4 min read

October 2025 bakery

A record of what has caught my attention and ideas I've been thinking about this month.
October 2025 bakery

A record of what has caught my attention and ideas I've been thinking about this month.

Half baked

Some written thoughts, but not enough for their own post.

One of my favorite new things to do with LLMs is to get them to snitch on each other. Here's what I mean.

In tab one I'll open Claude and use its research feature with prompt:

Please create a detailed integration guide for XYZ api.

Once the report completes I open up tab two again with Claude and its research feature with prompt:

Please assess the veracity of this integration guide.

Which produces something like:

The guide contains mostly accurate information with three significant errors and two important contextual gaps. Of 13 technical claims verified against official XYZ documentation, 8 are fully accurate, 3 contain errors, and 2 require important clarifications about API context.

Followed by the exact items that could not be verified.

Raw

Naked links

AI

The idea that LLMs could be distilled down to just the thinking bits, without so much memorized context was interesting. Also enjoyed the end section on teaching and AI tutors.

Claude Skills are awesome, maybe a bigger deal than MCP
Anthropic this morning introduced Claude Skills, a new pattern for making new abilities available to their models: Claude can now use Skills to improve how it performs specific tasks. Skills …

I like, "Claude Code is, with hindsight, poorly named. It’s not purely a coding tool: it’s a tool for general computer automation. Anything you can achieve by typing commands into a computer is something that can now be automated by Claude Code. It’s best described as a general agent. Skills make this a whole lot more obvious and explicit.""

Designing agentic loops
Plus coding agent PR stats and AI-assisted tooling that finds genuine security issues in curl

Always a good read - I like the joy of yolo mode.

Even if AI makes developers 30% more efficient, that is a total game changer.

I work on this!

Really good vibe coding workflow demo

Superpowers: How I’m using coding agents in October 2025
I used to write more

I started using Superpowers. It's very good.

Music

Their greatness cannot be denied.

Dear lord.

Software

Talking to the Bank of England about systemic risk and systems engineering
Patrick McKenzie travels to Threadneedle Street to brief the Bank of England on systemic risk and systems engineering.

Come for the talk on the CrowdStrike incident, stay for the dive into Tether and AI! This is one of my favorite Patio11 talks of the year. I really love when he reads one of his essays and interjects parenthetical remarks during the narration.

Home improvement lending with fewer bankers and more computers
Patrick reads his essay about the surprisingly competent financial machinery that makes buying windows painless.

Not really software, but a great essay.

Good real talk.

In fact, we already have an excellent example of (deterministic) agents making micro-transactions at scale: the entire digital ads ecosystem! Every time a human loads a webpage, an awe-inspiring amount of computation and communication happens in milliseconds, as an auction is run to fill the inventory on that page with an ad that is likely to appeal to the human. These micro-transactions are only worth fractions of a penny, but the aggregate volume of them drives trillions of dollars worth of value.

Craft and beauty

Admittedly impractical and irrational objects.

If only I looked this cool when I reconfigure my development environment. brew install neovim just doesn't have the same theatrics as wielding a circular saw. But if you think about it, it is essentially the same thing.

When design drives behavior
In some cases, design is what something looks like. In other cases, design is how something works. But the most interesting designs to me are when design changes your behavior. Even the smallest details can change how someone interacts with something. Take the power reserve indicator on the A. Lange & Söhne Lange 1 watch. The power res…

Meticulous craft.

We have become a society of convenience above all else.

Politics

"Fuck you Charles Murray... Don't cut that... I hate this guy. The day he dies our society will be marginally better because of it." This episode is not pay walled and is one of Matt and Sam's absolute bests. Strongly recommend a listen.