2025 In Review

Like last year , this is my somewhat rushed recollection of the year that was, and as usual it’s a mix of personal and professional reflections, with some thoughts on technology and trends thrown in for good measure. This saves me hours What it feels like right now It's pretty much like this Not as tidy in real life A messy desk is a happy desk... Almost Sometimes I wonder if I shouldn't stick to reading Our gaming setup, minus the portable streaming A bit more movement, please That Agentic Thing Yeah, AI has been kind of a topic over the past few years, and we’re now at a point where the hype is under so much pressure to become real that things are either turning into diamonds or starting to crumble into carbon dust. But it’s pretty obvious that we’re already going into a cultural feedback loop-this is what I got when I prompted for a magazine cover that was “about agentic AI”, and I think it’s the perfect reflection of our collective unconscious: One of my takeaways from the year is that people outside (or inside ) the tech circle keep falling into the trap of either anthropomorphizing AI or dismissing it out of hand-and both are the wrong way to deal with it, especially the Luddite take. Another is that a fair amount of folk are finally aware of the economic implications of AI not panning out and the bubble finally bursting, either at the hardware level (I am somewhat happy I don’t own any NVIDIA stock) or by OpenAI imploding in some way. But for people like me, AI has indeed been a force multiplier. There were three times this year where, for me, models stepped over different qualitative thresholds, and the upshot is that now, at the tail end of 2025, I finally trust them to help me accomplish real work-with proper guidance, of course (which arguably implies experience and forethought). A few examples: I now use GPT-5.2 (and the 5.x-codex variants) on an almost daily basis to generate scaffolding, review code and add the kind of uniformity and standardized logging that I would find too boring to hash out. One of the most noticeable inflection points has been around generating unit tests and UI components that I then polish. For me, it’s turned front-end development into a commodity-I routinely give Copilot a couple of URLs, tell it to “build a Preact UI for this API with that base CSS”, and all the drudgery just vanishes. I have started making a list of the gripes I have with particular pieces of software and just diving in and fixing them, even if I’m not familiar with that stack or tooling. I typically ask for an assessment of the codebase, then for guidance in structure and feature mapping, and then progressively zoom in until I can match my gripes with a section of the code which I then either hack at myself or write a brief on...

Preview: ~500 words