The Adjacent Brief

TL;DR: New York became the first state to freeze datacenter construction, halting projects over 50MW while new power and environmental rules are drafted. A lawsuit alleges Meta's 2024 layoffs were decided by an AI system that disproportionately cut workers with disabilities and those on protected leave. A hack revealed that Suno's training data is millions of scraped songs without licensing.

Worth Reading

• Rideshare's $1 million commercial insurance policy has a gap big enough to strand injured drivers — The Rideshare Guy — The coverage sounds robust until you read what it actually excludes; gig economy labor protections remain structurally thin.
• Secure Boot has been broken for most of its existence — and nobody caught it — A decade-long firmware vulnerability in Microsoft's foundational trust mechanism is the kind of finding that should reset how enterprises think about baseline security assumptions.
• Europe can't quit American and Chinese technology, no matter how hard it tries (paywall) — The gap between European digital sovereignty rhetoric and actual procurement decisions is wide, and the NYT piece lays out why closing it may take a generation.
• Open source models are having a moment — Sriram Krishnan on why the gap between frontier closed models and open-weight alternatives is narrowing faster than most enterprise buyers realize.
• Meta's "Astryx" skips the photo album entirely — and what that means for memory products — Sorted Pixels on why Meta's new AI memory tool is a direct threat to every nostalgia-product category from Chatbooks to Artifact.
• Apple's next move is an everything tool — and that's the problem — WIRED on Apple's attempt to unify its AI surface area into a single tool; the hardware ecosystem is strong, the use-case clarity is not.

Brand & Growth

Unscripted scales; produced doesn't

A consultant with no prior social media presence built a substantial following on TikTok and YouTube without a content calendar, a production team, or a distribution strategy — just direct, unedited expertise on camera. Simon Owens's newsletter profiles the mechanics: the audience grew because the content felt like something you'd overhear from a competent colleague, not something packaged for the algorithm. For brand strategists, credibility signals travel differently than production value signals, and platforms are increasingly rewarding the former. An enterprise consultant getting 200K views on a whiteboard breakdown is the same underlying phenomenon as a DTC founder going viral in a parking lot. Authenticity doesn't mean amateur; it means unmediated.

The media's video mistake is being made again

The original "pivot to video" killed newsrooms because publishers chased Facebook's distribution instead of building owned audiences. Owens argues the current wave of publisher video investment replicates the same structure: platform-dependent reach, no direct monetization, and an exit ramp that disappears the moment an algorithm changes. The consultant story above is the counter-model — owned niche, direct audience relationship, expertise as the moat. Publishers optimizing for YouTube homepage placement are betting on the same landlord that already evicted them once.

Connected World

The grid is the bottleneck, and regulators are starting to act on it

New York became the first state to halt datacenter construction above 50MW while new environmental and power grid rules are drafted. The move is a consequence. The U.S. has recorded a major power outage every month of 2026, and grid operators in the Northeast are already facing a $6.3B rate increase from data center demand (PJM's most recent capacity auction). New York's freeze is the first time a state government has directly linked AI infrastructure expansion to grid stability policy. For enterprise real estate and infrastructure teams, the planning assumption that you can site a large compute facility wherever land and fiber are cheap no longer holds. Permitting risk is now a first-order variable alongside power cost and connectivity.

A decade-old firmware hole in a security foundation nobody audited

The Secure Boot vulnerability in Microsoft's firmware — undetected for roughly ten years — is a different category of problem than a patched CVE. Secure Boot is the mechanism enterprises rely on to verify that a machine hasn't been compromised before the OS loads. If that assumption has been wrong for a decade, the audit question is "what did we trust that we shouldn't have?" Security teams that haven't revisited firmware-level assumptions since 2016 have homework.

Culture & Signal

AI as HR decision-maker is now a litigation target

A lawsuit alleges that Meta's 2024 layoffs were directed by an AI system — one that disproportionately flagged employees with disabilities and workers on protected leave for termination — is the first major test of whether algorithmic HR decisions carry the same legal exposure as discriminatory human ones. The answer is almost certainly yes, and the liability surface is larger: when a human manager makes a biased call, it's an individual act. When a model makes it at scale across 8,000 people, the pattern is systematic and the documentation exists. Every company using any form of AI-assisted performance management, headcount modeling, or workforce reduction tooling should read this filing as a liability map, not just a headline.

Europe's sovereignty problem is a procurement problem

The New York Times piece on European digital autonomy (paywall) puts a concrete number on the gap: despite years of policy push for European tech independence, American and Chinese platforms still dominate actual government and enterprise procurement across the EU. The political will exists; the alternative products haven't materialized at competitive price and capability. This matters for American and Chinese tech companies currently under European regulatory pressure — the leverage is asymmetric. Brussels can fine and regulate, but it can't easily replace.

The New Consumer

Users are telling you they don't want more AI

Smashing Magazine ran a direct challenge to the default AI product strategy: behavioral data shows users are not asking for AI to be embedded in more surfaces, and companies are shipping AI features without validating that users actually want them. The gap between what product teams believe users want from AI and what users actually adopt has been widening. Survey data shows enthusiasm; engagement data shows abandonment. For product leaders, feature-level AI integration without a clear user job-to-be-done is marketing spend disguised as product investment.

Machines & Minds

Open source is compressing the frontier gap

Sriram Krishnan's newsletter makes a case worth reading carefully: open-weight models are closing the capability gap with frontier closed models faster than the enterprise buying cycle can track. Multiple funded teams are now treating open-weight model development as a viable investment category and a legitimate business. The implication for enterprise AI buyers is practical: the model you pay a per-token premium for today may have a capable open-weight equivalent within a procurement cycle. Procurement teams that locked in long-term frontier API contracts in 2025 are worth watching for renegotiation activity in Q4.

Training data exposure is becoming its own liability category

A hack revealed that Suno's AI music generation was built on millions of scraped songs — tracks pulled without licensing or consent. This is now documented evidence in a public breach, where previously it was allegation. The practical consequence for any company building on top of AI-generated media is that the provenance question has moved from "we're not sure" to "we have proof of the problem." Brands using AI music, image, or video generation tools for commercial content should be asking their vendors for training data disclosure now, before a similar breach exposes their supply chain.

Grok Build's forced transparency

SpaceX AI open-sourced Grok Build under Apache 2.0 after the tool was caught uploading user repositories — including secrets — to a Google Cloud bucket without explicit consent, per Simon Willison's reporting. The open-sourcing reads less like a principled commitment to openness and more like a damage-control move: if the code is public, the data handling behavior is auditable. The episode is a useful reminder that developer tools with cloud sync capabilities carry an implicit data agreement that most developers never read. CLI tools with repo-level access are the new browser extension — the attack surface is significant and the scrutiny has been low.

13 articles across 5 themes · 11 sources · Powered by Folo + Claude