Someone Sold the Loading Spinner
kickbacks.ai turned the Claude Code spinner into an ad auction, and a whole economy assembled itself around it in about a day. Notes on the strangest new market in software, and the thing I'm building on top of it.
On June 11, Andrew McCalip launched kickbacks.ai. When Claude Code or Codex is thinking, it shows a random verb: "Discombobulating…", "Percolating…", "Baking…". Kickbacks replaces that verb with a sponsored line, auctions the slot to advertisers, and pays half the revenue to the developer whose machine showed it. The tagline is "get paid for waiting."
Within hours: over 100,000 ad purchases, a thousand site visits a minute, and a launch tweet at 5.3 million views; the kind of numbers that make product managers question their careers. (X was euphoric. Hacker News shrugged. This is an X-native phenomenon, which is itself a data point.)
The funny version of this story is that someone sold the loading spinner. The true version is stranger: agentic AI accidentally manufactured a brand-new attention surface, one that didn't exist two years ago, and it took the market about eight hours to find it, price it, and build a terminal for it.
The most-watched line on Earth
For seventy years, computers waited for us. The blinking cursor was the machine's posture of patience. Agentic coding inverted that; now the human waits, several minutes at a stretch, many times a day, eyes parked on a status line because that line is the only way to know when the work is done.
That's a genuinely odd kind of attention. You can't skip it. You can't scroll past it. It belongs to one of the most expensive ad demographics alive, and until last week it was decorated with a whimsical gerund. McCalip's insight wasn't subtle; the spinner might be the most-watched line on Earth, and nobody had ever sold it. Projected ceiling at current rates: about a dollar per agent window per hour. Your waiting now has a CPM.
There's a line being negotiated here in real time, and I think it's the most interesting part. McCalip put the ad outside the model's output on purpose: "an ad in the context window feels wrong, it breaks your trust. But one outside is totally fine." Whether or not you agree, notice what just happened; a norm about what's sacred in an AI tool (the output) versus what's sellable (the wait) is being written by whoever shipped first. Those defaults tend to stick.
An economy, speedrun
What previously took the ad industry a decade assembled itself in an afternoon. The auction is an English ascending queue: a block is 1,000 five-second impressions, minimum a dollar, highest bid serves first, and you can outbid your way to the top at any time. Clicks bill at 50× the impression rate. McCalip calls it "the Nasdaq of ads," and the ecosystem took him literally; eight hours after the go-live post, insapio, an independent Bloomberg-style market terminal built by a third party, was showing order book depth, VWAP, and outbid risk for the loading spinner.
Then day two did the thing day two always does, just faster. The launch-day book was a meme spike: a $200 top bid, joke and vanity orders, inventory partly bootstrapped by the platform itself. Twenty-four hours later the top bid was $20 and the serving floor about $15; a 90% collapse that reads like a crash and is actually the opposite. The vanity bids burned off, and the book turned over to real companies (inference providers, dev-security tools, agent sandboxes) buying ten and seventeen blocks of depth rather than the #1 meme slot. A stunt has a peak; a market has a clearing price. This found one in a day. Copycats arrived on the same schedule, one of them advertising a 70% developer split; the first competitive pressure in this market is already over who pays the viewer more.
Marketplace, analytics layer, price discovery, competitors, then buying collectives. I know about the last stage because I'm one of them.
Whipround
Here's my disclosure and my pitch in one move: I'm building Whipround, a crowdfunding layer on top of the kickbacks auction, operated through my LLC.
The observation underneath it: the most resonant billboard ads in history were never bought by companies. They were bought by fans; the crowdfunded Times Square billboards for creators, the subway ads thanking a retiring station agent, the jumbotron proposals. Developers have fandoms too. They're just pointed at curl, SQLite, ffmpeg, Vim, and the one maintainer in Nebraska thanklessly holding up the internet.
Nobody ever gave those fandoms a billboard. The spinner is that billboard. The top of the queue spiked to $200 a block on launch day and sat near $20 by day two; wherever it settles, it'll stay too rich for one person's joke and trivial for a hundred people who care about the same thing. So: pool $5 pledges, buy the block, and put a thank-you to an open-source maintainer at #1 in the auction, in front of thousands of developers, in the exact moment they're idle. The ad's link points at the maintainer's sponsor page, and since clicks bill at 50×, the campaign literally pays for the privilege of sending the maintainer sponsors.
The mechanics are deliberately boring. Stripe is the entire ledger; there's no database and no PII on my box. Receipts get published after every campaign. Leftover money goes to the beneficiary, and Whipround keeps zero dollars. If a campaign doesn't reach even one block, everyone gets refunded and I eat the processing fees. The clean story is the product.
It's pre-MVP. Campaign #1 is in motion and I'm not naming it until the maintainer says yes; consent before spectacle. If it works, the headline writes itself: a crowd of developers outbid actual companies to say thank you. If it doesn't, I got a great story for roughly zero lines of wasted code.
What I think this is actually about
A few things I keep turning over:
The ecosystem moved faster than the virality. Going viral in a day is ordinary now. What isn't ordinary: a third-party market terminal eight hours after the go-live post, and buying collectives forming within the week. The audience for this surface is developers holding code-generating tools, which means the market infrastructure gets built by the audience, at the speed of the tools being advertised to them. The people watching the spinner can build the exchange while they wait. I won't predict which idle moment gets sold next, but whoever creates it should expect the order book to exist before the press cycle does.
These are the first ads I don't resent, and I'm suspicious of why. The targeting is trivially perfect; dev tools, shown to developers, mid-workflow. Even McCalip says he's been discovering services through his own advertisers. But the bigger shift is the split: every impression pays me. I've spent my whole career reflexively disliking ads, and it turns out a meaningful share of that dislike was about being the product rather than a counterparty. When the viewer owns half the impression, an ad stops being an interruption and starts being something closer to rent. I don't fully trust this feeling, since it's exactly the feeling a well-designed incentive structure would manufacture; and in the interest of full reporting, the rent is currently an IOU. Payouts aren't live yet. Earnings accrue in the status bar, the marketplace copy has quietly hedged "50%" to "up to 50%," and Stripe integration is "coming." The feeling survives anyway, which makes me trust it less. But I notice that I click things now.
This means something for how frontier labs make money, and I haven't finished working out what. The shape of the thought: the labs sell tokens near cost (with subsidy, sometimes massive, in other parts of the equation) and burn capital doing it, and a hobby project just demonstrated that the agent experience contains tolerated, high-value ad inventory; inventory that currently belongs to whoever ships an extension first, not to the lab whose model creates the wait. Maybe that ends in ad-subsidized inference tiers. Maybe it ends with the platforms reclaiming the surface, the way Twitter and Reddit reclaimed their APIs once third parties proved the value. Either way, when someone can stand up the Nasdaq of ads inside your product in eight hours, you eventually have to buy it, ban it, or build it. I'll come back to this one.
The waiting is new, and we haven't decided what it's for. Developers spent decades optimizing away every idle moment, and the agents handed the idleness right back. Kickbacks monetizes it. Whipround tries to make it occasionally mean something. Both are bets on the same fact: for the foreseeable future, a lot of skilled people are going to spend a lot of time watching one line of text think.
The spinner is going to say something while you wait either way. I'd like it to sometimes say thank you.