DRAGON
the
blogCATCHERthe
blog

Nobody knows anything February 25, 2026

Here is Derek Thompson on AI and where it’s going:

[ … ] I am lucky to have par­tic­i­pated in con­ver­sa­tions about the future of AI with exec­u­tives and builders at fron­tier labs, econ­o­mists at AI conferences, AI investors, and other big­wigs at off-the-record din­ners where impor­tant truths can the­o­ret­i­cally be bandied about without risk. And if I had to pick three words to sum­ma­rize this col­lec­tive expert view of the future, I could not in a mil­lion years, or with a tril­lion tokens, find three words more suit­able than these: Nobody knows anything.

This was my asser­tion in the first edi­tion of my pop-up newsletter, but of course, one always wonders, is it just me who doesn’t know anything … ? Hearing Derek say it, too, makes me feel more confident.

Signs and portents February 24, 2026

I’m sure it won’t last long, but I just want to memo­ri­alize a strange event: on Tuesday, Feb­ruary 24, at approx­i­mately 8:10 p.m. Pacific Time … Google search … was DOWN!

I guess they Gemini’d too hard!

It was the best of times, etc. February 23, 2026

On his fab­u­lous new blog, Marcin Wichary links to a post of mine, the one where I wrote:

What makes the AI chat­bots and agents feel light and clean, here and now in 2026? Is it an innate archi­tec­tural resis­tance to advertising, to atten­tion hacks, to adver­sarial crud? No — it’s that they are simply new! The lan­guage models in 2026 are Google in 1999, Twitter in 2009. Their vast con­joined industry of influ­ence hasn’t yet arisen … though it is stirring.

Marcin’s view is char­ac­ter­is­ti­cally clear:

The AI com­mu­nity tends to say “this is the worst this will ever be” in response to criticism, but in a very learned sense, in many aspects it is also the best it will ever be.

This is useful: a sort of counter-mantra. The enthu­si­asts really do invoke “this is the worst AI will ever be” almost like a magic spell. Along the narrow tech­nical dimen­sion, the claim is cer­tainly true … but the tech­nical dimen­sion is not the only one, nor even the most impor­tant. There are com­mer­cial and eco­log­ical dimen­sions, too, and we can look at every other dig­ital system — literally every other dig­ital system! — for clues about tra­jec­to­ries from worst to best, best to worst.

Your claw does okay on the internet of 2026. How will it fare on the internet of 2029, which will surely be crowded with adversaries — hacker claws and trick­ster claws? Recall that the hackers and trick­sters are, as always, MUCH more moti­vated than you to field this software. Even if they are denied access to the tip-top models, their next-best instru­ments can com­pen­sate with volume and malevolence.

My strong hunch is that claw and coun­ter­claw will net out into a sort of cruddy detente. See exhibit A: email. Or exhibit B: the World Wide Web.

That’s the thing about technology: every­body gets it!

First time for everything February 23, 2026

My path as a pro­grammer has mostly been guided by avoidance: of complexity, fragility, exces­sive moving parts. I like my pro­grams simple, stateless, cool to the touch: mortar and pestle, not food processor.

So, it took thirty years … but I finally used a queue, and I finally used a ring. I am repli­cating data across three dif­ferent buckets, and the envi­ron­ment imposes con­straints that pre­vent me from doing what I’d nor­mally do, which is load the data into memory and slam it into all three loca­tions at once.

I can’t say I love the feeling of that queue out there … looming … growing … but it seems to be working. I’m proud of myself for fig­uring it out, even if I’m mostly just using Cloudflare’s ergonomic off-the-shelf offerings.

I’ve still never written a test, and I don’t plan to start!

The voice of the computer February 22, 2026

Here is the most spec­tac­ular demo of present AI systems: pull out your phone, ini­tiate Gemini’s live voice mode, say “please trans­late this con­ver­sa­tion between Eng­lish and Japanese”, and allow the system to act as a respon­sive and com­pe­tent interpreter.

ChatGPT offers a mode like this, too; it’s clear Google and OpenAI have both invested a ton in these features, and that both believe they will rep­re­sent a significant — THE significant? — inter­face to their models. Indeed, OpenAI’s upcoming devices are all premised on this.

Meanwhile, it seems odd to imagine “the voice of Claude”. I’m sure Anthropic could buy itself a voice mode, but this inter­face shouldn’t be mis­taken as a lan­guage model sand­wiched between voice recog­ni­tion and TTS — it’s sub­tler and more fluent than that. I believe (though of course I could be wrong) these voice modes are the result of sup­ple­mental end-to-end training.

I’ve demoed Gemini’s live voice mode sev­eral times, always to great reception, yet never actu­ally used it for a prac­tical purpose. I mean: I was just in Japan for two weeks! And I don’t speak Japanese! And I never used it!

I still imagine I might — it is really sharp — but, for me, it has remained a daz­zling demo, not a useful tool. The infi­nite bummer of “let’s both talk into my stupid phone” remains an insu­per­able barrier.

I’d love to know the usage sta­tis­tics that Google and OpenAI are seeing; I’d love to know if and how people are really engaging with the voice of the computer, straight out of Star Trek.

Artificial general economy February 20, 2026

Here’s a gimlet post from Benn Stancil, on this weird manic moment:

Ver­tigo is the sen­sa­tion of spinning. It is feeling dizzy, even when the world is standing still. If every Anthropic press release is all we talk about, have the robots not already taken over? If every com­pany is urgently rear­ranging itself around a work­force of agents, does it matter how well they score on the tests? Are we learning, by posting to our Sub­stacks and reading the discourse, or are we becoming obsessed? Is this takeoff, or just takes?

The new funnel? February 20, 2026

Here’s an inter­esting and sur­prising nugget in a recent post by Justin Duke, he of Buttondown:

From both onboarding sur­veys and overall quan­ti­ta­tive traffic: half of our cus­tomers come through word of mouth, one-fourth through search, and one-fourth through LLM-based traffic. (I’ll talk more about that last bucket in a later post.) When you divide that not by overall cus­tomer count but by revenue, it’s two-thirds word of mouth and one-third every­thing else.

I will be very curious to read that later post about the LLM traffic!

Justin’s 50% word of mouth, 25% search, 25% robo-brain feels both (1) very healthy, and (2) very futuristic — a char­ac­ter­is­ti­cally good balance.

Public service announcement February 9, 2026

I’ll be away from the blog for a few weeks, so let me just pin this to the top here:

I still write my own code, because I enjoy it.

I have never been a normal pro­grammer, and maybe you aren’t, either. So, let’s just do things the way we want — the way we like.

😌

Flood fill February 9, 2026

A new edi­tion of my pop-up newsletter has arrived: Flood fill vs. the magic circle, about AI automa­tion and its limits.

Also, the actual email I transmit con­tains a com­pendium of don’t-call-them-tweets, i.e., stray thoughts adja­cent to AI. These are fun to pull out of my notebook; con­sider them EXCLU­SIVE CONTENT, and an entice­ment to subscribe.

Four more edi­tions, then poof!

The music of the feeds February 6, 2026

The latest Disquiet Junto assign­ment is out, and this blog pro­vides a link in its cre­ative chain. The assign­ment is:

Write a piece of music emu­lating the dopamine engine that is social media.

Click through for Marc’s instructions, which are, as usual, provoca­tive and fun. I myself have responded to a grand total of one (1) Junto assign­ment; if I wasn’t headed out of town, I’d give this one a try, and double my output.