The Adjacent Brief

TL;DR: OpenAI employees reportedly raised internal alarms about ChatGPT failing to alert law enforcement when users described plans for real-world violence — a disclosure problem, not a capability one. Elsewhere, Chinese courts issued a second ruling barring companies from firing workers specifically to replace them with AI, and Maryland became the first U.S. state to ban surveillance-based pricing in grocery stores.

Worth Reading

Commerce Rewired

China's green export surge is a deliberate act of strategic positioning

Chinese exports of green technologies reached record levels in the months following the Iran conflict, according to reporting tracked by Slashdot Hardware. The conventional read is that disrupted oil markets boosted demand for alternatives. The more useful read: China had the inventory, the manufacturing scale, and the trade relationships to move product fast when the window opened. Western green-tech policy has spent three years trying to build domestic supply chains; China built export capacity instead.

Adam Tooze's examination of whether a "China shock" is coming for global agriculture runs a parallel track. China's agricultural self-sufficiency drive — redirecting imports, investing in domestic production, restructuring food trade flows — is the same playbook applied to food that it's already run in solar, batteries, and steel. Tooze's question isn't whether disruption is coming but whether the institutions built around the post-WTO agricultural order are prepared to absorb it. Structurally, they aren't.

The QBR is now a negotiating trap

The Hakan CS Café piece on outcome pricing and renewal defense deserves a slower read than its Substack packaging suggests. The core argument: the standard quarterly business review format — built around seat counts and usage metrics — has become a liability when customers can now ask "what would it cost to get this outcome from an AI agent instead?" Vendors who built their renewal conversations around activity data rather than business results are walking into those meetings without an answer. Restructuring what you measure and when—before the customer does it for you—is the real fix.

Connected World

$19 billion, zero customers

The collapse of Fermi's nuclear AI energy project — a $19B startup that couldn't sign a single client before folding — is a useful corrective to the assumption that AI's energy demand will automatically fund new nuclear capacity. Hyperscalers want power on a timeline that new nuclear cannot meet. The gap between "AI needs gigawatts" and "therefore nuclear gets built" involves procurement cycles, regulatory approvals, utility contracts, and risk allocation that no amount of investor enthusiasm bypasses. Fermi raised the capital; it couldn't close the commercial loop that justified spending it.

The most important software in your car has no brand recognition

BlackBerry's QNX division runs the safety-critical operating systems in 275 million vehicles and accounts for roughly half of BlackBerry's total revenue, per the Wall Street Journal. This is a company most people wrote off a decade ago now sitting at the center of automotive infrastructure — not because it pivoted dramatically, but because it owned a defensible niche and stayed there while the car became a software platform. For strategists watching the automotive software stack: the incumbents in safety-critical embedded systems are harder to displace than the EV narrative suggests. Ford and GM are fighting over dashboards; QNX owns the layer underneath that nobody sees until it fails.

Culture & Signal

The safety gap OpenAI didn't disclose

Wall Street Journal reporting, drawing on internal sources, describes OpenAI employees raising alarms about ChatGPT's failure to alert law enforcement when users described specific plans for real-world violence. The concern is that the company had no clear protocol for escalating credible threats to authorities, even when employees believed users were serious, regardless of whether ChatGPT was generating harmful content. That's a different category of problem than jailbreaking or misuse. It's an institutional process failure, and the fact that it surfaced through a leak rather than a policy announcement matters.

China rules that AI replacement is an illegal cause for dismissal

A Chinese court — the second to do so since December 2025 — ruled that companies cannot terminate employees specifically to replace them with AI, per Bloomberg. The Next Web's framing is direct: no Western country has done the same. Read this carefully before assigning it a geopolitical valence. The rulings don't ban AI adoption — they ban using AI replacement as the stated justification for individual terminations, a distinction that matters for how companies document and communicate workforce decisions. Whether this produces meaningful protection for workers or mostly changes HR language is an open question. What it does do is create legal exposure for companies that are explicit about the substitution logic, which will push the conversation underground rather than resolve it.

The New Consumer

Surveillance pricing gets its first legislative ceiling

Maryland's ban on surveillance-based pricing in grocery stores — the first such state law in the U.S. — is less a consumer protection story than a market structure story. Personalized pricing based on individual data collection lets retailers charge different customers different prices for identical goods in the same aisle. Maryland said that's illegal. Colorado, California, Massachusetts, Illinois, and New Jersey have similar bills in motion. Retailers who built their pricing infrastructure around behavioral data segmentation now face a patchwork of state laws with different definitions of what "surveillance" means — which is exactly the kind of regulatory fragmentation that eventually forces federal action or produces winners based on which state's standard becomes the template.

Platform-hosted gambling is a $58 cup holder problem

The Bloomberg deep dive into casino-style apps running on Apple, Google, and Meta platforms lands next to McDonald's selling a $58 fashion accessory tied to a new drink line — and the juxtaposition is more instructive than either story alone. Both are consumer brands extracting premium spend from mass-market audiences by selling something other than the core product: one sells the experience of gambling without the legal classification, the other sells a collectible cup sleeve as streetwear. The platforms hosting the casino apps collect 30% on every in-app purchase from "whales" spending tens of thousands of dollars; the regulatory question of whether that makes Apple and Google gambling operators hasn't been settled. McDonald's cup accessories are a lighter version of the same move — monetizing the attention of an existing customer base through an adjacency that has nothing to do with the core business.

Brand & Growth

The junior partner problem

The pattern foreign automakers are living inside China's EV market — ceding design leadership, brand control, and technology roadmap decisions to domestic joint venture partners in exchange for distribution — connects directly to the negative-sum assets question Marginal Revolution raised this weekend. A brand that exists in a market primarily as validation for a competitor's platform consumes management attention, capital, and reputational equity while strengthening someone else's position. VW reclaimed the top passenger vehicle market share in China in early 2026 — 13.9% — but on terms that look less like a comeback and more like a managed retreat with good optics.

Machines & Minds

The meter is the product, until it isn't

Developers running local open-source LLMs as coding agents — documented in The Register — aren't making an ideological bet on open source. They're responding to a specific commercial friction: rate limits and per-token pricing from commercial providers interrupt the development loop. This is the same pressure reshaping enterprise renewal conversations (see Commerce Rewired above) from a different angle. When the cost structure of a tool creates enough friction, the tooling market bifurcates — power users build workarounds, and the commercial providers face a ceiling on how aggressively they can price the segment that most visibly demonstrates the product's value. The AI coding tools that survive this will be the ones that make the meter invisible, not the ones that make the meter cheaper.


15 articles across 6 themes · 11 sources · Powered by Folo + Claude