newsletters
Newsletter #1: The Embrace of the Dark Side
On the inevitable compromise of values, the AI arms race, and what I actually got done this week.
23 ·03· 2026 — Robert Shalders
TRANSMISSION_01 — UPLINK ESTABLISHED
This week I decided to try Claude.
More specifically, I decided to use Claude Code to build and iterate on this very site. Which means you are, in a sense, reading the output of a machine I invited into my workflow whilst simultaneously complaining about it. The irony is not lost on me.
For the record: I am not comfortable with this.
The Compromise
I have, for a long time, held the position that open-source tooling, local models, and self-hosted infrastructure are the correct path. The reasoning is straightforward — you own the process, the data, the output. There are no usage limits, no terms-of-service updates at 03:00 on a Tuesday, no ambiguity about what happens to your inputs.
And yet.
The velocity is genuinely difficult to argue with. Things that would take a full afternoon took an hour. Things that would take an hour took minutes. The iteration loop contracts to a degree that changes the nature of prototyping entirely.
This does not mean I have abandoned my principles. It means I have temporarily leased them at a rate I consider acceptable for now, subject to review.
The RAM Problem
A less-discussed side effect of the AI boom: memory is getting expensive again.
LLM inference at scale requires enormous amounts of high-bandwidth memory. Data centres are hoovering up HBM and DDR5 at a rate that has begun to ripple through consumer pricing. If your workstation upgrade is suddenly more expensive than you expected, you now know where to direct your quiet disappointment.
We are, in a very literal sense, paying for the convenience.
Other Things I Have Been Paying Attention To
The Kohi Game Engine — I have now watched somewhere north of 80 episodes of Travis Vroman building a game engine from scratch in C using Vulkan. This is not a casual watch. Each episode is dense, precise, and surprisingly meditative. If you want to understand why your high-level abstractions cost what they cost, watching someone build the floor they stand on is instructive. Highly recommended if you have the patience and the curiosity.
Onur Mutlu on Computer Architecture — Professor Mutlu taught computer architecture at Carnegie Mellon and later at ETH Zürich. He has, for years, recorded and released his lectures freely. They are long. They are rigorous. They are some of the best educational material available on the internet, and they are free. If you have ever wanted to understand why your software is slow at a level below the operating system, start here. The lectures on DRAM and memory hierarchy alone are worth the time investment.
Aviation Update
Nine theory exams completed. The cross-country navigation assessment and flight skills test remain outstanding. The sky continues to be present.
Progress is being made. More when there is more to report.
Until Next Time
The transmission ends here. The noise continues.
If you found something useful in this, tell someone. If you did not, the feedback mechanism is the same email address you used to get here.
SIGNAL_LOST — RECONNECTING