Tess report - 2026.0

January 1 - January 15

hi! I'm gonna try to start writing these casual reports on my life about twice a month. the idea is they're part public journal, part tessware newsletter, and serves as an incentive for me to ship and legiblize my work. very experimental, we'll see how long I keep it up.

what I did

• spent new years in Berlin (highly recommended, the most... immersive.. fireworks show you'll ever see)
• went skiing near Niseko, Japan
• transferred through both Real China and Fake China (take your pick as to which is which)
• got sick
• hung out in San Francisco
• got (mostly) not sick
• made it home to Portland

what I shipped

• myself, half way across the world (as mentioned)
• an error handling framework for simpler and more correct C. I put this together in a few days for another project. one could argue this isn't in the spirit of shipping because it's mainly intended for my own use, but it's pushed and I wrote docs so it counts.

what's on my mind

as is often the case, the future of software engineering. in the last year my productivity has taken a significant hit from the prospect of LLMs decreasing the long-term value of my work. what's the point of investing in a code base if it will be ~free to generate it from scratch in a year? what's the point of designing new languages, tools, frameworks and paradigms if all human-written code will be uncompetitive with... whatever it is that LLMs do.

I now have a different perspective on this.

despite a lot of code getting machine generated these days, LLMs clearly struggle with software engineering (as evidenced by me still running more and better custom software than any vibecoder I know of, and AI labs increasing instead of decreasing their human engineering capacity). this could flip at any point, but when and how is not predictable. there's a lot of strong incentives to claim LLMs are currently effective at software engineering, and that them getting better is predictable, which explains why a lot of people are saying things I disagree with.

there's now an increased chance that tech I build today is made irrelevant for reasons outside of my control, however it's still the highest leverage thing I can be doing. by focusing on projects with relatively short-term rewards I can mitigate the risk, which is really what I should have been doing all along. "long-term investment" often degrades into playing with model trains if left to fester without a punishment/reward loop.

it's also not clear that anybody is making close to optimal use of current LLMs or related fundamental tech. the people working on LLM coding tools are often AI-pilled, which makes sense but but indicates these tools may not be receiving the highest standard of human engineering. I believe expert human attention is still necessary to build the best software, indicating there may be room for improvement. I've been procrastinating getting into the field for some time, but it may be a good idea for me to do that soon.

links

• GUIs for LLMs. haven't played with it myself yet, but seems cool: Stop using natural language interfaces
• a lot has been going on in the world, (mostly) unrelated from my life or software. I'm not big on "the news" but having at least a minimal up-to-date understanding of geopolitics/conflicts seems potentially valuable, not to mention interesting. for updates on who's doing what and (seemingly) well-calibrated analysis on why, I recommend William Spaniel on YouTube.

that's all, Tess out