The Adjacent Brief

TL;DR: Meta's new AI Muse image tool auto-enrolls public Instagram photos for AI training by default, continuing the platform's opt-out-first data collection approach. Enterprise AI adoption is running into a separate friction: companies are deploying "AI champions" to convert internal skeptics. And Wired reports the consumer subscription model that funded the first wave of AI products is under real pressure.

Worth Reading

• Suspecting AI cheating, Brown professor ordered an in-person final — scores dropped 50% — The gap between AI-assisted performance and actual comprehension is now measurable in a single semester.
• Argentum sees the capital stack, not silicon or power, as the missing layer in AI infrastructure — Financing structure, not engineering, is the real bottleneck in data center deployment.
• Agentic inference is pushing storage to center stage, and Solidigm is positioning accordingly — When AI workloads shift from training to inference, the infrastructure constraint moves with them.
• Meta patents an AI device that tracks emotions and monitors medication intake — Pair this with the always-on smart glasses story below; the product vision is becoming readable.
• YouTube overtakes Spotify as the UK's top podcast platform — Audio is winning on video rails; Spotify's distribution moat is leaking.
• AI search is eroding SEO's ability to deliver local and geographic outcomes — Local search was SEO's last durable stronghold; that claim is now in question.
• At Cannes, creator content has moved from rough-and-raw to primetime programming — WPP's Adam Shlachter on a format shift that's restructuring content budgets.

Brand & Growth

The AI champion playbook is an admission that the tool isn't selling itself

BCG's finding that 74% of front-line employees now use AI regularly (paywall), up from 51% a year ago, looks like a success story. Read the mechanism and it's more complicated. Companies are deploying organized internal networks of "AI champions" — enthusiasts tasked with converting skeptics — because top-down mandates haven't moved behavior on their own. That's a change management problem dressed in a technology frame. When a tool requires a dedicated conversion function to reach two-thirds adoption, the friction is real and the remaining third is the hard third.

Recruiters are becoming the cautionary tale they used to place

Recruitment firms are adding AI specialist hiring to their service lines to offset the revenue they're losing as AI automates the sourcing and screening work that was their core product. The firms that survive will be the ones that understood the shift before their clients did. Most won't. This is the same displacement pattern that hit travel agents and financial advisors, and the timeline is compressing in the same way.

Meta's opt-out default is the product strategy

Meta's AI Muse image tool uses public Instagram photos for training by default, with opt-out available but not foregrounded. This is not an oversight. Meta's competitive position in generative image models depends on training data volume, and its 2 billion-user platform is the data asset. The regulatory question is whether European data authorities treat this as a consent violation under GDPR; the business question is whether users notice or care enough to opt out in meaningful numbers. History says most won't.

Connected World

Power transformer lead times are the physical ceiling on AI's ambitions

The data center buildout has a mundane chokepoint that no amount of capital can immediately solve: average lead times for power transformers now stretch into years (paywall), up from months. Transformers are custom-manufactured, capacity-constrained, and not something you can fab on a new production line in a quarter. Every data center announcement made today is implicitly a 2028-or-later story. The companies that locked long-term transformer supply contracts 18 months ago made a bet that is now paying off.

Die casting comes home, and the garage is the proof of concept

Hobbyist-scale die casting equipment is becoming accessible in a way that was implausible five years ago. This matters less as a consumer story and more as a leading indicator: when manufacturing processes that required factory infrastructure migrate to garage scale, the next wave is small-batch industrial adoption. The same pattern played out with CNC machining, laser cutting, and PCB fabrication. Watch for the first serious product companies built entirely on sub-$50K casting setups.

Always-on glasses are a product decision Meta hasn't finished making

Meta is reportedly developing smart glasses with continuous recording capability — ambient capture rather than user-triggered clips. The technical challenge is manageable. The product challenge is whether anyone within ten feet of the wearer consents to being recorded, and whether Meta wants to own that liability. The emotion-tracking and medication-monitoring patents noted in Worth Reading above suggest the underlying product vision is expansive. The question is which pieces Meta ships first and in what order.

Culture & Signal

The right to repair wins a round, and the economics explain why it's lasting

A significant right-to-repair ruling came down this week, extending a pattern of legislative and regulatory movement that manufacturers spent years containing. The durable force behind right-to-repair is inflation, not activist pressure. When replacement costs are high and consumers are stretched, the political economy of repair-friendliness changes. Manufacturers that treat repairability as a brand attribute rather than a threat are getting ahead of a posture that regulation will eventually impose anyway.

Madison Square Garden's surveillance database is a preview of what ambient data collection looks like at scale

MSG tracked hundreds of celebrities using a database that categorized attendees by sexual orientation, exclusion status, and other personal attributes. This is the facial recognition plus CRM combination that privacy advocates have been describing as a theoretical risk for years. MSG ran it at one of the highest-profile venues in the world. The gap between what is technically possible and what is legally prohibited is still very wide, and most venues have neither MSG's legal resources nor its notoriety. Expect quieter versions of this to surface as facial recognition hardware costs continue to fall.

IAB's David Cohen calls the AI disruption of media structural, not cyclical

IAB's chief executive argued that AI is doing to the internet what the internet did to traditional media — redistributing attention and ad revenue away from incumbents toward the new infrastructure layer. Cohen's framing is notable because IAB represents the publishers and platforms most exposed to this shift. When the trade body's own leader says the disruption is structural and not cyclical, the member companies still treating it as a search-traffic optimization problem are misreading the brief.

The New Consumer

A Brown class averaged 96 on the take-home exam and 48 in person

A Brown University professor suspected AI-assisted cheating after a take-home midterm averaged 96%, then administered an in-person final that averaged 48.6%. The 47-point gap reveals a capability measurement story. The take-home score measured what AI can do; the in-person score measured what students retained. Every institution running online or take-home assessment is now sitting on a version of this data they haven't looked at yet. The Ars Technica follow-up in Worth Reading gets at the harder question: if students are optimizing for the credential and not the comprehension, the assessment format is the problem, not just the behavior.

A $185 device is selling the idea that your phone is the problem

The Light Phone-adjacent $185 device positioned as a "modern Palm Pilot" functions as a market signal rather than a mass market product. The consumer willing to pay $185 to reduce phone dependency is the same consumer buying analog notebooks, dumb watches, and screen-time management subscriptions. The category is real and growing, but the business model hasn't found its form yet — because the people who most need less screen time are the least likely to seek out a hardware solution.

Parents are worried about AI dependency, but survey data only goes so far

Half of parents in a new survey say they worry their children rely on AI too much. This is aspiration data, not behavior data — it measures concern, not action. The Brown University story above is what the behavioral version looks like. The parental anxiety is real, but the children in question are not going to be convinced by concern; they'll be shaped by whether institutions — schools, employers, colleges — design assessments and evaluations that require demonstrated capability rather than output alone.

Commerce Rewired

Consumer AI subscriptions are losing their subsidy

Wired's model behavior column argues that the golden era of AI subscriptions is ending. The underlying logic: labs priced consumer tiers below cost to drive adoption during the novelty phase. As behavioral novelty fades and usage patterns normalize, the subsidized price-to-value ratio gets harder to sustain. The labs that built durable enterprise contracts during this window are positioned differently from those that built large consumer subscriber bases on subsidized compute. Enterprise contracts justify themselves against workflow ROI; consumer subscriptions justify themselves against perceived value, which is a softer foundation.

Pricing platforms are expanding because pure pricing is a thin moat

Forrester's analysis of pricing platform convergence describes vendors absorbing adjacent retail functions — inventory optimization, promotion management, demand forecasting — into what were pure-play pricing tools. This is a standard software consolidation move: the pricing vendor that also handles adjacent workflows is stickier than one that handles pricing alone. For retailers evaluating pricing tech, the buy-vs-build calculus is shifting. The vendor pitching "we'll replace your pricing tool" is now often pitching "we'll replace four tools." That's a different procurement conversation.

14 articles across 5 themes · 13 sources · Powered by Folo + Claude