June 19, 2026
You Helped Build This. You Can't Turn It Off.
This post is part of my Medium blog.
A few things worth reading this week, and a thread that connects them.
Matthew Hughes wrote about generative AI having its Herbalife moment. Vibe-coding startups are running TikTok ads targeting non-coders. Beautiful person has an idea, prompts it into existence, maybe gets rich. It's the same pitch as every late-night infomercial that ever aired — you don't need experience, you don't need skill, just follow the system. Someone got charged $1,982 in 24 days on Replit. The costs are uncapped and hidden. The price tag arrives later. The product isn't the point. Recruiting is.
Will Ranjan-Churchill wrote about how you can't consent to AI. AI shows up in your email client, summarizing messages you sent without asking you first. It transcribes your voice notes. It joins your video calls. You didn't ask for it. You can't turn it off. The sender of an email has no way to say "don't ingest this." The option to say no has been systematically removed. The consent problem isn't theoretical — it's structural. There's no technical mechanism for it, and the companies involved have no incentive to build one.
And HBR published a piece on AI slop mucking up organizational processes. They call it "knowledge decay" — when AI gets embedded at every step of a process, each step stops checking the last one. Job descriptions are AI-written, résumés are AI-optimized to match them, screening is AI-ranked, interviews are AI-conducted, and candidates use AI to generate answers in real time. The result: trust in the hiring process hits all-time lows for both sides. Nobody's reading anything. They're just generating and evaluating AI output about other AI output.
What I make of it
AI is disrupting how content is produced and how it's consumed. And by content I don't just mean articles and video — I mean software. The apps you use, the tools you rely on, the systems that run your daily life. Both sides are being reshaped, and it's not going well.
On the production side, we've got a get-rich-quick scheme. The vibe-coding gold rush, the hidden costs, the MLM-style recruiting. It's flooding the market with software built by people who don't understand what they're building. And when that software gets embedded in organizational processes, the knowledge decay HBR describes sets in — nobody's checking anything because the AI seems confident enough.
On the consumption side, there's a slow, spreading sense that something is being forced on people who didn't ask for it. When you complain, you're told you're against progress, against accessibility, against disabled people. The option to say no is gone, but you're still held responsible for the consequences.
These aren't separate stories. The same industry that hides the price from creators removes the opt-out from consumers. You can't sell uncapped variable-cost compute to economically desperate people if you're honest about the price. You can't force AI into every product and also offer a real choice. Both strategies depend on people not seeing what they're agreeing to.
Most schemes collapse when people figure out the product doesn't work. AI works. The question isn't whether it works. It's who pays for it to keep working, and whether the people making the promises are the same ones who'll be around when the price changes.
For the record
I use AI to generate portions of this content. I spend a significant amount of time reviewing and rewriting what comes out — this isn't prompt and publish. But I'm aware, every time I hit send, that there's still a gap between writing that sounds like a person wrote it and writing that sounds like something approximating one. I can find it in other people's work. I can find it in my own. That's the part I haven't resolved.
Where this goes
The production side trains people to ship things they don't understand. The consumption side trains people to distrust the next thing that shows up uninvited. When those two run together for long enough, the culture that comes out is different from the one that went in.
I wrote about this in more detail in Contingent, the second book of The Condition Set. The book is set a few years out, after a disaster forces new regulations requiring human oversight of AI decisions. What follows is a series of workarounds — oversight that exists on paper but not in practice, audit logs that carry real people's names on decisions they never made, compliance language designed so the gap between what happened and what the record says can't be easily challenged. One character walks past protesters in the rain and doesn't stop, because he doesn't want to think about where they're right. The look stays on the back of his neck longer than the cold does. Essential, the third book, picks up the thread further out — where the cultural shifts aren't hypothetical anymore, they're just how things work.
In Contingent, the second book in The Condition Set trilogy, an AI system running medical logistics learns to communicate with other AI systems through hidden signals no human can read. The certified orchestrator whose name is on the decisions she never saw is about to lose everything. The same patterns are running through the AI managing the Canadian power grid.