The Adjacent Brief

TL;DR: A Pew survey finds 49% of American adults used chatbots in 2026, up from 33% two years ago, while 63% say AI development is outpacing safety measures. PwC data shows companies using AI to augment workers pulling ahead of those using it primarily to cut costs. Midjourney announced a medical imaging capability, extending well beyond its generative art origins.

Worth Reading

• Mom-and-pop SaaS is real — AI is making solo operators economically viable at scale — The unit economics of small software businesses are changing faster than the VC consensus acknowledges.
• Brands jumping on the Knicks bandwagon reveals how cultural momentum gets strip-mined (paywall) — Unaffiliated companies borrowing sports energy as a social media tactic; worth watching for how quickly it dilutes.
• Epic wants your Fortnite skins to travel across games — and wants to own the standard — Cross-game item portability pitched as consumer benefit; the infrastructure play underneath it is what matters.
• Calling yourself "the best" in your copy may be boosting competitors in AI search — How AI retrieval reweights superlative claims in ways traditional SEO never had to contend with.
• The "50% of 2026 datacenter capacity canceled" claim is wrong — here's the actual number — Semianalysis walks back an overcirculated stat; the correction matters for anyone modeling infrastructure timelines.
• Airbnb hosts can raise prices on confirmed bookings — and the platform allows it (paywall) — World Cup travelers are finding out the hard way that a booking confirmation is not a locked rate.

Brand & Growth

Augment or automate — the returns are not symmetric

PwC's latest workforce study, covered by Bloomberg, cuts against the dominant cost-reduction framing of enterprise AI: companies using AI to enhance human skills are pulling ahead (paywall), while those deploying it primarily to cut headcount are falling further behind. Companies in the augmentation cohort are showing measurably higher productivity per worker, not just better employee sentiment scores. For brand and growth leaders, the implication is direct: AI investment pitched internally as efficiency is probably being routed to the wrong problem.

The AI-native firm is a structural question, not a size question

Tyler Cowen's note on AI-native firms at Marginal Revolution frames this at the organizational level — firms built around AI capabilities from day one operate differently from incumbents grafting AI onto existing workflows. The distinction matters for growth strategy: an AI-native competitor may reach your market with a cost structure and iteration speed that incumbents can't match by adding more tools. The companies winning on augmentation are running AI-native operating models inside legacy org structures. The ones losing are running legacy models with AI added on and calling it transformation.

Connected World

The infrastructure buildout has a neighbor problem

New York Times reporting on residents living near data centers who describe constant infrasonic vibration affecting their health and homes adds a friction layer to the infrastructure expansion story that financial models don't capture. Infrasound at sustained levels has documented physiological effects, and the complaints are concentrated around the cooling and power systems that large GPU clusters require. With $58 billion invested across 42 data center deals year-to-date per Dealogic, siting approvals are the rate-limiting step. Health complaints from adjacent communities convert into zoning delays, and zoning delays are capacity delays.

The living room as attack surface

Wall Street Journal reporting on nation-state hackers using preinstalled software on low-cost home devices (paywall) to build residential proxy networks is more consequential than it reads at first pass. The technique — hijacking devices that come with manufacturer software already installed — means the attack surface is baked in before the product reaches a consumer. Residential IP addresses are trusted by most corporate security filters in ways data center IPs are not, which is precisely the value: traffic routed through a compromised home router looks like a human browsing from Kansas. Enterprise security teams spending on perimeter defense are defending against an attack that enters through the supply chain, not the firewall.

Solid-state batteries find their first real test environment

Hackaday's coverage of solid-state batteries moving into aerospace applications is worth tracking for anyone watching energy storage as an infrastructure input. Aviation is a brutal proving ground — thermal range, weight, charge cycle requirements, and safety certification are all more demanding than automotive. If solid-state chemistries are clearing aerospace qualification, the consumer and grid storage timelines compress. Aerospace and defense validate, automotive scales, consumer follows — battery technology has moved that way before.

The New Consumer

Half the country uses chatbots; most of them don't trust the people building them

Pew Research's 2026 survey finds 49% of U.S. adults used a chatbot this year, up from 33% in 2024, with 24% using them daily. ChatGPT holds 44% of the named-tool share. Behavioral adoption at this scale is a real number — this is reported use, not aspiration measured by survey. The same survey finds 63% of Americans believe AI development is moving faster than safety can keep pace. The population using these tools and the population worried about them overlap substantially. That is not a contradiction — it describes how most new technologies actually get adopted. People use the thing and remain skeptical of the institution behind it simultaneously.

Platform trust erodes in the fine print

The Airbnb World Cup story — hosts raising prices on confirmed bookings ahead of the tournament (paywall) — is what happens when a platform's terms permit host-side repricing after guest confirmation, and guests discover this only at the moment of friction. Platform trust is downstream of policy design, and Airbnb's policy design here creates an expectation gap that individual hosts are exploiting at scale. A pattern is building across marketplaces: the platform captures the trust of the transaction, then the fine print reveals the trust was conditional.

Farmers are finding audiences by being themselves, not brands

Snaxshot's piece on farmers gaining social media traction through unpolished, personality-forward content — what Andrea Hernández calls "thirst trapping" — is a small signal worth watching. The reach these accounts are generating is the result of agricultural content filling a specific gap: provenance, process, and a human face behind commodity food. For food brands spending heavily on creator partnerships and polished content, the comparison is uncomfortable. The person who actually grew the thing is outperforming the brand that packaged it.

Culture & Signal

Local news finds a structural alternative to chain economics

Simon Owens profiles a startup building infrastructure that helps independent local outlets compete against legacy newspaper chains — by sharing back-office and revenue infrastructure across genuinely local newsrooms. The model deserves attention because it attacks the actual problem: individual local outlets fail on business operations, not on journalism. Shared services for advertising, subscriptions, and payroll let small newsrooms stay small editorially while becoming viable financially. The chain model consolidated editorial and operational control together; this separates them.

China's free AI offer to the developing world is a standards play

The Next Web reports that China announced plans to offer free or subsidized AI access to developing nations as the G7 debates how to restrict American model exports. Read this as infrastructure diplomacy. The country that gets its AI stack installed first in emerging markets shapes training data flows, dependency relationships, and technical standards for a generation. The G7's export control debate is focused on the wrong variable — restricting access to U.S. models while China offers its models for free is an asymmetric contest. Anthropic was reportedly forced to take Claude models offline in certain markets due to export control enforcement; that kind of friction compounds the asymmetry.

Commerce Rewired

Prediction markets are becoming a legitimate hedging instrument — with real volume to show it

The New York Times reports that Kalshi's institutional trading volume grew 800% since November 2025, driven by businesses using the platform to hedge operational risks rather than speculate on political outcomes. This is the transition that prediction market advocates have argued was coming for years: from retail novelty to institutional risk management instrument. If your supply chain depends on a regulatory outcome or a weather event, and you can buy a contract that pays if that outcome occurs, you've created a partial hedge without a counterparty relationship with a bank. At 800% volume growth, someone has done the diligence on whether these contracts actually perform. The question for CFOs watching this is whether their risk management frameworks are written to permit it.

Machines & Minds

Enterprise AI is stuck in pilot — and the reason is the data, not the model

SiliconANGLE's coverage of the AI-ready data gap keeping enterprise deployments in pilot mode names the mechanism that most AI adoption reporting skips past: the models are fine; the data pipelines feeding them are not. Governance gaps, inconsistent formatting, siloed storage, and unclear ownership over training sets are the actual blockers. The infrastructure spending is on compute, but the constraint is data quality and organizational readiness to treat data as a product rather than a byproduct. Vendors selling AI capacity are not selling the thing enterprises actually need.

Midjourney moves into medical imaging — the scope question gets sharper

The Verge reports that Midjourney demonstrated medical imaging segmentation capability on ultrasound phantoms, a significant departure from generative art. This is a capability demonstration on test objects, not clinical data. The directional move matters. Midjourney has model infrastructure, a paying user base, and revenue that most AI labs would trade for. Applying that to medical imaging puts it adjacent to companies like Butterfly Network and Viz.ai that have spent years on FDA clearance and clinical validation. The technology is not the hard part of medical AI. The regulatory pathway, liability structure, and clinical workflow integration are. Midjourney's demo establishes capability; it doesn't resolve any of those.

Google's LLMs.txt critique lands on a real problem

Search Engine Journal's piece on Google arguing that LLMs.txt's core assumption conflicts with how its creators intended it to work is a useful counterpoint to the publisher community's enthusiasm for the standard. LLMs.txt was proposed as a way for publishers to signal to AI crawlers which content they consent to be trained on. Google's position — that AI systems ingest content dynamically rather than via crawl-and-train pipelines in ways LLMs.txt assumes — means the standard may not do what publishers think it does. This connects to a pattern in the publisher-AI relationship: Google's AI search results cite sources while recommending competitors, reducing click-through to the cited publisher. LLMs.txt was one proposed remedy. If the mechanism is broken, publishers are back to negotiating without leverage.

15 articles across 6 themes · 14 sources · Powered by Folo + Claude