The Adjacent Brief

TL;DR: Hyperscalers issued $155 billion in unsecured bonds through May — 45% more than all of 2025 — as debt markets funding AI infrastructure run well ahead of demonstrated returns. UK police were ordered to stop using AI for court documents, Meta's app served AI-generated news clickbait to users, and schools across the US pushed back on ed-tech adoption.

Worth Reading

• Google gives sites an AI search opt-out — but withholds the data that would make it useful — A choice without information amounts to liability management dressed as transparency.
• Microsoft's AI products aren't finding commercial traction — Copilot has the distribution but not the adoption. The gap between install base and daily active use is where the story lives.
• AI productivity gains are real — but slop is eating the surplus — Gary Marcus on why output volume and output quality aren't moving together.
• The renewals that blow up are the ones no one can see coming — When sales controls account access, customer success is flying blind at renewal time.
• The AI boom is becoming an entrepreneurship boom — Azeem Azhar on how falling model costs are shifting value creation downstream, toward application builders rather than foundation model providers.
• Google Chrome tests putting AI search results in front of organic links — The browser becomes the interface; publishers become the background layer.
• Steak 'n Shake's X account went full culture-war pundit — Brand voice as political positioning. Works until the boycott math doesn't.

Brand & Growth

The AI clone as professional facade

C-level executives and Harvard professors are now using trained AI digital twins to manage calendars and take meetings on their behalf (paywall), the New York Times reports. The pitch is efficiency. The operational reality is more complicated: when the person in the meeting isn't the person who agreed to the meeting, accountability for decisions made in that exchange becomes genuinely murky. An AI twin that schedules, responds, and commits on your behalf is a liability wrapper with your name on it.

Uber's internal experience surfaces this from a different angle. A piece in Nate's Substack on AI token cost management uses Uber's early AI budget blowout as the case study: the company burned its entire AI allocation ahead of schedule because no one had set usage guardrails before the tools went live. Uber's coding agents now produce roughly 1,800 code changes per week, with 95% of engineers using the tools monthly — but the early cost overrun is the detail worth sitting with. Adoption at scale without instrumentation produces a bill that arrives faster than the value does.

LinkedIn's post-cringe era is a product decision, not a culture shift

The New York Times piece on LinkedIn courting high-profile influencers frames this as a vibe change. It reads better as deliberate platform repositioning. LinkedIn has a business problem: it is the dominant professional network with nowhere left to grow in its traditional mode. Recruiting influencers to produce "post-cringe" content — personal storytelling, opinion, culture commentary — is a bid to increase session time and ad inventory. The risk is the one every platform faces when it widens the content aperture: the feed that made it useful for job searches becomes indistinguishable from every other social feed competing for the same attention.

Culture & Signal

The institutional veto on inconvenient science

The American Diabetes Association ejected five researchers from its annual conference for distributing copies of published journal articles — peer-reviewed papers, not self-published claims. The official grounds involved conference distribution rules. The papers in question reportedly challenged dietary guidelines aligned with major ADA sponsors. Whatever the procedural rationale, the optics are bad in a specific way: the conference floor became a space where published science required permission to circulate. For anyone tracking how institutional science is perceived by skeptical publics, this is the kind of story that does lasting damage to the credibility of the bodies meant to protect research integrity.

AI in the courtroom hits a hard wall

Several UK police forces have been ordered to stop using AI to prepare court statements (paywall) after concerns that inaccurate outputs could compromise legal proceedings, the Financial Times reports. This is a narrower and more consequential ruling than generic "AI makes mistakes" coverage suggests. Court statements are evidentiary. An AI-generated inaccuracy in a witness summary is grounds for case dismissal or wrongful conviction, not a customer service failure. The UK's move puts a floor under the liability question that other jurisdictions haven't yet drawn. Expect this to land in US legal proceedings within the year.

The screen-free school is a values statement, not a tech policy

Persuasion's piece on the anti-tech rebellion in schools documents a movement of parents and administrators pulling back from ed-tech tools, citing screen time, privacy, and doubts about pedagogical effectiveness. The telling detail is that the resistance is strongest in affluent communities whose kids have had the most exposure, rather than in low-income districts worried about device access. This is consistent with a pattern running through this arc for months: the families best positioned to adopt technology are increasingly the ones choosing not to.

The New Consumer

The feed gets worse on purpose

Meta's main app is surfacing AI-generated clickbait articles in users' feeds, the Verge reports. These aren't third-party posts that slipped through moderation — they appear to be content Meta's systems generated and served deliberately. The mechanism matters: if AI-generated content can be produced at near-zero marginal cost and engagement metrics are comparable to human-written posts, the incentive to fill the feed with it is nearly irresistible. The cost lands on users and advertisers, not on Meta. Brands running display inventory against this feed should be asking what content their ads are appearing next to, because the answer is changing faster than most brand safety tools can track.

Social attention is fragmenting below the platform layer

TechCrunch's look at the next generation of social apps beyond Instagram profiles platforms built around interest-based discovery rather than follower graphs — tools where you find content before you find accounts. The structural difference matters for creators: follower-graph platforms reward those who already have scale; interest-graph platforms can surface a new account with zero followers if the content matches intent. For brands, this is the more important shift. Instagram optimization — high production value, recognizable aesthetic, consistent posting cadence — is a playbook built for a world where the algorithm rewards reach accumulation. Interest-graph platforms reward specificity.

Barter as a behavioral signal

Marginal Revolution flagged a surge in barter markets across categories — cleaning services trading labor for robot training data, goods-for-goods exchanges in local communities, skills trading in professional networks. The phenomenon runs from practical (a cleaning startup acquiring first-person video of home interiors by offering free or discounted service) to structural. When cash-equivalent transactions feel expensive and trust in platform intermediaries is low, peer exchange re-emerges. Worth watching as a leading indicator of consumer confidence in the underlying economy.

Commerce Rewired

The bond market is writing a very large check on AI's future

Hyperscalers issued $155 billion in unsecured bonds year-to-date through May (paywall), Bloomberg reports — more than 45% above 2025's full-year total, with some AI infrastructure bond sales running four times oversubscribed. The oversubscription is the part worth sitting with. Credit markets are known for demanding collateral. A 4x oversubscription means institutional investors are actively competing to fund AI infrastructure debt at scale. The Bank of England's Andrew Bailey, warning this week that AI may need to be rationed due to power supply constraints, offers the limit case: if energy becomes the binding constraint before the revenue model matures, those bonds are backed by capacity that can't run at projected utilization. Doubleline and Oaktree, per the Bloomberg reporting, are already positioning for that scenario.

Detroit's EV problem doesn't have a domestic solution

Chinese EVs face 125% cumulative tariffs and a proposed Senate ban on direct US market entry, but the Next Web's analysis argues Detroit's most viable path may be partnership rather than competition. The cost structure on Chinese EV production — battery chemistry, manufacturing scale, supply chain integration — isn't something US automakers can replicate through domestic investment alone on any near-term timeline. The tariff wall keeps the product out but doesn't close the cost and technology gap. A partnership model would give US brands access to the technology while giving Chinese manufacturers a route around trade restrictions. Whether that clears regulatory review under current trade policy is a separate question.

Connected World

3D-printed housing at European scale

Europe's largest 3D-printed apartment building completed in the past week is worth noting for what it represents in the construction cost equation. Yanko Design's framing is characteristically maximalist, but the underlying signal is real: additive construction at residential scale reduces formwork, reduces labor hours, and compresses construction timelines. Housing supply in European cities is constrained primarily by cost and regulatory approval speed, not by demand. A technology that materially lowers the per-unit build cost is worth tracking, even if one building doesn't prove the model.

13 articles across 5 themes · 12 sources · Powered by Folo + Claude