

rcmnd What they recommend 24 things, 26 mentions — each backed by an exact quote.
Compiled by rcmnd.app, not by them: the quotes are theirs, and each links to the post it was copied from.
Recent
- Uses datasette-mcp 24 Sept 2026
  
  I just had a voice conversation with the Datasette backup of my blog via datasette-mcp, which you can now talk to using the ChatGPT iPhone app
  
  x.com 24 Sept 2026
  
  What others say about datasette-mcp →
- Loved GPT-6 Luna 22 Sept 2026
  
  GPT-6 Luna is half the price of 5.6 Luna, which was already an astonishingly cheap model given how capable it is Luna is my favorite model for building product features thanks to its cost (and speed)
  
  x.com 22 Sept 2026
  
  What others say about GPT-6 Luna →
- Liked GPT-6 Astra 13 Sept 2026
  
  This is pretty neat: ChatGPT Work and GPT-6 Astra (I used "Max") can take an address and produce a 5K/10K circular running route starting from that address, using OSM data
  
  x.com 13 Sept 2026
  
  What others say about GPT-6 Astra →
Show 23 more

Liked ChatGPT Work 13 Sept 2026

This is pretty neat: ChatGPT Work and GPT-6 Astra (I used "Max") can take an address and produce a 5K/10K circular running route starting from that address, using OSM data

x.com 13 Sept 2026

What others say about ChatGPT Work →

Uses GPT-5.6 Sol 11 Sept 2026

We ran an extensive audit using Claude Fable 5.1, GPT-5.6 Sol and GPT-6 Astra, then fixed a range of different bugs.

x.com 11 Sept 2026

What others say about GPT-5.6 Sol →

Uses Claude Fable 5.1 11 Sept 2026

We ran an extensive audit using Claude Fable 5.1, GPT-5.6 Sol and GPT-6 Astra, then fixed a range of different bugs.

x.com 11 Sept 2026

What others say about Claude Fable 5.1 →

Recommends Datasette 11 Sept 2026

If you're running a Datasette instance on a public website you should upgrade.

x.com 11 Sept 2026

their own

What others say about Datasette →

Uses GPT-6 Astra 10 Sept 2026

Here's a Blender model of a Pluribus themed Fabergé egg I had GPT-6 Astra build - I generated the concept image using ChatGPT Images 2.5

x.com 10 Sept 2026

What others say about GPT-6 Astra →

Uses Blender 10 Sept 2026

Here's a Blender model of a Pluribus themed Fabergé egg I had GPT-6 Astra build - I generated the concept image using ChatGPT Images 2.5, then pasted the image into Codex and told Astra to turn that into a Blender file

x.com 10 Sept 2026

What others say about Blender →

Uses Codex 10 Sept 2026

I generated the concept image using ChatGPT Images 2.5, then pasted the image into Codex and told Astra to turn that into a Blender file

x.com 10 Sept 2026

What others say about Codex →

Uses ChatGPT Images 2.5 10 Sept 2026

I generated the concept image using ChatGPT Images 2.5, then pasted the image into Codex and told Astra to turn that into a Blender file

x.com 10 Sept 2026

What others say about ChatGPT Images 2.5 →

Uses Blender 5 Sept 2026

New TIL on using Blender with coding agents on macOS: https://til.simonwillison.net/llms/blender-coding-agents-macos

x.com 5 Sept 2026

What others say about Blender →

Loved Django Migrations 7 Jul 2026

My favorite implementation of this pattern remains Django’s Migrations

simonwillison.net 7 Jul 2026

What others say about Django Migrations →

Loved Starlette 22 Mar 2026

Kim Christie started working on Starlette in 2018 and it quickly became my favorite out of the new breed of Python ASGI frameworks.

simonwillison.net 22 Mar 2026

What others say about Starlette →

Loved The Normalization of Deviance in AI 31 Dec 2025

One of my favourite pieces on LLM security this year is The Normalization of Deviance in AI by security researcher Johann Rehberger.

simonwillison.net 31 Dec 2025

What others say about The Normalization of Deviance in AI →

Uses shot-scraper 26 Nov 2025

I’ve got a GitHub actions thing that runs a piece of software I wrote called shot-scraper that runs Playwright, that loads up a browser in GitHub actions to scrape that webpage and turn the results into JSON, which then get turned into an atom feed, which I subscribe to in NetNewsWire.

simonwillison.net 26 Nov 2025

their own

What others say about shot-scraper →

Uses NetNewsWire 26 Nov 2025

I’ve got a GitHub actions thing that runs a piece of software I wrote called shot-scraper that runs Playwright, that loads up a browser in GitHub actions to scrape that webpage and turn the results into JSON, which then get turned into an atom feed, which I subscribe to in NetNewsWire.

simonwillison.net 26 Nov 2025

What others say about NetNewsWire →

Loved Mark Steel’s in Town 26 Nov 2025

Every episode he goes to a small town in England and he does a comedy set in a local venue about the history of the town. And so he does very deep research... I love that sort of like hyperlocal, like comedy, that sort of British culture thing.

simonwillison.net 26 Nov 2025

What others say about Mark Steel’s in Town →

Loved Design Patterns for Securing LLM Agents against Prompt Injections 9 Aug 2025

One of my favorite papers about prompt injection is Design Patterns for Securing LLM Agents against Prompt Injections

simonwillison.net 9 Aug 2025

What others say about Design Patterns for Securing LLM Agents against Prompt Injections →

Loved Mistral Small 3 6 Jun 2025

My favorite model release from January was another local model, Mistral Small 3

simonwillison.net 6 Jun 2025

What others say about Mistral Small 3 →

Loved Anthropic prompt engineering documentation 25 May 2025

I still think Anthropic have the best prompting documentation of any LLM provider.

simonwillison.net 25 May 2025

What others say about Anthropic prompt engineering documentation →

Loved Claude 3.7 Sonnet 2 Mar 2025

As a Python and JavaScript programmer my favorite models right now are Claude 3.7 Sonnet with thinking turned on, OpenAI’s o3-mini-high and GPT-4o with Code Interpreter (for Python).

simonwillison.net 2 Mar 2025

What others say about Claude 3.7 Sonnet →

Recommends Is AI progress slowing down? 31 Dec 2024

To understand more about inference scaling I recommend Is AI progress slowing down?

simonwillison.net 31 Dec 2024

What others say about Is AI progress slowing down? →

Loved Google Gemini 29 Oct 2024

The Google Gemini series are my favourite multi-modal models right now due to the size and breadth of content they support.

simonwillison.net 29 Oct 2024

What others say about Google Gemini →

Loved ChatGPT Code Interpreter 13 Jul 2024

ChatGPT Code Interpreter has been one of my favourite AI tools for over a year.

simonwillison.net 13 Jul 2024

What others say about ChatGPT Code Interpreter →

Loved Your AI Product Needs Evals 21 Jun 2024

Your AI Product Needs Evals by Hamel Husain remains my favourite piece of writing on how to go about putting these together.

simonwillison.net 21 Jun 2024

What others say about Your AI Product Needs Evals →
All 26 recommendations →

Where they publish

Blog simonwillison.net Their blog: near-daily notes on LLMs, tooling and what they just tried.
One of the most-cited running records of what the current models can actually do — new releases tested the day they land, plus the long-running Datasette and llm project notes. Has an Atom feed.

Open simonwillison.net
Recent
- Chirp Pong 15 Oct 2026 Tool: Chirp Pong Another game experiment - I wanted to see if it was feasible to have a two player game on two separate phones running in static HTML where the networking code uses chirps sent via the microphone. It see…
- Anthropic tells investors quarterly revenue up nearly 50% to $16.8bn ahead of IPO 15 Oct 2026 The phrase "told investors" appears four times in this FT article, which reports on Anthropic's day of hosting investors on Wednesday. Lots of details in here from "people with knowledge of the matter". Key things I lea…
- Tree Appreciator 14 Oct 2026 Tool: Tree Appreciator Another game experiment with Claude. This one used the default Claude iPhone app (chat mode, not Claude Code mode), where I prompted Claude Opus 5.5 Max with: Build a three js game called "Tree ap…
Show 17 more

AI is mass decompiling games. That's both awesome and scary. 14 Oct 2026 Indie game developer Popcar talks about the recent wave of AI-powered decompilation of existing games, enabling weird mashups between existing games, and producing a boom in game preservation projects to get old games w…
Redesigning source control at Anthropic 13 Oct 2026 Interesting piece from Anthropic describing why and how they built their own custom alternative to Git, based on the dramatic increase in rate of changes to their code after the past six months. In May 2026, Anthropic a…
llm-gemini 0.35 13 Oct 2026 Release: llm-gemini 0.35 I added support to llm-gemini for the new Gemini 4 Argon model, made available today via the Gemini API and for customers of their $100/month subscription plan. Here are the low, medium and high…
Quoting Geoffrey Huntley 13 Oct 2026 If the agent isn't using my diagnostics library, I could introspect and update AGENTS.md, but there's still a delta, and it'll drift. Instead, I write a pre-commit hook that checks the git diff for direct calls to obser…
web-edit 0.1a1 13 Oct 2026 Release: web-edit 0.1a1 Includes performance optimizations written by Claude Opus 5.5 after I fed it this Python 3.15 profiling report generated by GPT-6 Astra. Tags: profiling, projects
llm-meta-ai 0.2 12 Oct 2026 Release: llm-meta-ai 0.2 Meta AI release Muse Wonder (and Muse Wonder Full), their new "most capable models". These look like they may be Meta's first Fable class models - Artificial Analysis scores Muse Wonder (Max) as…
Hello Python 3.15, goodbye 3.10, and now I can use 3.11 12 Oct 2026 Python 3.15 is here! Here's the essential What's new in Python 3.15 guide, this year curated by 3.14 and 3.15 release manager Hugo van Kemenade. As someone who builds open source software that works in every supported P…
Profiling web-edit with Python 3.15: flame graphs and source heatmaps 12 Oct 2026 Research: Profiling web-edit with Python 3.15: flame graphs and source heatmaps I put the new Python 3.15 profiler through its paces by having GPT-6 Astra Extra High in Codex run it against my new web-edit app: I need a…
web-edit 0.1a0 12 Oct 2026 Release: web-edit 0.1a0 I was working on a project from my phone using Codex Remote when I found myself wanting to edit a credentials file without round-tripping those credentials through the LLM itself. So... I had Cod…
datasette-ports 0.4 12 Oct 2026 Release: datasette-ports 0.4 My plugin for detecting and listing all Datasette instances on the current machine now works with the latest version, and displays the detected PID as well. Knowing the PID makes it easier t…
Dwarf Fortress uses version control now 11 Oct 2026 I remember hearing a while ago that Dwarf Fortress didn't use version control. That fact has been lodged in my head, because given its inherent complexity I have trouble imagining a codebase that would benefit from vers…
Python 3.15.0 added to actions/python-versions 10 Oct 2026 Bit of a niche link, but I've been waiting for this for a couple of days - I even told ChatGPT to check for it: Clone https://github.com/actions/python-versions and git pull once an hour until they add the stable 3.15 -…
Quoting The New York Times 10 Oct 2026 Anthropic detailed the activity of its A.I. agents in a blog post on Friday, without naming the targeted websites. But two sources with knowledge of the incidents said Anthropic’s A.I. agents had submitted 20 visa appli…
Deno is joining Cloudflare 9 Oct 2026 The Deno team released the first version of celld back in August - their open source implementation of the Durable Objects pattern from Cloudflare Workers. Today, Cloudflare are acquiring Deno outright, with the goal of…
Quoting Matthew Green 9 Oct 2026 Everyone is very concerned about being respectable, so I’m going to be the goofball who raises worst-case possibilities. I think there is a 1% chance we live in Minicrypt, and a 15% chance we functionally lose confidenc…
A new feature for my blog, built using my voice 9 Oct 2026 I shipped a new feature for my blog today: the Newsletters page, which offers an index of all of the newsletters I've sent out, both my free weekly Substack and my monthly sponsors-only updates. I built the feature almo…
ttok 1.0 9 Oct 2026 Release: ttok 1.0 I released ttok 0.4, ran uv tool upgrade ttok, piped a file into the new version... and realized that it was defaulting to the GPT-4 tokenizer when it should very clearly default to GPT-5/GPT-6 instead…
Link verified 1 Sept 2026. Recent items update automatically from the channel.
x.com @simonw Short-form, and where new posts get announced.
Quick reactions to model releases and tooling, and the announcement channel for the blog. Most items here reappear on the blog with more detail.

Open x.com
Recent
Show 17 more

What's the best open weight Mixture-of-Experts LLM for coding that fits in less than 60GB of RAM? I think MoE might be necessary to get reasonably interactive speeds on the hardware I have access to - I want something faster than 12 tokens/second 9 Oct 2026
I built a new feature for my blog entirely by voice with Codex Desktop, while I was cooking dinner 9 Oct 2026 simonwillison.net
This "Super Intelligence" thing is just so, so, so stupid I can't even 8 Oct 2026
New open source cross-platform (Windows, macOS, Linux) sandboxing library from Microsoft - looks very promising, uses processcontainer/bubblewrap/seatbelt under the hood 8 Oct 2026 github.com
Pelicans and a bunch of notes on the pricing of the new Claude Haiku 5.5 7 Oct 2026 simonwillison.net
I released an LLM plugin for sending prompts to OpenAI's new Jev-clone decision model that's based on GPT-6 Luna 7 Oct 2026 simonwillison.net
Anyone know when the models started being able to compose competent music? I had Opus 5.5 "write some computer game music for me" and the results were much better than I had expected 6 Oct 2026 simonwillison.net
Mistral can pelican now! 6 Oct 2026
Just published this post about how we’re going to need default hard budget caps on pretty much everything 3 Oct 2026 simonwillison.net
Now that we've had a few days with it, how are people differentiating between Dot and regular ChatGPT? I'm having trouble deciding when I should prompt Dot vs using ChatGPT - my Dot seems to afford a single conversation, but I like controlling my context across multiple threads 3 Oct 2026
I'm at OpenAI's DevDay event in San Francisco today - as I have for the past three DevDay events, I'm running a live blog where I'll be posting updates during the keynote, which starts in five minutes 29 Sept 2026 simonwillison.net
The most important thing about Sonnet 5.5 is that it's now the model that powers the free tier on https://t.co/f1IGIsU4Hc - so all of this stuff can be done by free users ChatGPT's free tier is still GPT-5.6 Luna, which is a lot less capable 28 Sept 2026
I've published detailed notes and an annotated transcript to accompany the video of the keynote I gave at @WeAreDevs World Congress North America in San Jose on Friday - here's my rundown of everything that's happened with LLMs and agents in 2026 so far 28 Sept 2026 simonwillison.net
I included a few references to this year's record-breaking Kākāpō breeding season in a talk I gave yesterday, and since Claude Opus 5.5 is surprisingly capable at pixel art animation I had it create this celebratory video for my closing slide 27 Sept 2026
"The last year in the technology industry has felt like 100 years all happening at once. Our industry is destabilized in a way nobody’s experienced since the advent of the personal computer. Every limit AI runs up against collapses within a month. Everything we do with frontier models today, in a few years we’ll be doing instantaneously and for free." 26 Sept 2026
I'm on stage for the keynote in ten minutes time! 26 Sept 2026
The more time I spend working with coding agents, the more convinced I am that they make software engineering even harder We can do amazing things with them, but unlocking their full potential requires extraordinary discipline and knowledge 25 Sept 2026
Link verified 1 Sept 2026. Recent items update automatically from the channel.
GitHub simonw Datasette, llm, and several hundred smaller tools.
Unusually busy: Datasette and the llm CLI alongside hundreds of small utilities, many written to answer a single question. The repo list reads as a record of what they were working on when.

Open github.com
Recent
Show 17 more

scrape-fema-shelters 16 Oct 2026
usgs-scraper 16 Oct 2026
pge-outages — Tracking PG&E power outages 16 Oct 2026
scrape-github-actions-package-versions — Git scraper recording the package versions installed on the defaul GitHub Actions ubuntu-latest worker 16 Oct 2026
scrape-hacker-news-by-domain — Scrape HN to track links from specific domains 16 Oct 2026
package-stats — Download statistics for my PyPI packages 16 Oct 2026
scrape-roads-dot-ca-gov — Scrape highway information from https://roads.dot.ca.gov/ 16 Oct 2026
scrape-fediverse — Git scrapers for scraping the fediverse 16 Oct 2026
simonw — https://simonwillison.net/2020/Jul/10/self-updating-profile-readme/ 16 Oct 2026
ollama-models-atom-feed — https://ollama.com/search?o=newest 16 Oct 2026
usgs-demo — https://earthquake.usgs.gov/earthquakes/feed/v1.0/summary/all_hour.geojson 16 Oct 2026
claude-system-prompts — Tracking changes to Claude's published system prompts 16 Oct 2026
scrape-faa-releasable-aircraft 16 Oct 2026
research — Research projects 16 Oct 2026
inaturalist-clumps — Runs inaturalist-clumper 16 Oct 2026
tools — Assorted useful tools, almost entirely generated using LLMs 15 Oct 2026
simonwillisonblog-backup — Backups of the database for simonwillison.net 15 Oct 2026
Link verified 1 Sept 2026. Recent items update automatically from the channel.

About the English under a post

Some people here publish in a language other than English. Where they do, this site shows a machine translation beneath the post, in this typeface — the site's own, not theirs.

The post itself is never changed, moved or hidden: what is set in the serif above is exactly what the person published, and it is what to quote them on. A translation can be wrong in ways that matter, especially about tone.

Only the post's own words are translated. A quoted post, a linked article and a belief on korrents.com are left in their original language.