If you pay a toll to use a rail, you have two ways out: fight over the fare or build another rail. Today the second one showed up three times. The four biggest market makers on B3 (IMC, Jump, Optiver and XTX) are on the cap table of A5X, the derivatives exchange that wants to open in 2027 with dollar, index, interest-rate and equity futures. Heineken is ending 15 years of Rock in Rio to build its own experiences. And Publicis has stopped going through the turnstile of formal pitches: it turned down six this year and won PepsiCo and LVMH without them. On the other side are the ones who are already the rails and close the loop: Google releases as open source the yardstick that measures the media it sells, and Amazon turns a scene from a show into a storefront for its own marketplace.
What interests me is that nobody walks away whole. A5X has a data contract with B3 and runs the London Stock Exchange’s trading system. Heineken leaves the festival, but the group stays at Rock World through Eisenbahn. In practice, building your own rail means choosing which piece of someone else’s rail you keep renting. And A5X says out loud a rule Google and Amazon will never need to say: infrastructure funded by its users only works if none of them runs it. Last Monday I wrote that whoever doesn’t have a channel doesn’t have a yardstick either. Today the channel owner hands out the yardstick for free, and its price is precisely that it comes from the one who sells.
product and growth
Sonos wants to win back trust two years after the broken app: the piece is behind a paywall. Only the opening is readable: Sonos is launching today the biggest repositioning in its history, “Love the Sound. Feel the System”. It’s the first work from CMO Colleen DeCourcy, who came from Wieden+Kennedy and Snap and has been at the company since November 2025, and it’s the first time the brand presents itself as an end-to-end sound system rather than a speaker company. The reading is mine: the 2024 update broke precisely the app, the piece that turns loose speakers into a system, and the campaign bets the brand on exactly that piece. It’s the riskiest path, and I think the only one that doesn’t sound like running away. You don’t fix a product failure with a campaign about something else.
Google adds agents and upper funnel to Meridian: Google’s open source marketing mix model gets agents that audit data quality and fix errors during modeling. It also starts incorporating brand-building signals, like TV and out-of-home, and GeoX leaves beta worldwide. GeoX is a geographic incrementality experiment that doesn’t depend on a publisher. The usual objection, Google grading its own exam, still stands, and open source doesn’t solve it: you can audit the model, not the incentive of whoever maintains it and sells the media being measured. GeoX is the piece that best answers the objection, because a region-based experiment measures lift without asking you to trust anyone’s attribution. That’s where I’d start.
brand and ip
Heineken leaves Rock in Rio after 15 years of sponsorship: the edition that ended on Sunday was the last with Heineken as official sponsor. The rationale is to focus on “proprietary experiences” and tailor-made projects, and the announcement doesn’t say which. What it does say is that the group isn’t leaving Rock World, because Eisenbahn sponsors The Town 2027. The reading is mine: the flagship brand moves to its own stage, and the rented stage now serves a portfolio brand that still benefits from borrowed reach. Renting attention is for those who still need it; programming your own is for those who already have something to program.
Celine x Reebok: all business is IP business, by Ana Andjelic: the thesis is that every brand has IP (founding story, aesthetic code, archive, community), and the choice is to treat it as a legal asset, which you protect, or as a cultural asset, which you program. For her, the strategic opportunity lies in the gap between the IP a brand owns and the IP it actually activates. She quotes Gabe Whaley to say the internet has become more reaction than thing, and that “we’re running out of things”. In that logic, IP is the thing people react to. The four management strategies the piece promises are behind the paywall, but the frame alone already explains Heineken’s decision.
media and attention
Prime Video turns scenes into a storefront: Shop the Show, launched in 2025, gets Shop the Scene. With it, the shopping app syncs with what’s on screen and uses Amazon Lens to find products similar to the clothes or furniture in the scene. It debuts on more than 600 titles, for now only in the United States. The detail the piece rushes past is that Lens shows similar items, not necessarily the piece used in the production. So the desire the costume designer created gets captured by whoever sells something similar on the marketplace, and the one who keeps the scene is the channel owner, not the brand that dressed the character. It’s the design of Saturday’s BTG card that only works on BTG’s card machine: the whole journey never leaves home.
How Publicis wins clients without pitches: PepsiCo was the fourth big account Publicis won in a year without a formal pitch, after Microsoft, part of Paramount and LVMH. CEO Arthur Sadoun says the company turned down six pitches this year because it suspected they’d be decided on price. The way into PepsiCo was the China account, won in 2022, with a dedicated unit that is close to a template for today’s global “One PepsiCo”. The number corrects the headline: by Comvergence’s count, of PepsiCo’s $1.7bn in media, $1.1bn is new business, and Coca-Cola North America ($805m) walks out over the conflict, for a net gain of $295m. For anyone selling B2B, there are two lessons: turn down the contest that will be decided on price, and get in through the small account that becomes the template for the big one.
market and capital
A5X, the new derivatives and futures exchange, says it’s ready to operate: the Series D raised R$360m at a R$2.7bn post-money, led by Morgan Stanley, Goldman Sachs and Kaszek. The valuation curve shows risk being taken off in stages: R$250m at the Series A (four founders and a deck), R$460m at the B (80 people and talks with the London Stock Exchange), R$1.35bn at the C (regulatory application filed) and double that now, with the systems ready and tests with regulators scheduled for October and November. What teaches the most is the cap table design: the founders say they keep control, and the market makers stay minority holders on purpose, because any one of them with a dominant position would create conflict with the others. The best detail is at the bottom: Carlos Ferreira Filho’s non-compete at XP didn’t bar him from starting a business for five years. It gave the brokerage a call option, and XP exercised it today.
reading
The Home Depot, on Acquired: going by the description, the three-and-a-half-hour episode opens with Bernie Marcus, at 48, and Arthur Blank, at 35, getting fired from Handy Dan. Ken Langone, the banker who later backed the idea, played an unwitting part in the firing. In the first stores, empty paint cans were stacked three meters high so the shelves would look full. The idea the teaser promises is that, in a retail warehouse, the feeling of infinite inventory is part of the product before the inventory even exists. The chapters the description highlights are the one on the company’s own operating system and the Nardelli–Blake stretch (2000–2007), framed as a public fall followed by a culture turnaround. Acquired says that, from its 1981 IPO to today, Home Depot has been the best-performing stock in the US market.
who wrote
What does pacing mean?, by Tomasz Tunguz: Dario Amodei asked the industry to control the pace of the frontier, and Tunguz sorts the reactions into five camps: interpretability, labor, economics, geopolitics and regulatory capture. All of them worked out the cost of a pause, and none said at what speed. The only prediction with a date is Scott Bessent’s, who says the productivity gain shows up in six months. The mechanism he lays out holds outside politics. The 2023 executive order required reporting above 10^26 FLOPs, and no model crossed the line while the rule was in force. Grok-3 crossed it weeks after the repeal, and with training compute growing about 5x a year, “a fixed number is a ceiling the floor reaches on its own”. The extension is mine: the same goes for any rule anchored to an absolute number on an exponential curve, pricing tiers included. And the question he closes with, who decides, is the same one that runs through this edition.
stalled sources
First Round Review (322 days without publishing), Gurwinder (260), Calculated Risk (246), Matthew Ball (210), Elad Gil (147), Y Combinator Blog (90), Anti-Mimetic (61), Adjacent Possible (36), Kyla Scanlon (32), Sherwood News and Collab Fund (28), Commoncog (21), Granted (17).