A federal judge in Ohio ruled on Friday that J.M. Smucker can keep suing Trader Joe’s over the shape of a frozen sandwich, and the reason given matters more than the case: twenty years of resources spent made the customer associate the design with the brand. The shape became an asset because someone paid for it for two decades. The whole day turns on that math — what it costs to stop renting and start owning. Wiz decided to buy leads at R$25–30 so it would no longer be “the middleman’s middleman” in consórcio; Belron wants a valuation above €30bn for owning 3,500 units in 40 countries; a16z expanded its growth fund to $8.5bn because, in the stated thesis, an AI startup reaches the growth stage faster and burns more capital. None of the three is buying technology. All of them are buying the relationship with whoever pays.
The counterweight came from the two items that measured before asserting. The paper reviewed by Alpha Architect shows GPT-4 getting the direction of the immediate reaction to corporate headlines right in 93.3% of cases — and the strategy’s Sharpe falling from 6.54 in the fourth quarter of 2021 to 1.22 in the first five months of 2024, as the tool stopped being rare. Baymard graded seven telecom sites across 3,100 scores and didn’t find a single one above “mediocre”. On one side of the day is what accumulates and is therefore defended; on the other, what reproduces with a prompt and therefore decays. The useful question for anyone building is which of the two sides their own product is on.
launches
Behind the build: Generative plugins and shaders at Figma — today’s update isn’t the animated shader: it’s the decision to let what the agent generated be published to Figma Community (or privately inside the organization, on Organization and Enterprise plans), and to allow downloading the code to edit it in a third-party agent. In other words, Figma bet on distribution instead of lock-in — what the user creates can leave, but the channel where everyone else will find it is still theirs. The account of the process has the most usable detail: vibe-coded prototypes worked for designers to challenge their own assumptions, but didn’t work as a reference for developers, PMs and stakeholders, so the team went back to static mocks with every state exploded into view to get aligned, and only then moved to a coded prototype. Two months to Config, fifteen people at the start, no PRD and no crit.
product and growth
Wiz cuts out middlemen to make consórcio weigh more on the balance sheet — Wiz brought origination and quota sales in-house, set up a 23-person telesales operation in São José dos Campos and started buying leads, keeping the partner channel close by. The unit economics are stated in the piece and are worth the exercise: a mature operation converts around 3% of leads, and the math only works with cost per lead between R$25 and R$30 — which gives, in simple arithmetic, a CAC between R$833 and R$1,000 per quota sold. The accounting detail is the most instructive: in direct sales revenue is recognized at signing, but the administrator’s commission comes in over ten to twelve months, so the vertical will show up in results before it shows up in cash. This lands in a bad quarter (segment net revenue at R$32.4m, down 15.4%, and Ebitda margin from 61.6% to 47.9%), which is the classic moment when someone decides to buy the end of the chain instead of going on renting it.
Telco UX Benchmark: 3,100+ performance scores — Baymard manually graded seven carriers (AT&T, Xfinity, Lebara, Orange, Vodafone UK, Bell and Verizon) against 500+ parameters, and only AT&T escaped “poor”. The mistakes are the same in all of them: proprietary plan names and technical specs shown as labels with no explanation, contract length missing from the listing, inconsistent units across devices, 4G and 5G coverage on separate maps. The read for anyone building isn’t about telecom: when an entire category fails at the same point, the company’s internal vocabulary leaking into the interface has stopped being a bug and become the sector standard — and that’s exactly why nobody fixes it.
brand and ip
The Battle For the Square-Circle: Trader Joe’s Fails to Dismiss Smucker’s Uncrustables Lawsuit — judge John Adams denied the dismissal of the suit filed in October 2025 and also refused to transfer the case to California, writing that Smucker devoted significant resources over twenty years to building and protecting the Uncrustables brand. Trader Joe’s defense is a quick course in positioning inside out: it claims its sandwich isn’t round but a “squircle”, a square with rounded corners, and that the competitor’s current packaging, launched in 2024, is too new to support federal trademark dilution. What you learn is about the nature of the asset: trade dress doesn’t protect the shape, it protects the accumulated association with it — and association is the only moat the competition can’t reproduce inside the same quarter. Smucker is asking for sales to stop, plus restitution, fees and the handover of product and packaging for destruction.
media and attention
The Telegraph’s next podcast play: video, events and paid memberships — the Digiday piece came through truncated in the dossier, so I’ll stick to what the headline announces: newly bought by Axel Springer, the Telegraph wants to turn podcasts into a repeatable franchise that runs across video, live events and paid subscriptions, with a video-first approach. The numbers cited up top: podcast audience 161% higher year over year and an average of 2.4 million weekly views on YouTube across the three daily shows (“The Daily T”, “Ukraine: The Latest” and “Iran: The Latest”). The stated reason is the same as everyone’s now — search and social no longer deliver reliable traffic, so the way out is converting borrowed attention into a direct relationship before the platform changes the algorithm again.
market and capital
Carlos Brito prepares Carglass’s owner for an IPO of more than €30bn — Belron (Autoglass, Carglass, Safelite) posted €6.7bn in revenue in 2025 serving more than 17 million vehicles, and is weighing a listing in Amsterdam next year aiming above €30bn, according to Bloomberg. That’s more than four times revenue for a windshield replacement business, and the story holding up the multiple isn’t growth — it’s operating margin going from 21.2% to more than 25% by 2028, with revenue growing at mid-to-high single digits. Brito, who quintupled AB InBev’s value, was hired in 2023 to do in a service company what he did in a beverage one. The Brazilian detail is a reminder of how much an option is worth before the thesis becomes obvious: in 2015 Belron sold 60% of Carglass Brasil for R$4.8m and has since carried the operation under the equity method.
a16z expands growth fund to $8.5bn days after launching a new $1.1bn fund — the fifth growth fund grew by $1.75bn on top of the $6.75bn raised in January, and comes right after the Machine Age Fund, $1.1bn dedicated to AI hardware — chips, memory, networking and storage. David George’s rationale is the part that matters: in the AI era companies reach the growth stage faster, consume more capital and at higher valuations than in previous cycles. If the thesis is right, the practical consequence for anyone building is that the capital range that defined “growth stage” has moved up a level, and getting there sooner doesn’t mean needing less money — it means needing more, earlier.
Can ChatGPT Forecast Stock Price Movements? — Alpha Architect reviews the Lopez-Lira and Tang paper (Journal of Financial Economics, 2026), which gave GPT-4 only the company name and the headline across 159,137 observations from 4,123 US companies, in a period after the model’s knowledge cutoff. The result worth stealing outside of finance: a 93.3% hit rate on the direction of the immediate reaction, and even so almost none of it is capturable, because the move happens while the news is arriving. What’s left is the one-to-two-day drift, 34 basis points a day before costs, which turns into a loss at 20 bps of round-trip cost — and whose Sharpe collapsed from 6.54 to 1.22 as the tool went mainstream. The lesson for product is direct: forecast accuracy and capturable value are two different things, and any AI feature sold as “predicts X” has to answer where exactly the user gets to act on the prediction.
reading
#431 How Henry Singleton Worked (Founders, David Senra; today’s feed carried the episode without a url) — the episode reads Distant Force, George Roberts’s memoir about Teledyne, alongside The Outsiders, by William Thorndike (2012), a book that practically came out of this case. The central fact is the same as the rest of the dossier: over the years Singleton bought back more than 90% of Teledyne’s shares, because the cheapest asset he could find to allocate capital to was the company itself. Two lines explain the method — “our accounting is set up to maximize cash, not reported earnings”, and, on the buyback fashion among the Fortune 500, “if everyone is doing it, there must be something wrong with it”. For anyone building product the transplant is this: the metric you choose to optimize defines what the company becomes, and Singleton preferred quarterly earnings to wobble over distorting the operation to please an analyst.
the rest in one line
The wave of copycat suits in CPG continues: Mondelez against Aldi over packaging similarity last year, and Buc-ee’s against Nut Huggers in November 2025 — aggressiveness that earned the chain a John Oliver parody on Last Week Tonight, the “Buc-off” brand, with an explicit invitation to be sued.
stalled sources
First Round Review (310 days), Gurwinder (248), Calculated Risk (233), Matthew Ball (198), Elad Gil (135) and Kwokchain (132) still haven’t published. Among those that normally show up every week, missing are Naval (62), Investment Idiocy (61), Benedict Evans (55) and Anti-Mimetic (49).