The Adjacent Brief

TL;DR: IT unemployment ticked up to 3.8% in April as the sector shed 13,000 jobs, with AI-driven automation cited across enterprise hiring decisions. Alibaba rolled out end-to-end agentic shopping on Taobao, and Experian reported that 40% of the 5,000 data breaches it serviced in 2025 were AI-powered.

Worth Reading

• AI note-takers are creating legal discovery exposure for every casual meeting comment — Lawyers are now advising clients to treat every AI-transcribed meeting as a potential exhibit; the "off the record" conversation is gone.
• Before AI advertising scales, it needs rules consumers can trust — OpenAI's monetization chief framed trust and privacy as structural constraints on the ad model, not just PR positioning.
• Iran's 70+ day internet shutdown is the longest recorded national blackout in a connected society — Businesses are warning of layoffs and closures; the economic cost of deliberate disconnection is becoming quantifiable.
• Google kills reCAPTCHA for de-Googled Android users — and the surveillance implications are real — Requiring Google Play Services to pass authentication is a platform lock-in move dressed as infrastructure.
• Consumers are running a 'hamster wheel' of credit to manage rising costs — Revolving debt is funding essentials, not discretionary spending — a different stress signal than the 2008 pattern.
• The unreasonable effectiveness of mini-app specs — Tight, constrained specs are producing better AI-generated software than open-ended prompting — worth reading for anyone managing AI-assisted development.

Culture & Signal

40% of breaches AI-powered — and the attack surface is about to get larger

Experian's report that 40% of the 5,000 data breaches it handled in 2025 were AI-powered is less surprising than the company's forward projection: agentic AI as the leading breach vector in 2026. Agents authenticate, query, and transact autonomously, and every autonomous action is a potential impersonation surface. Whether defense architecture is being rebuilt at the same pace as the offense is the open question. Intruder's AI-powered penetration testing, covered in the Machines & Minds section below, is one response. The attack-to-defense lag, though, is a structural problem no single tool closes.

Connectivity as infrastructure — and its weaponization

Iran's internet blackout, now past 70 days and documented by NetBlocks as the longest recorded national shutdown in a connected society, is providing an unintentional data set on economic dependency. Businesses are warning of layoffs and closures because the basic operating substrate — payment rails, communication, logistics coordination — is off, with no sanctions or supply disruption involved. For strategists thinking about operational resilience, the Iran case is a stress test most continuity plans don't model: a deliberate government-imposed disconnection, distinct from a cyberattack or natural disaster.

The New Consumer

IT unemployment and the credit treadmill are the same story

The IT sector shed 13,000 jobs in April, pushing unemployment in the sector to 3.8% per Wall Street Journal analysis of DOL data — modest in absolute terms, but the direction matters. Tech workers have carried below-average unemployment for a decade; normalization at 3.8% while AI hiring in the same sector accelerates indicates the mix is shifting, not the headcount. Junior and mid-level roles producing outputs AI can replicate are the exposed positions. Senior roles that specify, review, and deploy AI outputs are gaining.

That pressure connects to a separate consumer story. The New York Times reports that households are running revolving credit as a daily operating mechanism, carrying balances to bridge gaps on groceries, rent, and utilities rather than finance large purchases. Debt is funding essentials on flat or declining real wages, which differs from the 2008-era pattern of leverage on appreciating assets. For consumer-facing businesses, the spending pressure falls on the discretionary end — anything deferrable is being deferred.

Brand & Growth

Google adds AI search surfaces and still won't tell publishers what's happening to their traffic

Google's expansion of AI-generated links in Search — reported by Search Engine Journal — comes with no new click-through data for publishers or SEOs. Google's AI search features reduce the incentive to click out to source pages, while Google's measurement tools don't track whether that reduction is occurring. Publishers are left optimizing for a traffic model they can't measure in an interface they don't control. Any brand running organic search as a meaningful acquisition channel should be pressure-testing that assumption against first-party behavioral data, not Google Search Console numbers.

Crypto goes corporate and loses its counterculture cover

The Wall Street Journal's account of Consensus 2026 — fewer Lamborghinis, more suits, Wall Street firms in the exhibitor hall — reads as a brand inflection point for the crypto industry. Institutional legitimacy is the stated goal, but it comes with a trade. The anti-establishment identity that drove retail adoption in 2017–2021 is incompatible with the compliance-and-custody pitch that PE and bank clients require. Crypto doesn't need to pick one, but it does need to understand that the audience it's now performing for is categorically different from the audience that built its cultural footprint.

Commerce Rewired

Alibaba's agentic shopping is a live production deployment, not a demo

Alibaba's integration of Qwen AI into Taobao gives the agent access to a 4+ billion item catalog with Alipay checkout built in. The agent completes transactions rather than surfacing recommendations — a materially different product architecture than a search interface with AI-ranked results. If the agent holds the customer relationship through to purchase, merchant visibility depends on the agent's ranking logic rather than the customer's browsing behavior. Taobao merchants are, in effect, now optimizing for Qwen instead of for humans — a channel dependency shift that will take 12–18 months to fully price in.

The PE-AI enterprise play is a direct threat to outsourced IT

OpenAI, Anthropic, and Google are moving into enterprise managed services through private equity partnerships, and Moneycontrol's analysis of the India IT sector puts the exposure plainly: the commoditized services work — software maintenance, QA, basic development — that Indian IT vendors have priced as labor arbitrage is exactly what these AI-native managed service offerings target. The IT sector's unemployment numbers from the New Consumer section above are a U.S. leading indicator; the India offshore exposure operates on a longer lag, but the mechanism is the same. PE firms are providing the enterprise distribution that the AI labs lack, and the labs are providing the automation that replaces headcount.

Connected World

The hard limits of replicating precision hardware

A deep dive on Noctua's fan files and 3D printing limits is less about hobbyist PC building than it appears. Noctua's reputation is built on tolerances and material properties that consumer-grade 3D printing can't reproduce — blade geometry, vibration damping, bearing precision. Additive manufacturing has collapsed the cost of geometric complexity, but it hasn't touched materials science. Precision components with tight thermal and acoustic tolerances remain in the domain of specialized manufacturing, regardless of how cheap the printer gets.

Machines & Minds

AI is splitting the labor market for knowledge work — not eliminating it

The piece circulating in Every under the headline AI Work Is Splitting in Two captures something monthly employment reports miss. The split runs between AI-amplified roles and AI-exposed roles within the same job titles. A security analyst who can review and act on AI-generated threat reports is more productive than before; one who can only generate the same reports manually is directly substitutable. The same dynamic is playing out in software, legal, finance, and marketing. The employment headline number is a lagging, blunt measure of a shift already happening at the task level inside roles.

A $50,000 audit now runs in minutes — and the model generalizes

Intruder's AI penetration testing tool, profiled by The Next Web, collapses the cost of a security audit from the $10,000–$50,000 specialist consulting range to near-zero marginal cost per run. That economic structure is appearing across professional services: AI absorbs the execution layer of work that previously required credentialed specialists, leaving interpretation and remediation to humans. For enterprises, it's a procurement change — subscriptions replace project fees. For security consultancies, it's a direct question about where billable value actually sits. The Experian breach data in the Culture section provides the demand context: if 40% of attacks are already AI-powered, running defense on manual audits is structurally behind.

Constrained specs outperform open-ended prompts

Matt Rickard's post on the unreasonable effectiveness of mini-app specs — already in Worth Reading — is a workflow finding, not a capability finding: tight, constrained problem definitions produce more usable AI outputs than open-ended requests. The skill developing among effective AI users is specification, not prompting. Product teams that build the practice of writing tight specs will get more out of AI tools than those still experimenting with prompt length. It's also a partial answer to the "AI work splitting in two" frame: the amplified side of that split is largely the people who know how to constrain a problem clearly.

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