What I’m thinking about: The AI job-market statistics I routinely write about just grew faces at my brother’s wedding…and what my own family reveals about the real-time experiment we’re ALL living inside.
My brother got married(!) a couple of weeks ago. (I had the honor of serving as Best Man, and my daughter was one of the flower girls)
(The wedding itself was wonderful and the healthy kind of reminder that nothing digital competes with a room full of family and friends…pure nourishment to the soul, and the longevity research backs that up.)
Somewhere between the toasts and the dance floor…the AI job market pulled up a chair.
It skipped the headlines and charts this time and showed up as two conversations with cousins I see maybe once a year.
When the labor data shows up at your family’s table
Three years ago we celebrated a cousin’s son graduating college with dreams of animation and design work. That was early 2023, when ChatGPT was brand new and almost nobody was pricing in what AI might do to creative careers (AI-derived artwork was laughably bad, all pseudo-body horror,4-fingered hands and waists sporting an extra leg.)
At the wedding I got the update. Three years later, still no traction toward animation, just lower-paid local work while the target industry drifted further out of reach. His parents called it a double hit (a brutal job market for young people, colliding with AI arriving in full force). I could tell it was an uncomfortable topic, and I didn’t pry.
Another set of cousins brought a sharper contrast. Their older son (age 21) wants into animation as well, with no path opening yet. Their younger son (age 19) is pairing coding and engineering WITH AI, in a program reporting a 90%+ job placement rate (as relayed to me, and the direction tracks with the data below).
Those two brothers stand on opposite sides of the clearest split in the 2026 labor data.
Automation vs. augmentation…the fork in the data
Stanford’s Digital Economy Lab, working with ADP payroll data, found employment for young workers in the most AI-exposed jobs fell ~6% between late 2022 and mid-2025, while older workers in those same roles GAINED 6-9%. Stanford’s 2026 AI Index separately pegs employment for software developers aged 22-25 down nearly 20% from its 2024 peak.
The mechanism matters more than the headline. Where AI automates the task itself (producing the deliverable end to end, with routine animation and graphics squarely in that bucket), entry-level hiring declines. Where AI augments the work (a human directing, checking, and iterating), employment holds steady or even grows.
The split runs at the task level rather than the job level, which is why two brothers aiming at adjacent creative-technical careers can land on opposite sides of it.
Zoom out and the entry rungs are visibly thinning…new-grad unemployment hit 5.7% in late 2025 (versus 4.2% for all workers), with entry-level postings down ~35% since early 2023.
My conviction has only hardened…every job now needs AI paired with it (whether through direct AI implementation, managing AI employees, and/or managing humans who are implementing AI), and a moat built around your skill set, which is what my AI coach Callan Faulkner calls becoming an AI architect. Patrick O’Shaughnessy (host of the Invest Like the Best podcast) has a frame for this, this is who you’re up against, because the person competing for any seat is increasingly applying these ever-improving tools daily.
My wife knows it’s AI…and keeps watching
The content side of this experiment runs inside my own house.
My wife recently found AI-produced episodic videos in the romantasy space (think A Court of Thorns and Roses territory), some with six-figure view counts (e.g., The Rise of the Lycan Queen)…not far off from how some video game cutscenes are shot. The channels are open about the AI-based production. She knows, tells me she shouldn’t like it…and keeps watching, because it’s aimed straight at her tastes.
I can’t claim immunity either. The Wheel of Time is my favorite book series, and Amazon cut the adaptation short (right after its best season…*shakes fist*). If someone built a scene-by-scene walk through the actual books via AI…not even top-notch, just visually interesting…I know I’d be drawn in (setting aside the very real IP hurdles, since any legitimate version would need the rights holders on board.)
That points at the stranger frontier, AI content for projects no studio would realistically greenlight, or even content generated for an audience of exactly one (the token requirements are cost-prohibitive currently, but within the next couple of years…?)
The slop aimed at my 3-year-old
The adult version is at least a choice…the kids’ version gets decided by an algorithm.
YouTube earns more watch time than any other streaming platform, and when my 3-year-old daughter watches, the recommendations drift toward content that sets off my radar. Basic animation, repetitive sounds, endless variations of cartoon cars dropping down slides (some videos on that channel have half a billion views!). Nothing inappropriate, no bad language, and still my gut reaction was immediate…this is AI slop, and my daughter shouldn’t be consuming it.
The data says that radar is calibrated. A February New York Times investigation found YouTube’s algorithm routinely serving toddlers AI-generated video, a Kapwing analysis found ~21% of Shorts shown to new users contain AI content (with the kids’ category among the densest), and AI-dedicated channels are pulling an estimated 63 billion views a year (YouTube’s own CEO has named managing slop a 2026 priority).
Attention is the most valuable commodity in the world (yes, more so than time), and this material is engineered to harvest it by the hour.
A single pattern runs underneath all of it, one I’ve started calling the Great Migration to the Unprintable…value flees whatever AI can print (content, code, first drafts, entry-level output) and concentrates in what it can’t (earned judgment, verifiable trust…and the dirt we underwrite). The aforementioned wedding (with REAL LIVE people) itself sits on the unprintable side of that ledger, too.
The gap is wider than you’d think (and it pays)
On the business side, I recently started an AI diagnostic for the marketing arm of a multi-billion dollar multifamily firm. Sharp people, real brand equity, and six figures of annual recurring revenue from a mastermind they launched THIS year.
They’re running essentially zero AI workflows, with no data attribution at all, succeeding in spite of the absence, which to me is the exciting part, with the upside sitting right there, un-harvested.
There’s no forcing function (yet) for most teams to adapt, and fully-optimized AI implementation is likely still years from being table stakes in many industries. For the operators who can actually deploy these workflows (and validate them…the Check column from the framework two weeks back, which never leaves human hands), that gap is the opportunity.
(Admittedly I’m biased, since I now sell AI implementation services…which is exactly why I can say plainly the demand is real and growing.)
The two kings keep cutting prices
The two frontier labs (Anthropic and OpenAI) are at each other’s throats (add in the constant creep of the Chinese open source models as well), to the benefit of everyone using the tools. Naval Ravikant has made the point that a single AI king would mean monopoly pricing…two kings means a knife fight on cost and capability.
Anthropic’s Fable model (released in June, powering our recent SLC Deal Engine breakthroughs) set the bar. OpenAI answered on July 9 with GPT-5.6 at roughly half Fable’s wholesale price (AI usage is billed per token, roughly 3/4 of a word, metered separately for text in and text out), and independent benchmarks put it at comparable intelligence for ~a third of the cost per task. Fable still leads on the hardest coding work, but the pricing pressure is doing its job.
OpenAI’s cheekiest move hit my own desktop last week. ChatGPT offered to import my Claude projects, Cowork chats, and Claude Code history directly…a raid on the walled-garden lock-in (the strategy behind Apple’s decades-long march toward market cap dominance) aimed at users like me, with mountains of data (and projects) built inside Claude. The import wasn’t perfect when I inspected it, but the play is brilliant, and it forces Anthropic to keep earning my seat.
Image generation is running the same race (the cartoons in this very newsletter sharpen week over week…produced via ChatGPT, which is currently in the lead for images). Which brings everything full circle to my cousins’ sons, because the tools eating routine animation jobs are the same ones handing any individual creator a studio’s toolkit.
We’re all in the experiment
I don’t have a clean answer on where this lands. Candidly, I don’t see how we avoid a widening gap between the haves and have-nots without significantly more governance or safety nets than anything currently on the table…and I’m watching it reach my family (more fortunate than most), which tells me how broadly it will be felt.
What I do know…we’re living in a real-time experiment, all of us as data points, and the variables you actually control are your skills, your judgment, the risk you’re willing to incur, and whether AI is paired across those areas. My younger cousin picked his variables well. So can the rest of us.
Real connection, earned judgment, verifiable trust…the unprintable side of the ledger doesn’t depreciate.
=====
If you’re an experienced operator with routine deal flow looking for a capital partner that runs every deal like a controlled experiment (data in, earned judgment validating every output, downside protected before the first dollar moves), let’s talk. We write checks from $50K+, we close 100% of the deals we commit to, and our national underwriting sits firmly on the unprintable side of the ledger.
