Follow buttons here act as you, not as @admin.
| Label | Kind | State | Action |
|---|---|---|---|
| @ricardo | local | — | |
| @rmdes | local | — | |
| @paul | local | — | |
| news.rss.chat | source | — | — |
| scripting.com | source | — | — |
| euwatch.micro.blog | source | — | — |
| www.techdirt.com | source | — | — |
| simonwillison.net | source | — | — |
| www.manton.org | source | — | — |
| feeds.kottke.org | source | — | — |
| rss.beehiiv.com | source | — | — |
| www.404media.co | source | — | — |
| brilliantmaps.com | source | — | — |
| shellsharks.com | source | — | — |
| www.platformer.news | source | — | — |
| podstandards.org | source | — | — |
| zacharykai.net | source | — | — |
Google Maps Changes Lake Ontario to Lake America. Totally pathetic and stupid. Maps are wayfinding tools and this toadying political change will help no one find their way.
OpenAI announced ChatGPT Work on July 9th, and have been furiously iterating on it ever since. It is an extraordinarily confusing and very powerful product. Here's what I've figured out about it so far.
The more interesting version of ChatGPT Work is the one that runs in the cloud. This can be accessed via chatgpt.com or through the ChatGPT mobile apps. Let's call it Work Cloud.
If you install the ChatGPT desktop app - the app that used to be called Codex - you gain access to a thing called ChatGPT Work that can access files and run programs directly on your computer. Let's call that one Work Local. This one feels more like regular Codex re-skinned to be less intimidating to non-software-developers.
(Update: Work Cloud is also available from the ChatGPT desktop app, via a Where should this chat run? dropdown.)
For the rest of this article I'm going to talk exclusively about Work Cloud.
Right now, ChatGPT Work (in both flavors) is available only to $20/month and up subscribers. Free users and $8/month Go users do not have access.
The interface for accessing Work is a tab selector, which presents it as an alternative to Chat:

The obvious question is when should I use Chat, and when should I use Work?
OpenAI's official answer to that question is:
Use Chat when you want an answer, explanation, brainstorm, or short draft. Use ChatGPT Work when you want ChatGPT to complete a task with a clear outcome, such as a brief, deck, analysis, recurring update, workflow, or file you can review and use.
I find that almost entirely useless, because I've been using regular ChatGPT Chat for all of those task categories for years!
The better question then is what features does Work have that are missing from Chat?
After extensive experimentation I think I've mostly figured that out:
In Work, you get the option to pick GPT-5.6 Sol, Luna, or Terra, each with Light, Medium, High, Extra High, Max, or Ultra reasoning levels. You can also pick GPT-5.5 at Light, Medium, High, or Extra High.
These look to be the same models that are available through the OpenAI API.
Chat offers a different selection: 5.6 Instant, Medium, High, Extra High, and Pro (actually Extra High and Pro are only available for $100/month+ subscribers - $20/month subscribers cap out at High). It doesn't explain if those are Luna or Terra or Sol (I'm assuming Sol?). 5.6 Pro appears to be exclusive to Chat, with no equivalent in Work.
My current understanding from using Codex is that Ultra is a special mode that more eagerly delegates to sub-agents.
I believe ChatGPT Work sessions are billed against your Codex allowance, while ChatGPT Chat Sessions get their own, separate allowance. This may help explain the model availability differences.
As a long-time fan of the Code Interpreter pattern - pioneered by OpenAI in 2023 - this is by far the most exciting feature of ChatGPT Work (Cloud) for me.
The code execution environment can now talk to the rest of the internet!
ChatGPT Chat can't do this - if you ask it to install additional software packages or interact with websites or APIs that access will be blocked by the container proxy.
(Weirdly, back in January it grew the ability to install packages, but that doesn't seem to work any more. I wish they had better changelogs!)
Claude's equivalent container has allowed restricted internet access since it launched last September. Claude can install packages from PYPI and NPM and clone repositories from GitHub. But that is about it: the allowlist of domains is very short.
ChatGPT Work allows a whole lot more than that. It can be configured with a specific list of allowed domains, but the default appears to be open to all.
This makes Work an incredibly useful tool. You can have it clone GitHub repositories, install their dependencies, then use them to interact with the rest of the web!
Another killer feature of ChatGPT Work is the browser tool. ChatGPT Work can launch a full Chrome instance, load websites, fill out forms, and take screenshots.

If a site requires sign in the browser can prompt you to take over and enter both passwords and 2FA codes, without round-tripping those credentials through the model itself.
It can even run JavaScript against the DOM of loaded pages. I prompted:
Load simonwillison.net in your browser and extract the headings using JavaScript
ChatGPT Work fired up a browser instance and ran the code:
await tab.playwright.evaluate(() => { return Array.from(document.querySelectorAll("h1,h2,h3,h4,h5,h6"), heading => ({ level: heading.tagName.toLowerCase(), text: heading.innerText.trim().replace(/\s+/g, " "), id: heading.id || null })); });
This feels a lot like my shot-scraper javascript tool, only now I can access it on my phone!
ChatGPT Chat gets a fresh filesystem for each chat session. These cannot be accessed from any other session.
In ChatGPT Work each session gets its own scratch folder - named something like /workspace/scratch/e00a0a017944 - but each of those are persisted across sessions, so you can access files from previous chats. I have 171 folders in /workspace/scratch right now!
As far as I can tell that /workspace volume is mounted to all Work sessions that are currently running - file edits from one can be instantly seen by the others. They don't seem to share the same process space though, and localhost servers running in one can't be accessed from another.
ChatGPT Work has the ability to build and deploy entire websites, using Cloudflare Workers. These can have HTML and JavaScript and can run server-side features too, including stateful features on top of Cloudflare D1 and R2.
Here's a simple site I built with this feature:
london-pelicans-in-her-piety.simonw.chatgpt.site

My prompt was:
Figure out all of the places in London with a pelican in her piety, then turn that into a JSON file and build a ChatGPT sites site about them
(A pelican in her piety is a fascinating piece of medieval Christian imagery - once you know about them you'll find them all over the place.)
These sites default to being private to the user that created them, but you can make them public and (on team plans) share them with other specific individuals.
There's not much to say about this one. ChatGPT Chat can't run sub-agents. ChatGPT Work can. This is very much a power-user feature: if you are running a complex project that can benefit from multiple parallel agents working together, Work can do that.
Another feature that seems to have migrated from regular ChatGPT to ChatGPT Work at some point. You can prompt ChatGPT Work like this:
run a search to see if Waymo have announced a launch date for Half Moon Bay every day at 8am
This will schedule a prompt to run on that frequency. These prompts can decide that nothing interesting has happened, or they can decide to notify you of some new information.
Update: Actually this seems to work in ChatGPT Chat as well.
It's still worth noting here though, as it can be used in conjunction with other ChatGPT Work exclusive features. You can set a scheduled task to update a ChatGPT Site on an hourly basis, for example.
An open question for me right now is how safe all of this stuff is.
My lethal trifecta model warns about the risks inherent in any agent system that combines access to private data with exposure to untrusted content and a way to communicate stolen information back to an attacker.
ChatGPT Work combines all three!
I'd love to hear more from OpenAI about how they protect ChatGPT Work sessions against prompt injection attacks. I expect their answer is the same auto-review mechanism as Codex.
Figuring this all out took way more work than it should have.
I think there are two key problems here:
If the ChatGPT Work documentation included the exact system prompt and tool descriptions used by the agent I wouldn't have needed to write this post.
Shortly after publishing this article I had an idea. I started a fresh Work session and prompted:
Build a site that lists every one of your tools - nearly grouped into categories - and for each one explain what it does. Try to exactly duplicate arguments and tool descriptions where possible. Design aesthetic should be technical docs, minimal flare
Here's the site it built, which includes details of 223 registered tools - though 6 of those are from my own personal MCPs served via datasette-mcp.
I noticed that the only browser-related tool in the list was web.run, which has methods for running searches, opening URLs, and clicking links, but didn't look like the full story in regards to headless browser automation.
This made me suspicious that something was missing, so I told the ChatGPT Work session that built that tools reference site:
Add full copies of every skill to the website (separate pages linked to from the homepage)
It turns out ChatGPT Work uses a lot of skills - 44 in fact!
The control-browser skill explains how the browser works:
Run browser setup code through the Node REPL
jstool. In this environment the callable tool id typically appears asmcp__node_repl__js. [...]The ability to interact directly with the browser is exposed through the
browser-clientruntime via theagent.browsers.*API. Before trying to interact with it, you MUST emit and read the complete documentation returned byawait browser.documentation()in one go.
So I told Work:
Add the full output of await browser.documentation() to the bottom of the /skills/control-browser page
And now you can read that on /skills/control-browser as well.
A few more interesting Skills:
.docx filesimage_gen tool.xlsx, .xls, .csv, .tsv
You are only seeing the long-form articles from my blog. Subscribe to /atom/everything/ to get all of my posts, or take a look at my other subscription options.
Goliad State Park. Finally getting back to my state park quest. Summer is too hot, but I’m testing a new portable air conditioner.
County courthouse in Lockhart.
Successful launch of NASA's Nancy Grace Roman Space Telescope:
Roman pairs a large field of view with crisp infrared vision to explore vast swaths of the sky and probe deeply into cosmic history. This flagship mission will help astronomers explore dark matter, dark energy, and worlds outside of our solar system, known as exoplanets.
🚀
I've been watching Season 1 of The Pitt, and as its reviews said, it's great drama, and there are lots of episodes. I like shows like this because they're simulated communities, we come to know the characters, and think about them even when we're not watching. Before too long we'll have AI programming that makes these shows like software, the stories assemble around the characters, and then are generated the same way ChatGPT generates the art we call slop, which will someday not be a pejorative the same way sick and junk aren't negative terms today. So we'll be able to call up new episodes of The Wire or The Sopranos or The Pitt any time we like. We'd be missing the fun of discussing the latest episode with friends, or maybe we'll have bot-invented friends too who will simulate having watched the episode you just watched, which leads us to another masterpiece of entertainment, The Matrix. On the other hand, I've watched The Wire at least five times all the way through, and even though I know what they're about to say in a lot of places, I do forget the sub-plots so at least some of it can be new each time. And btw, finally an episode of Silo that's was surprising that I didn't see coming. I read the books, so I feel like I should watch the show, but until the latest episode I was uninterested in any of the characters. Finally there's a plot to think about!
Watched: Coyote vs. Acme. So much fun. They nailed it. 🍿
An absolutely delicious temperate cloud-free day in the Catskills, so I got up from the computer and enjoyed.
Ahhh, relax with the sights and sounds of the Windows 98 Disk Defrag Simulator.
Working on something new for blog book detail pages. I've been wanting to do this for years, finally getting around to it.
Anthropic announced a limits change on X in the most convoluted way possible:
Starting September 14, we're permanently raising standard weekly limits in Claude Code by 25% for Pro, Max, Team, and seat-based Enterprise plans. Until then, the current 50% increase will be in place.
Compared to today, this works out to a 17% reduction in weekly limits on Claude Code. We’re working on exciting changes that will make it feel like you’re getting more from Claude, while having more visibility and control of your usage. Can’t wait to share them.
I had to read it a few times. It's actually bad news.
Voting is now open for the 2026 Tiny Awards, featuring 10 sites that celebrate “the best of the small, poetic, creative, handmade web”. You have until Sept 25 to vote.
As a practical example of Standard.site in Micro.blog, you can view my latest blog post in Standard Reader. It pulls the Markdown from the PDS, and also shows Bluesky replies if any.
After sleeping on the Cursor + OpenAI news from last night, I think it points to a broader question of openness. I don't actually use Cursor, so it doesn't affect me, but I'm interested in the dynamics of the industry.
There were two ways to use GPT-5.6 inside Cursor:
What isn't possible, but perhaps is needed, is a solution in the middle. A portable subscription that any IDE can use to integrate with a model provider's flat-rate plan without having to negotiate a deal.
You can use a $20 OpenAI subscription with OpenClaw and Xcode. You probably should be able to use it with Cursor too, even if it doesn't give you as much access as from within Codex.
Many people believe that current AI prices are heavily subsidized and as soon as we're all addicted to using AI, the prices will increase. In fact, the opposite trend is already playing out. Prices are a fraction of what they were just a couple years ago. And as costs drop, Jevons paradox kicks in.
What seems untenable in the long run is token pricing. It is hard to wrap your head around or even estimate when building apps. Is this new feature going to be nearly free or expensive? How much is this refactor in Cursor going to cost?
Netflix is the same price whether you watch two shows a week or 20 shows. With enough subscribers, it balances out. AI has much higher marginal costs — Netflix bandwidth doesn't burn through GPUs — but as those costs fall, maybe the frontier labs can figure this out too. What seems missing is a flat-rate subscription that follows you across whatever tools you choose to use.
New photo journal web app Moments has added Micropub for posting. It also includes custom domain names and RSS feeds. This means you can post from apps like our own Sunlit, or add the posts automatically to the Micro.blog timeline.
Amazing what can be done with just feeds and a posting API.
Heat waves are disproportionately harming people with schizophrenia. “The bodies of people with schizophrenia, especially those on medication, simply regulate temperature very differently than others do.”
Late Friday night news from OpenAI. They are winding down their partnership with Cursor:
This decision was incredibly tough, as we care deeply about our models being broadly available for developers. We are making this choice because we cannot be confident that SpaceX will use our technology within our terms of service, based on our experience with Elon Musk's companies violating contracts.
Sort of makes sense with all the bad blood between OpenAI and Elon Musk, but the announcement is confusing. This is going to burn some goodwill at a time when things are going well with developers.
9to5Mac has a summary of yet another response from OpenAI. OpenAI doesn't believe any specific trade secrets are even alleged. From the filing:
Apple all but admits that it has not yet identified the alleged trade secrets, promising an “iterative process” of discovery, after which it will tell Defendants and the Court what the alleged trade secrets are.
The hearing is October 1st. OpenAI's DevDay is September 29th. Hopefully we'll see the new device before the lawsuit goes much further.
Manton wants to know why Marc Benioff chose the name Claudeforce for his brand-merge of Salesforce and Claude. My two cents. Claude is a great name. I think that’s ultimately what he bought. His customers just want to be able to check the box – we’re using AI. Boss is happy. When I was coming up as a young dude in tech, the old saying was “No one ever got fired for buying IBM.” Today, in 2026, it’s Claude in place of IBM. My father worked in marketing at IBM in Armonk, so I heard a lot about their strategies that meant the IBM brand would be kicking ass forever. All things must pass. But in the early days of the PC industry, IBM sure had the power. Kind of like the US going to war with Iran in 2026. In the end IBM wanted to get rid of Microsoft, but Billg had the last laugh.
I thought it was a dead pixel, but there's actually a tiny chip in the glass of my MacBook Pro screen. Annoying in the nitpicky, small life problems kind of way. Really want to keep this machine another couple years.
Got notice from Linode that 8 of my servers need to rebooted in the next month. A couple of them have 2-year uptime! There won't be downtime except for one of the servers, might be a short blip, which I'll schedule ahead of time.
I've been randomly reading and listening to physics topics, not sure why, including lectures by Richard Feynman. Something finally clicked with understanding relativity and spacetime. I sort of got it before, but not really. 🤯
I don't get the Claudeforce name. Why tie your product to a single model? That's someone else's brand. And whenever a competitor like OpenAI does something good, your own product will feel out of step too.
I wonder if Apple will ever merge App Store Connect into developer.apple.com. Pretty weird that things are still fragmented. ("Oh, you want to update an app ID? Better go somewhere else.")
It's a minor nitpick. Just the whole backend feels stagnant because they have a monopoly on app distribution.
Myke Hurley blogging about the upcoming campaign for St. Jude:
We are now within reaching distance of $5 million raised. What an incredible achievement in less than 10 years. I don’t think I could have imagined what we would all achieve together. But time and time again, the Relay community has stood by us and believed in this cause.
Great story at Cartoon Brew about Peter Cullen's voice acting work. I had never really thought why he was so good as Optimus Prime:
His brother’s idea of heroism became central to Cullen’s interpretation of the character and arguably had a tremendous impact on generations of children who could look up to a strong leader who didn’t rely on bravado or overt displays of aggression.
Rest in peace.
Ok, we’re gonna go a little old school today. I’m talking the late 1990s and early 2000s, back before permalinks (sorry), RSS, and social media turned blog posts & tweets into their own little discrete units packaged to be easily passed around and things were less homogeneous. Back then, blogs — at least the ones I loved — tended to be more of a stream of consciousness thing, just an unordered and uncurated list of whatever happened to pop into people’s heads or cross their screens.
So, I’ve currently got at least 120 tabs across 20+ windows open in Chrome. I know, I know, I’ve gotta get a tab manager, but this is where I am right now. Each of these tabs contains something interesting and potentially bloggable, but some of them have been sitting there for months and with moving stuff consuming most of my days right now, I don’t have time to sit down and give them each even a Quick Links treatment. So I’m gonna do them rapid-fire in the hopes that you might find some of it interesting. Here we go, let’s empty some tabs!!
The winners of the 2026 iPhone Photography Awards. Via my pal Jodi, I think. I love this one, which rhymes with this famous photo (Behind the Gare St. Lazare) by Henri Cartier-Bresson:

Ways of Treeing, “Meditations on time and rebellion against the straight line”.
The quiet grief of adult friendship. “Adult friendship became one of the most emotionally significant and least discussed losses of modern life.” (Via swissmiss; have I linked to this one before?)
Hugh Howey (author of the Silo books) on how the self-publishing window has closed (or at least narrowed). “We had less than 20 years of writing being hard and publishing being easy, and we will never get that window back.”
I haven’t done a media diet post since November (and may not again?) but I’ve been eagerly watching s03 of Silo, listening to Toni Morrison read Beloved (great book but what a fucking gift to hear Morrison read it), and enjoyed the new Spider-Man, a series that is at its best when the three leads just chill and talk.
“Artist and researcher Maggie Coblentz fermented miso in orbit, chasing a deeper question: What makes a miso a miso—or a person a person—away from Earth?”
I Eat the Stars: How to Live Fully and Beautifully in a Collapsing World by Sarah Wilson.
Messy Nessy’s Cabinet (always a good source of interesting links) redesigned recently.
These Dutch strip islands are cool.
How art invented humanity (instead of the other way around).
I don’t watch enough videos by Phil Edwards these days: “Why are bowling alleys across America replacing 2,000-pound mechanical pinsetters with nylon strings?”
Is democracy without elections possible? Could a lottery be better? I actually watched this a couple of months ago but never got around to writing it up. More on this from Alex Guerrero.
I really wanted to get to this one but just never could find the time: Historia Civilis on The History of the City.
This might be a minority view, but my YouTube algorithm is actually pretty good at showing me stuff I’m interested in. Case in point: when I was getting the link for the city video above, I noticed a newish video from Tom Scott about Bletchley Park and the secrecy of the Intelligence Factory.
The Spectacle: Has Taking the Perfect Photo Ruined Tourism?
Fordite is an artificial gemstone made up of dozens of layers of automotive paint that built up and hardered over time.
This material is typically found in the paint booths of auto factories, where cars were spray-painted by hand. As the paint sprayed over vehicles, excess paint would build up on the equipment and surfaces around the painting area. Over the years, these layers of paint would harden, creating colorful, agate-like structures.
Fordite is no longer being produced: “The switch from solvent-based paints to more environmentally friendly, quicker-drying paints in modern automotive plants has virtually ended the creation of new fordite, making it a finite material.”
Infinite Digest: a collection of infographics about DFW’s Infinite Jest.
Inge Druckrey: Teaching to See, “a 40-minute crash course in design thinking”. Found this via Edward Tufte.
Debbie Millman on the fleeting euphoria of success (YouTube).
March Forth is a feature-length documentary about Reginald Dwayne Betts, who was incarcerated at 16; “after his release, Betts went on to become an acclaimed poet, author, lawyer, activist and MacArthur Genius Award winner”. Trailer.
The Professor Who Taught People How To Think (1962). This looks great; shades of Feynman.
Screenshots of various operating systems (1983-2005). See also The Virtual OS Museum (“1,700+ pre-installed operating systems, ready to run”).
The Unsolved Mystery of Lorem Ipsum. Got this from Andy so you know it’s good.
Speaking of Andy, he recently posted about Amazon’s decision to shutter Mechanical Turk. Kinda shocked it was still going tbh.
2018 PBS special hosted by Michael Keaton: Why Mister Rogers Still Matters Today. I’m gonna make some time to watch this one after my move is done.
Love the premise of this: Mediocre Movies That Changed My Life. Eddie the Eagle!
Visual Instruments: An Exploration of Making Music Out of Images. This is a fun little web toy to play with.
I wanted to do a whole post on the artist’s palettes they have in the collection at the Met.

A Living Museum of the Comic. “A living museum of where the comic came from — its Western and Eastern roots, the women who drew the early funnies, the Europeans who invented its grammar, and the golden age of the American color cartoon. Every artifact is public domain, restored by us, and free to explore.”
From Kurzgesagt: The Impenetrable Barrier Trapping Us in the Solar System. “From the vast empty space to the laws of physics themselves, there are many reasons why the dream of space travel might stay just a dream.” Alt title: Why Humanity Will Never Leave The Solar System.
The Dingo Fence | The Longest Fence in the World. “In the heart of Australia, the continent has been cut in two by the longest fence in the world in a bid to protect livestock from dingoes. Dingoes have divided the nation for the past 150 years, but with new research coming to light, is it time for this to change?”
“I rebuilt 1901 by Phoenix from scratch to figure out what makes it so good.” I love this song and will make time to watch this soon.
The Deadly Legacy of HIV Truthers (2015). “Even as the medical community reached a consensus that HIV caused AIDS, a counter-movement was emerging, claiming that HIV didn’t exist, or that the virus existed but was harmless. The symptoms of AIDS, according to some of these people, were actually caused by HIV therapies themselves.”
Recent Howtown: So… did NASA find life on Mars? “Awkwardly, we’ve now collected the most promising extra-terrestrial biosignature in a tube that might spend the rest of eternity sealed inside a dead rover.”
I love these little icon collages and creations from Font Awesome (FA serves the icons I use on KDO). Starry Night! Sudoku! Chef Vader! Business Fett!
More than 200 Ansel Adams photos, mostly of National Parks, that are free to view & download. Via Open Culture, one of the gems of the open web.

The Crazy History of the Shower. I think that’s self-recommending.
Jamelle Bouie saw The Odyssey and shared his thoughts about it on YouTube. From his Letterboxd review:
The Odyssey or, what if you did THE DARK KNIGHT in ancient Greece?
This, to be clear, is not a bad thing. Nolan is covering the same thematic territory he did nearly 20 years ago and more recently with OPPENHEIMER. And it is compelling! A kind of Burkean liberalism that sees the heroism of “great men” as a necessary bulwark against disorder and chaos but understands the great men as merely men — who themselves bear some responsibility for the chaos — and sees heroism as a terrible burden to bear, more akin to a curse than a crown. There are no parades or honors. In the end you bear the scars and fade away, an exile remembered in legend.
Anyway, I am glad Hollywood keeps giving the man hundreds of millions of dollars to make his didactic and idiosyncratic blockbusters. Keep ‘em coming!
Yeah, exactly. I feel like most pans of The Odyssey fell into two main buckets: 1) it wasn’t long enough — pacing off, didn’t include my favorite scene, etc. — which was entirely due to 3-hour runtime constraints of IMAX projectors, and 2) I wish Christopher Nolan hadn’t made a Christopher Nolan movie.
Debbie Millman again: Is the World Getting Uglier? First brand she mentions is a Tesla Cybertruck, a completely tasteless monstrosity designed by Elon Simpson that I still laugh at every time I see one on the road.
Due to drought in Britain, you can see “gardens that disappeared hundreds of years ago starting to show themselves again through the grass”.
Artisans make punchcards by hand for the looms that weave beautiful banarasi sarees.
Ok, that’s all for now. I didn’t close all of the open tabs though…maybe I’ll do some more in a few days. How about you? Do you have anything interesting to share with the rest of the class that’s been languishing for a few weeks in a browser tab?