Six items and one unpaid invoice. This week everybody announced a transformation and left the money exactly where it was. The WFA rebuilt how agencies get paid and filed rebates under "separate issue." Infillion attacked walled gardens while laying bricks for its own. The Trade Desk cut 575 people and then asked shareholders to reset the options of the people who stayed. Dentsu found a new way to sit on both sides of the table. Everybody moved the furniture. Nobody moved the money.

Programming note: Sukkos begins tonight at sundown. Your correspondent will spend the holiday eating in a temporary structure with a leaky roof in the Arizona heat, which still puts it ahead of most ad tech business models. Offline through Sunday night. Gut Yom Tov.

The World Federation of Advertisers says labor-based agency compensation has collapsed to 17% of the market, down from 54% fifteen years ago. Fixed fee and output models now lead at 35%, and labor-plus-performance deals have more than doubled to 23%. The WFA's Laura Forcetti says clients care about "the quality, impact and performance of the work," not headcount or hours, and that AI is making time "an increasingly weak proxy for value." Agency relationships now last 4.3 years, up from 2.0 in 2018. The study, done with Agency Media Solutions, was a voluntary online survey fielded in May with representatives from 69 multinationals. It does not say how many people answered.

Then the report reaches rebates and calls them "a separate issue." 63% of marketers do not adjust agency pay against media volume rebates, up from 60% in 2022. Among those who do, 25% say rebates offset less than 10% of fees, and only 7% report an offset between 10% and 30%. So the industry spent fifteen years rebuilding the fee around value while the money agencies collect from the other side of the trade stayed off the scorecard entirely. Last week ADOTAT asked six holding companies how big their principal media and non-product income actually is. None of them answered. Media planning is still 41% labor-based, by the way, so the hours haven't died either. You can't pay your agency for value and let the seller pay it too. That isn't a separate issue. That's the issue.

Infillion bought Foursquare last Friday, terms undisclosed, and CEO Rob Emrich says the company will stay an "active acquirer." Understatement. The rollup already includes TrueX, Gimbal, Fysical, InStadium, UberMedia, Drawbridge's media business, MediaMath out of bankruptcy for $22 million, and in February, retail purchase data company Catalina. Foursquare brings 100 million points of interest, data from 250 million U.S. devices and 16 billion human-verified check-ins, with Apple Maps, Uber, Snapchat, Spotify and every major holding company as customers. Emrich's pitch: "Walled gardens and retailer apps can only measure what happens inside their own environments."

Now look at what he's assembling. Catalina knows what you bought. Foursquare knows where you went. MediaMath buys the ad. One company sells the media, targets it and grades whether it worked, which is the Roku homework problem from last week with a shopping cart attached. Foursquare keeps selling to outside customers "for now," and Infillion made the same assurance about Catalina while also telling AdExchanger that Catalina data would eventually be available only through its own platform. That is the textbook definition of a walled garden. Meanwhile nobody has said how store visits and purchase records get joined, at what level, or under whose consent, the exact questions the FTC asked in its location data cases against X-Mode and InMarket. And the check-ins that make Foursquare's panel worth buying come from Swarm, the app Infillion says it might open-source.

This month The Trade Desk cut 15% of its staff, about 575 people, with a memo calling its people its greatest asset. This week it filed for a special shareholder meeting to reprice more than 15 million stock options. The backdrop, courtesy of AdExchanger: the stock is down 90% from its high seven quarters ago, the company is about to be dropped from the S&P 500, and nearly the entire C-suite has turned over in a year. Independent analyst Richard Kramer, a longtime bear, says Wall Street treated TTD's campaign revenue like SaaS recurring revenue until the revenue declined and the music stopped, and he's worried about institutional knowledge walking out the door.

A repricing is a do-over for underwater options. Shareholders who bought at the top don't get one. This is the same company that spent $242 million on buybacks in the first half of 2026, buying its own stock at prices it now apparently considers too high to motivate the people holding options. The layoffs cut the cost base. The repricing hands the survivors a fresh strike price. Before anyone votes, shareholders should ask two things: which option holders get the reset, and whether any of the 575 people shown the door had options worth resetting.

IAB Europe surveyed 50 ad executives across 44 European markets this spring and summer, and Digiday reports that 58% expect agentic ad buying to reach "operational use or scale" within a year. Company size is the whole story: 86% of respondents at firms with more than 500 staff already run their most advanced agentic systems at a level where people and agents plan, delegate and execute together. At smaller firms, 48%.

Fifty people across 44 markets is barely more than one per country. And "operational use or scale" puts a pilot and an industry in the same checkbox. The actual number of the week came from Brian O'Kelley's Scope3, now renamed Apostra and pivoting to agent-to-agent ad buying, which told Adweek its agentic business has managed $1.2 million in media spend over the last three months. Credit where due: Apostra is the only company in this conversation that published a dollar figure, and it's an honest early-stage number. It is also a rounding error on a single brand's CTV flight. The survey measures belief. Apostra measured money. Believe the one with the invoice.

Dentsu signed on as the primary agent in the Americas for brands that want to tie into Kodansha's anime catalogue, including Attack on Titan, Ghost in the Shell and Blue Lock, covering promos, tie-ins, advertising, events and experiential. Dentsu's Caroline Hughes says the deal is about "offering our clients access" to culturally embedded IP. The deal builds on what MediaPost calls Dentsu's already dominant position in anime marketing inventory, complete with its own proprietary research, data and identity platform. The anime market is projected at $60 billion by 2030, with the U.S. at nearly $3 billion today.

Read the word "agent." Dentsu represents the rights holder selling the IP. Dentsu also represents the brands buying it. Dentsu owns the research that says which anime works and the identity platform that measures whether it did. That's principal media in a cel-shaded coat. It connects straight back to item one: when the agency gets paid by both sides and 63% of marketers never net it out, the client should ask whether Dentsu earns anything from Kodansha on deals it recommends, and whether that shows up in the client contract. "Access" is what the seller's agent calls it. The buyer's agent would call it a markup.

Google is sending its Tensor Processing Units into orbit on SpaceX's Transporter-18 rideshare October 1, with Planet Labs, to test whether the chips can survive launch forces, radiation and extreme temperatures. It's the next step in Project Suncatcher, which Google announced in September 2025 with prototypes planned for early 2027, so it's running ahead of schedule. MediaPost also reports an FCC update on September 18 accepting the initial phase of SpaceX's plan for the first 100,000 "Starmind" AI computing satellites, and says the technology lays the groundwork for hyper-localized ad delivery and lower infrastructure costs by the mid-2030s.

Everything in the actual mission is about whether the hardware survives. Space radiation causes bit flips, data errors from particles hitting the chip. Ad tech finally has a discrepancy excuse that isn't the SSP's fault: cosmic rays. The industry can't reconcile a CTV impression between two companies on the same planet, and it's already pitching hyper-local delivery from orbit, a decade out, resting on a test that hasn't launched yet. Ask about the hyper-local ads after the chips come back working.

Paramount settled with 12 state attorneys general who'd tried to block its $81 billion merger with Warner Bros. Discovery, clearing the way for the deal. Meta's Muse agent is pulling heavy downloads, and OpenAI is reportedly building rival features. Mozilla papered New York with posters skewering big tech executives, the rare ad campaign aimed at the people who own the ad business. And IKEA celebrated 30 years of its blue Frakta bag with Ogilvy Italy and street photographer Salvatore Matarazzo: ten snapshots of the bag as a cat bed, a toddler's paddling pool and everything in between. The only campaign this week where the product demonstrably does what the ad says. It's a bag.

Six stories, one pattern: the transformation is real and the money didn't move. The WFA changed the fee and kept the rebate. Infillion attacked the wall and started building one. The Trade Desk cut the people and reset the options. Dentsu found the talent and represents both sides. The agentic market forecast scale and reported $1.2 million. And the ads are headed to space, pending a radiation test. Everybody's reinventing. Check who's still getting paid.

How we reported this: items are drawn from trade coverage published September 18 to 24, including MediaPost, AdExchanger, Adweek, Digiday and Creative Bloq; the WFA and IAB Europe figures are as reported by those outlets and not independently obtained, and neither study discloses a respondent count large enough to generalize from; The Trade Desk repricing figure comes from its SEC special meeting filing as reported by Adweek. Disclosures: Infillion owes the author $500 on an unpaid invoice; Brian O'Kelley is a regular ADOTAT source. Any reply from a named company will run as a dated update.

ADOTAT+ members got the File this week. You got the headline.

Here's what's on the other side. The company that announced a "new" pause ad pipe on Tuesday, and the public document from three months ago that says otherwise. The one number in the industry's favorite pause ad study that never makes it into the pitch deck, and why it guts the "high attention" claim. The headline statistic every trade outlet repeated this week that we could not find a single published method for. The "preference" finding that turns out to come from a very friendly room. The case study with a triple-digit lift and no base underneath it.

Then the part your SSP rep won't bring up at lunch: how the same pause moment can show up in your bidder more than once, with different fees each time. A claim-by-claim scorecard you can forward to your buying team. And the ten questions we're putting to Nexxen, which work just as well on any seller who uses the word "proven" in your next meeting.

Your competitors on the buy side read it this morning. Join ADOTAT+ and read it before your next CTV plan goes out the door.