Weekly Thing 331 / RFC, Security, Tokens
Good morning! ☕️
At this point you might start thinking that I’m now sending these emails every other week. Nope… just had another busy weekend and wasn't able to get everything together in time. So here I am again with two weeks worth of links.
We are enjoying a lovely extended fall here in Minnesota. It is very odd to still have so many leaves on trees this late in the season. It makes for lovely walks though.
I hope you have a great weekend and see you next week! 😉

Maple tree giving us a great show of Fall colors. 🍁
October 18, 2025
Cannon Lake, MN
Notable
Claude Skills: Customize AI for your workflows Anthropic
These new "skills" for Claude create a whole new way to extend the capabilities of an LLM.
Skills are:
- Composable: Skills stack together. Claude automatically identifies which skills are needed and coordinates their use.
- Portable: Skills use the same format everywhere. Build once, use across Claude apps, Claude Code, and API.
- Efficient: Only loads what's needed, when it's needed.
- Powerful: Skills can include executable code for tasks where traditional programming is more reliable than token generation.
Think of Skills as custom onboarding materials that let you package expertise, making Claude a specialist on what matters most to you.
First with MCP, and now with Skills, we can create solutions that mix the AI non-deterministic solutions with the traditional software solutions. Simon Willison says they are are awesome, maybe a bigger deal than MCP which is saying a lot.
ChatGPT Atlas
I've spent some time playing with ChatGPT Atlas and it is pretty wild. It is notable to me that it blends Memory in with ChatGPT itself, so right out of the gate Atlas is able to contextualize the web for me with what it knows of me from my previous interactions with ChatGPT. I’m still getting used to what I can do having ChatGPT right in the web but my early tests were both compelling and wrong. I asked it to count things on web pages and it didn’t do a great job. But when I asked it to do research based on what I had been browsing it was phenomenal.
Scripts I wrote that I use all the time
I love this collection that Hahn has put together and shared with the world. He has identified a bunch of use cases I have never even considered. You can also read this just to get a feel for how customizable Unix command lines are. You really can create your own tooling to make it do exactly what you want to do.
Summary of the Amazon DynamoDB Service Disruption in Northern Virginia (US-EAST-1) Region
On Monday, Oct 20, AWS had a massive failure that took many services across the Internet offline. The core issue was a failure in DNS automation.
Many of the largest AWS services rely extensively on DNS to provide seamless scale, fault isolation and recovery, low latency, and locality. Services like DynamoDB maintain hundreds of thousands of DNS records to operate a very large heterogeneous fleet of load balancers in each Region. Automation is crucial to ensuring that these DNS records are updated frequently to add additional capacity as it becomes available, to correctly handle hardware failures, and to efficiently distribute traffic to optimize customers' experience. […] The root cause of this issue was a latent race condition in the DynamoDB DNS management system that resulted in an incorrect empty DNS record for the service's regional endpoint (dynamodb.us-east-1.amazonaws.com) that the automation failed to repair.
Without DNS no system knows how to find any other systems. Additionally AWS automation for many other services relies on DynamoDB. This was a particularly bad failure state.
Measuring Engineering Productivity - Off By One
The "measuring" in the title here is misleading since it really isn't about quantification. I'd say this is "managing" productivity and is spot on with what I've seen nearly all high performing teams do. There is a cadence and structure to engineering operations and this captures it well.
RFCs: Blueprints of the Internet | Broadcast
It has been a long-time since I've thought about RFCs but this article brought the memory back. RFC, or request for comments", were documents shared online to help form consensus and input on how protocols should work. They are what the Internet Engineering Taskforce create. TCP, IP, UDP, SMTP, HTTP, DNS — these were all created via RFC. I linked every one of those examples to their actually RFC. I love that the pages are simple text files. Every email you have ever received, every web page you have ever visited, every interaction you have ever had on your phone with some service works only because these RFCs exist.
I wish we still had this in our culture. Imagine if there was an active RFC process for Agents or MCP. I love that there is an RFC 8700 - Fifty Years of RFCs .
Agents 2.0: From Shallow Loops to Deep Agents
Agent "architecture" is a quickly moving topic and the state of the art is changing quick, but this approach to making more complex agents is a good example.
Moving from Shallow Agents to Deep Agents (Agent 1.0 to Agent 2.0) isn't just about connecting an LLM to more tools. It is a shift from reactive loops to proactive architecture. It is about better engineering around the model.
Implementing explicit planning, hierarchical delegation via sub-agents, and persistent memory, allow us to control the context and by controlling the context, we control the complexity, unlocking the ability to solve problems that take hours or days, not just seconds.
This space will see a ton of experimentation.
Roc Camera
It used to be hard to make fake images and now anybody can make images that are impossible to prove if they are real or fake. This camera is using a combination of hardware and software to create images that are proven to be real.
By combining attested sensor data, zero-knowledge proofs, and a tamper-proof environment, we've built Roc Camera to capture verifiably real photos.
A lot of cryptography work being done in here. This camera seems like a project not a product but I think it is cool to see the approach. Adobe and a bunch of existing companies are working on C2PA which is intended to solve this same problem and solves the whole workflow through image editing as well.
Why I code as a CTO
I get asked this question from time to time and I generally say no, a CTO should not be for their job. There is a lot more nuance to that question though for the folks that are really in it. My more specific answer is assuming you have a technology team that is more than a dozen or so people, then no, your CTO should not be taking sprint points. Notice I didn’t say coding, I said taking sprint points. This article was a good read because the author essentially highlights that he focuses on coding tasks that are not sprint points. He's working around the periphery of the code or on other efforts entirely.
Perhaps a better way to approach this whole question is to say Should CTOs practice their craft? The answer to that is a 100% yes, and in fact isn't really optional. I can’t imagine how a CTO could drive an AI strategy without using it deeply. Your craft is critical to leading. Yes, you should do that. And if your particular craft is coding, then you should do that in some capacity that doesn't mess up your team, like taking sprint points would.
toon: 🎒 Token-Oriented Object Notation – JSON for LLMs at half the token cost
Developers everywhere use JSON as the default way of handing data around including to LLMs. If you want proof of that take a look at the APIs for LLMs where you will find API methods to force the return message from the LLM to be JSON. This project was born out of a desire to be efficient with tokens and realized that you could format the data using a different style and get equally good results to JSON with approximately a 50% reduction in token consumption. Super interesting and for projects at scale could be a big deal.
More friends, more division: Study finds growing social circles may fuel polarization
I’m a big believer in Dunbar's number and I think it shows up all over the place in society. I particularly think the intersection of social media and the limits suggested by Dunbar's number are something we've ignored to our detriment.
This article caught my eye in a big way because it is talking about similar concepts — count of friends in certain levels. It suggests that our "close friends" number has risen, and that changes dynamics in notable ways.
Tolerance plays a central role. "If I have two friends, I do everything I can to keep them--I am very tolerant towards them. But if I have five and things become difficult with one of them, it's easier to end that friendship because I still have 'backups.' I no longer need to be as tolerant," explains Thurner.
Super interesting theory.
Easy 6502
Most developers work at very high layers of abstraction built on frameworks and APIs that are themselves already abstract. I often comment that most Python developers have no idea what is happening in memory when they use a Dictionary. One way to go at that is to gain an appreciation for how the CPU is really working by doing some Assembly code. This is a great way to learn about it right in the browser.
Claude Code: What It Is, How It's Different, and Why Non-Technical People Should Use It
This is a wonderful article that goes through in detail a complicated analysis and shows how you can use Claude Code to bring AI to the task. I've used Claude Code many times but only for programming efforts and looking at source code repositories. This article shows a completely novel way to use Claude Code for a project, not programming, and how you can build extra powers into that project. Torres is keeping this simple and approachable so she doesn't extend that doing the work the way she describes also has many other versions like sharing and version control.
Anyway, if you are not a programming but curious to use Claude Code for a more complex analysis this is a great approach.
Journal
Writing blog posts backdated a couple of days to when the thing actually happened is the most extreme form of not writing for the timeline. It is a gift to my future self when scanning my on this day page to have things show up in the right spots.
I’ve been using a Libre 3 Plus continuous glucose monitor and paying more attention to the specific foods that cause blood sugar spikes for me. By making tweaks to what I’m eating, when, and the order of foods I’ve lowered by baseline from 90 to closer to 70 with less spikes.

Driveway markers are in. Let it snow! 😊

Tammy led the way while we all helped and the shed is in better shape than it ever has been! I didn't get a before picture -- but you could not see any of the floor before.

Triple smash burgers tonight! 🍔

Figured out some support for these double wicks in the XL candles.

Started with 8 gallons of wax and 60 candles later…

Making Things 4 Good candles!

Supporting Membership
Hey there, Weekly Thing enthusiasts! 🌟 We're on an exciting mission to support the incredible Electronic Frontier Foundation (EFF), champions of digital freedom and privacy. So far, we've rallied together to raise an impressive $298.93, and with your help, we can soar even higher! With 27 weeks to go until we send our full contribution to the EFF, and just 23 amazing Supporting Members leading the charge, now is the perfect time to join our ranks. By becoming a Supporting Member, you're not just contributing funds—every single penny goes directly to the EFF to fuel their vital work. Let's make a powerful impact together and ensure a free and open internet for all. Ready to take the leap? Jump in and become a Supporting Member today!
| $4 monthly | $40 yearly |
Briefly
🎁 Gift article. Article about one of our favorite musicians as she gets ready to release a new album this week. See also list of Brandi Carlile shows. → Brandi Carlile Climbed Music’s Peak. Then She Had to Start Over. - The New York Times
This is a smart move by Google to bring their market leadership in Maps to Gemini and open up a whole bunch of use cases that could be created with that rich dataset. Question is will this MCP-like tool capability be made available to any other models? Hmmm… → Grounding for Google Maps now available in the Gemini API
I’m an F1 TV Premium subscriber and think it is great. Early read on this is that we will still have all those capabilities. I’m wondering if I may get to watch the driver view from my Vision Pro next year? 🤩 → Apple is the exclusive new broadcast partner for Formula 1® in the U.S. - Apple
Using the blockchain to store immutable assets to be distributed via malware is innovative and probably highly effective. → North Korean Hackers Use Blockchain to Hide Crypto-Stealing Malware | eSecurity Planet
Why in the world would you send data back and forth between satellites unencrypted? This seems bonkers to me. → 🛰️ SATCOM Security
Non-linear gauge to drive the outcome that is best. Delightful example. → When design drives behavior
I highlighted Sky in Weekly Thing 321. Their product was gobbled up by OpenAI before it was even launched. → OpenAI acquires Software Applications Incorporated, maker of Sky | OpenAI
New feature makes a ton of sense and opens up some more interesitng AI use cases for enterprises. This kind of capability is for sure the direction we will go with AI in companies, and how you'll get to a point where all meetings and interactions get recorded so that data doesn't leak out and lost. → Work smarter with your company knowledge in ChatGPT | OpenAI
Make a server in your basement feel like it is a robust platform that you can easily push projects to. Very cool! → Disco - Self-hosting that doesn't feel like self-hosting
Cool collection of space-related logos. → Space Exploration Logo Archive
e-bike's are generally pretty awesome and they just keep getting better. This take from Rivian breaks some assumptions — notably the pedals aren't even connected to the drivetrain! → Rivian’s Also e-bike is like nothing you’ve ever seen before | The Verge
This looks like a fun game. It reminds me of the early days of SimCity before it was layered with so many features (enshittified!) that it became stupid and overly complex. Building some subways could be a really fun simulation game. → Subway Builder
I think Apple should stop anything and everything they do with advertising. They aren't any good at it and it is completely damaging to their brand. Just stop. ✋ → Daring Fireball: Gurman Reports That Apple Is Preparing to Sell Ads in Maps Starting in 2026
Lovely article that is super fun about how your computer actually gets itself running. Some stuff in here I had never thought of myself — like what happens before the BIOS itself is actually loaded. I dig these kind of walk throughs. → The Linux Boot Process: From Power Button to Kernel | 0xkato
Hosted option for storing bookmarks. Every link you read in the Weekly Thing I manage using Pinboard which is a similar service. This one has a lot of features I wouldn't need but some may like. → Linkwarden - Bookmarks, Evolved
I could bang on for days about how much I love RSS and algorithm-free feed readers. Here is yet another open source and self-hosted option. I love seeing continued diversity in this ecosystem. → srssly: Minimalistic RSS server and ui
A bunch of different tools to do things with maps. → GeoUtil — Free Online Map & Geography Tools
Simple and pretty easy to read way to filter through JSON data. → JSON Query - a small, flexible, and expandable JSON query language
One of the leading local SaaS companies goes private. → Francisco Partners to buy Jamf in $2.2 billion take-private deal | Reuters
A haiku to leave you with…
Tokens cut in half,
JSON's light for LLM—
Savings in each byte. 💡
Would you like to discuss the topics in the Weekly Thing further? Check out the Weekly Thing on Reddit. 👋
Want to share this issue with others? The link is…
👨💻
This work by Jamie Thingelstad is licensed under CC BY-SA 4.0.
My opinions are my own and not those of any affiliates. The content is non-malicious and ad-free, posted at my discretion. Source attribution is omitted due to potential errors. Your privacy is respected; no tracking is in place.