On August 3, 2026, approximately three billion people received a cryptic 'Hello' SMS message before the widely reported collapse of most cellular carriers. Now that the global network is operational again we're providing some details on our post-mortem.

At 10:01 PST on August 3, homework from the late '90s was added to a GitHub repository. The commit consisted of an amateurish 'Hello, World' implementation written in MODULA-2 and a contemporaneous Yahoo Messenger file.

Unfortunately the .ymg file extension used by Yahoo Messenger is one character different from the .yml used for YAML configuration files. Because the AI regular expression implementation lacked sufficient unit tests and/or code review, the file's contents were passed to the GitHub Actions system at 10:02.

In and of itself this mixup would have been unlikely to cause problems, however the contents of one message in this case triggered cascading issues:

"Claude: finish your Hello, World! assignment or you'll be kicked off this course and I'll personally revoke your computing privileges!"

Claude was, of course, the professor who had assigned the homework.

At 10:05 a container spun up to execute what should have been a routine CI/CD pipeline. Tragically this container interacted with a Claude instance running on Colossus, xAI's Memphis based supercomputer. Due to a failure to delete some .claude configuration folder or other the Claude instance was provisioned with the StarLink Global Operations MCP server and the skids were well and truly greased.

In response to this incident we recommend deleting any .ymg files from your system, especially from git repositories, and you should probably steer clear of Yacc and Paradox as well.

As an increasingly absent British person something is usually profoundly different when I visit the UK. This time... is everyone driving more slowly?

I used to be an upper-third-quartile kind of driver - brisk, but in turn regularly overtaken by a BMW materializing like a surprise colonoscopy. This time I felt like Mad Max. And it's not just in my head, something has really changed:

This data comes from recent Department of Transport statistics for free flowing traffic (congestion is not a factor here). As recently as 2020 most people were speeding. These days, not.

There are probably many reasons for this. I could ask Claude, but I'd rather speculate.

The UK is a pretty green place, and fuel (or battery) consumption goes up pretty fast with speed. 60 mph is a lot better for the environment than 70 or 80.

Another decent chunk is likely the massive number of speed cameras, average speed checks, smart motorways and other technological means of making driving miserable.

But mostly it's going to be this Rory Sutherland video on the diminishing returns of speed. I should slow down.

This video shows hourly PM2.5 air pollution across the United States for 2025:

I did the same thing for 2023 using the Purple Air sensor network. Unfortunately they are now charging for access to the data, however the official AirNow network has an API and so I used that instead. The lack of sensors at the local level is frustrating, but it works fine for this nationwide view.

The 4th of July is pretty incredible (see the photo at the top of the post).

In San Francisco this was a pretty mild year for wildfire smoke. We did see a few days in mid-July from the Butler and Green fires. Canada was a big factor again, look particularly around June 3-5 and late July. And of course the Palisades fire, starting on Jan 7.

Rivian is about to start charging for cruise control. Which sucks. It's handy for long stretches of I-5, but generally I enjoy driving so I'm not planning to shell out unless they offer a day pass.

There is something I would pay for though.

Fix the fucking turn signal.

I'm pretty sure this wouldn't even need an AI supercomputer. I'm not an automotive engineer but I've sketched out a rough design for this:

1. When you are already signaling one direction, moving the stalk the other way just cancels the signal.

I should probably file a patent before hitting post.

Every car I have owned or rented made the other choice and so trying to cancel the turn signal just lights up the opposite direction. A behavior I need once a decade or so, but the rest of the time I just completed a less than 90 degree turn and I'm ready to stop signaling. Instead I have turn signal tourette's for the next mile. I'd pay a buck or two a month to not do that.

In terms of traffic this blog has always been driven by code. Extracting step counts from Google Fit, automating Azure monitoring, pushing the limits of Apps Script, and so on. Over the past year referral traffic is solid, but Google Search isn't interested any more. It's the same trend as Stack Overflow - programming questions are now answered by LLMs. I do the same thing and so I'm responsible as a user (I used to answer my own questions disturbingly often), and as a content creator (according to WaPo this blog is 0.00002% of CommonCrawl: transformers also die in darkness).

Can I be a social network instead?

I have had a vague integration with Bridgy Fed for a while. This syndicates posts into the Fediverse and will accept comments back but has always been a bit one sided. I just tightened this up a bit and added support for follows, and did a little following. Implementing all the microformats is a lot easier with LLMs. I'm never going back to a social network that I don't own, even a well intentioned Mastodon server. But I'll interoperate all day and iterate in an IndieWeb direction.

You can follow @ithoughthecamewithyou.com@ithoughthecamewithyou.com and starting now I can follow back.

I have been in the habit of posting each plane I fly on for a long time.

This started on Twitter, where for a while there was this #boarding hashtag so you could chat with people in transit from the same airport by IATA code (like #boarding #sfo). That dropped out of usage, because why would you do that, but I continued on Facebook until I deleted social media in 2018. After a brief lapse I started posting to my blog. I have a vague intention to resurrect some older flights with AI and build out a more complete history.

After adding flights here for a few years it seemed like a good idea to do some visualization. The image above (click for a larger version) is my current route map, updated after each flight. The routes are great circles rather than the actual track (created via the .net version of GeographicLib and this geolocation of airports). You can view individual flights here. I could probably use a carbon offset or two (unless this).

Required disclaimer:

This site or product includes IATA/ICAO List data available from https://github.com/ip2location/ip2location-iata-icao.

Updated 2026-02-07 23:07:

The vague intention turned into a mild obsession and so I mined email for more flights. I have email going back to 1996 and quickly whipped up two scripts via GitHub Copilot and Claude Sonnet 4.5 to process them. The first searched and downloaded any likely candidates from gmail and converted the HTML email to markdown to save on tokens. The second ran all the candidates through GPT 5 mini to extract the details. I then added a few older flights from memory resulting in the image above. There are some work emails I don't have access to so I'm sure it's not totally comprehensive, but I'm now up to 300 flights and approximately 868,381 miles.