The Algorithm Noticed First: Reading My YouTube History Fourteen Months Later
Fourteen months after Digital Breadcrumbs, I pulled the whole log instead of skimming it — 373 items over thirty days. The three pillars swapped places, the music turned from listening into making, and the claim I was proudest of turned out to be false.
In July of last year I wrote a piece called Digital Breadcrumbs, where I looked at my YouTube history and found three tidy pillars: technical exploration, classical music, and productivity optimization. I concluded, rather generously, that my viewing history was "a map of how I think, learn, and grow," and that "understanding our patterns allows us to shape them intentionally rather than being shaped by them unconsciously."
That's a nice sentence. It is also the kind of sentence you write when you've looked at the data for twenty minutes and liked what you saw.
So I went back. Not to the vibes this time — to the whole thing. Thirty days, August 9 through September 8: 373 items, 249 full videos, 124 Shorts, roughly 81 hours of runtime. Then I re-read the old post next to it.
Some of it held up. The part I was proudest of did not.
The pillars swapped places
The 2025 post opened with this: "The largest chunk of my recent viewing centers on AI development, specifically around Claude Code, AI agents, and coding workflows."
Not anymore. Of the 249 long-form videos in the last thirty days, AI and technical content accounts for about 13%. Music accounts for 41%.
That's not a rounding difference. That's a different person's account. Music is now the dominant thing I do on YouTube by a factor of three, and I didn't notice it happening. I would have told you, confidently and wrongly, that I was still mostly watching AI content.
The most-watched channels tell the same story from another angle. Nate B Jones (AI strategy) leads at 10 videos. But right behind him: Valentina Lisitsa at 9, Deep Persian at 9, MunomaMusic at 7, Meghan The Piano Teacher at 5, Deutsche Grammophon at 4. Add them up and the pianists, the synth-cover channels, and the piano teachers bury the AI shelf.
The tell: 47 videos about gear
Here's the part that actually matters, and it's the reason I wanted to write this one.
In the old post I drew a clean line between the two halves of my viewing: "While the technical videos are about building and creating, the musical content is about appreciation and absorption. It's the difference between being a maker and being a receiver."
That line is now false, and it's false in the direction I've been trying to move.
Forty-seven items in the last thirty days — 32 of them full videos — are about musical instruments. MiniLab 3, MPK Mini, OP-XY, OP-1 Field, Launchpad, KeyStep 37 MK2, Roland JUNO, Fantom, TR-8S, SP-404, the MPC Key 37, the NuPhy keyboard. Alongside them: live-loop covers of New Order, Depeche Mode, Eurythmics, Haddaway, Ace of Base, a-ha — dozens of them, each one a person showing you how they made the thing on the box in front of them.
And then there's Meghan The Piano Teacher, whose five videos in my history read like a syllabus I enrolled in without telling myself: "Is there a RIGHT way to learn piano?" "How to Get Addicted to Practicing Piano (According to Science)." "7 Piano Myths that Need to Die." "How to Spot a Piano Scam."
The classical listening didn't go away — one Monday night I went fifteen deep on piano performances, Lisitsa into Rachmaninoff into Chopin into a Moonlight Sonata third movement, the kind of run I'd have filed under "appreciation" last year. But it's now surrounded on both sides by content about producing the sound rather than receiving it. Gear to buy. Covers to reverse-engineer. Practice routines. Theory. Scam-avoidance for people shopping for a piano.
The receiver became a player. The algorithm figured that out well before I sat down to write about it.
The technical thread grew up
The AI viewing shrank, but it also changed character, and the change is more flattering than the shrinkage.
Last year my history was full of titles like "Claude Code — 47 PRO TIPS in 9 minutes." This year the AI content is almost entirely architecture and strategy: conference talks from AI Engineer ("Why We Killed Our Multi-Agent Pipeline," "Forward Deployed Engineering 101"), Discover AI on agentic skills and knowledge graphs, "AI Agent Memory: 7 Types Explained," a breakdown of Claude's memory release, and a plain "Graph Engineering in 5 minutes."
I am, at the moment, building a knowledge graph and a memory system. My watch history is no longer a list of tips I might use someday. It's a reading list for a problem I'm currently stuck on. That's the healthiest signal in the whole dataset: I stopped consuming tutorials at roughly the rate I started needing them less.
The uncomfortable column
Now the part that doesn't flatter anybody.
Last year I wrote: "Quality Over Quantity: I tend to watch longer-form content (20–45 minutes) rather than quick hits, suggesting a preference for depth over breadth."
Thirty days of actual data: 124 of 373 items — one in three — are Shorts. Median video length is 11 minutes. Only 70 of 249 videos ran past twenty minutes; 102 were under ten. The "preference for depth" I awarded myself in 2025 was a claim about my self-image that I never checked against the log.
And the deeper problem isn't the Shorts. It's that a meaningful slice of the long-form content is watching about doing. Forty-seven gear videos is not one hour of playing. "How to Get Addicted to Practicing Piano" is a video you can watch instead of practicing. "5 AI Prompts To Fix Your Entire Life In 1 Day" is in there too, and I'd love to tell you I watched it ironically.
This is the actual shape of the consumer-to-creator transition, at least mine: it doesn't start with making things. It starts with consuming a different genre of content about making things. That's real progress — you can't fake the shift from "watch a performance" to "watch someone build a patch" — but it's a stage, not an arrival. The gear videos are the on-ramp. They are also, if you let them be, a very comfortable place to park.
The third pillar moved into the body
The old "productivity optimizer" pillar didn't disappear. It relocated.
Where 2025 was habit-stacking and "learn 10x faster," 2026 is ADHD accommodations, Huberman clips on memory and sleep, mitochondria, brain fog, concussion recovery, morning protocols. Twenty-two videos in the health-and-attention cluster. The optimization target moved from the workflow to the wetware — from how do I organize my day to what is actually going on in my head, and can I work with it instead of against it.
That reads to me like someone who has stopped believing that a better system will fix it.
The algorithm as mirror, revisited
Last year I wrote that the algorithm "doesn't just show me what I've watched — it anticipates the connections between disparate interests." I meant it as a compliment to the machine, and slightly as a compliment to myself, for having interests worth connecting.
Fourteen months of data suggests a duller and more useful reading. The algorithm isn't anticipating anything. It's a lagging indicator with a very short memory — it noticed the music shift because I made the music shift, and it amplified it because amplification is all it knows how to do. It's not a mirror of who I am. It's a mirror of what I did last week, held very close to my face.
Which is exactly what makes it useful, and exactly why the 2025 read was too kind. A mirror shows you the practice sessions you skipped as faithfully as the ones you did. It logs 124 Shorts with the same neutrality it logs a Chopin concerto. If you only look at it when you're in the mood to be impressed, you'll find something impressive — I did, last July, and I published it.
The honest version of my old closing line is smaller and harder: understanding your patterns doesn't let you shape them. It just removes your excuse for not trying.
I'll check again in a year. The number I'll be watching isn't the AI-to-music ratio or the Shorts count. It's whether the gear videos have started losing to the thing they're supposedly preparing me for.
What would thirty days of your history — the whole log, not the flattering skim — actually say about the gap between what you're becoming and what you're watching people become?