The Adjacent Brief

TL;DR: Back-office workers — HR, billing, payroll — are emerging as the clearest near-term target for AI displacement, per new reporting from the Times, as enterprises simultaneously wrestle with AI cost models that don't fit existing FinOps frameworks. News publishers are moving to block AI crawlers by default, and Molly White launched a tracker for AI political spending.

Worth Reading

• Seattle bans new data centers for a year to study their impact — A 9-0 council vote with mayoral sign-off expected; the first major U.S. city to pump the brakes on infrastructure expansion rather than compete for it.
• SpaceX plans orbital AI computing tests by late 2027, has requested permission for up to 1 million data-center satellites — The most ambitious answer yet to data center geography constraints, though regulatory approval for a million satellites is its own problem.
• GM bets on energy storage for data centers and the grid — GM's battery play extends beyond vehicles; the data center market is now large enough to anchor an entirely separate business line.
• Data center modernization unlocks AI budget headroom — Enterprises finding AI spend capacity inside existing infrastructure rather than incremental capex — worth reading alongside the FinOps piece in today's Commerce section.
• The CLM market has a messaging problem, not a capability problem — Forrester's diagnosis applies well beyond CLM: enterprise AI vendors that can't explain value in one sentence are losing deals they should win.
• Brands are chasing the wrong 5% of shoppers — Channel Factory's Nerissa Valdellon on why performance-obsessed media buying systematically ignores the 95% of buyers not currently in market.
• The brand architecture behind Audemars Piguet x Swatch — A clean case study in how luxury uses collabs to expand addressable market without eroding the mothership's positioning.

Brand & Growth

The micro-SaaS bet is really a bet on who owns the spec

AI is enabling a new wave of micro-SaaS entrepreneurs who can build functional software products without engineering teams — the Next Web piece profiles founders shipping vertical tools in days rather than months. The economics look attractive on the surface: near-zero build cost, subscription revenue, defensible niche. The risk is that what's easy to build is also easy to replicate, and the moat shifts entirely to distribution and customer relationships rather than the product itself. Micro-SaaS has always been a distribution game; AI lowers the build cost without changing that math.

Connected World

The power problem has a vehicle-shaped answer, maybe

GM's vehicle-to-grid push frames EVs as distributed energy assets that can feed power back to data centers during peak demand — a genuine engineering idea that conveniently justifies GM's sodium-ion battery investment and positions the company as an AI infrastructure play rather than just an automaker. Whether V2G actually scales to data center power needs is unresolved; the grid coordination challenges are substantial. But the framing is deliberate: GM is arguing that its EV fleet is infrastructure, not just transportation, which changes the policy conversation and the procurement one.

ASML's stock tells you where semiconductor money isn't going

ASML is up 64% year-to-date (paywall) but trails the broader U.S. chip sector — Bloomberg's read is that capital is rotating toward advanced packaging and processes downstream of lithography rather than into the EUV machines ASML sells. AI hardware investment is concentrating in the assembly and interconnect layers that turn raw chips into usable systems. For anyone mapping the AI hardware supply chain, the money is moving closer to the finished product.

The $2 trillion infrastructure problem no one has a clean answer to

The Next Web's profile of the engineer working on AI infrastructure constraints puts a human face on a problem that mostly gets discussed in aggregate: the gap between AI compute demand and the physical infrastructure — power, cooling, fiber — that needs to exist before that compute can run. The $2 trillion figure is an estimate of what closing that gap requires. Seattle's data center moratorium (in Worth Reading) is a local expression of the same tension — communities bearing infrastructure costs that accrue elsewhere.

Culture & Signal

Publishers are making a choice, not a statement

More news sites are now blocking AI crawlers by default — Search Engine Journal reports Reuters and Time are among those requiring explicit allowlisting rather than opt-out. The shift from opt-out to opt-in is the operative detail: it changes the default relationship between publishers and AI companies from permissive to restrictive, and places the negotiating burden on the AI side. Whether this produces meaningful licensing revenue or just gets routed around is unresolved, but the posture is hardening across the industry.

AI as a chemistry lab for criminal networks

The Financial Times reports that the EU is warning of AI-boosted chemical synthesis helping European drug gangs develop new designer drug precursors (paywall) that evade existing blacklists faster than regulators can update them. This is the liability hedging problem made concrete: AI systems with chemistry capabilities don't distinguish between pharmaceutical R&D and precursor synthesis. The enforcement lag is measurable: new substances consistently hit markets faster than the scheduling process can respond. This pattern will surface in regulatory hearings before the end of the year.

Tracking AI's political spend before it becomes infrastructure

Molly White's Tech Influence Watch launches as a public tracker for AI industry political spending and lobbying — the kind of transparency tool that matters most when what's being tracked is still legible. AI lobbying is growing fast enough that a few years from now, the flows will be too large and distributed to map cleanly. White is building the baseline while the numbers are still small enough to count. For strategists watching regulatory trajectory, this is a useful early-warning resource.

The New Consumer

Siri finally works, which raises different questions

The Verge's hands-on with Siri AI under iOS 27 is notable less for what Siri can do than for the reviewers' surprise that it works at all. Apple spent years delivering a voice assistant that was worse than its reputation; the WWDC 2026 rebuild apparently crosses a basic utility threshold — shopping lists, schedule coordination, household task management — that the previous version couldn't. Apple used "AI" 28 times in the keynote, fewer than Google, which fits their pattern of letting the product speak while competitors narrate. The real test is whether these features hold up at population scale rather than in reviewer conditions.

Reddit won Google's algorithm, which means Google needs Reddit

Reddit gained top positions across virtually every niche following Google's May core update, extending a pattern that's been building for over a year. The mechanism is straightforward: Google's quality signals reward human-generated, community-vetted content at exactly the moment when AI-generated articles are flooding the web with plausible-but-unreliable alternatives. Reddit wins because it's structurally difficult to fake at scale. For brands, the implication is that Reddit presence is no longer an optional social channel — it's where Google sends people who want information they can trust.

The back office is the first real displacement story

The New York Times reports that back-office administrative roles — HR, billing, payroll — face near-term displacement from AI automation (paywall) ahead of more visible categories like software engineering. This fits the actual deployment pattern: enterprise AI buyers prioritize processes that are high-volume, rules-based, and don't require customer-facing judgment. Back-office work is all three. The displacement narrative has largely focused on coders and creatives because that's where the demos are. The actual adoption curve is running through functions that don't have public advocates and don't generate coverage until the layoffs are announced.

Commerce Rewired

Chinese capital controls have a crypto workaround, and it's working

The Financial Times reports that Chinese investors are using tokenized stocks purchased with stablecoins like USDT (paywall) to replicate positions in U.S. IPOs — including SpaceX — that Beijing's capital controls would otherwise block. The mechanism: buy a synthetic version of the asset on a crypto platform, bypassing the formal cross-border capital flow that would trigger scrutiny. This matters because crypto is providing utility the formal financial system isn't — specifically, access to assets for investors whose governments have decided they shouldn't have it. Regulators in both countries are paying attention.

FinOps wasn't built for AI's cost structure

Enterprise finance teams are rebuilding FinOps cost measurement models because AI infrastructure spending doesn't behave like cloud spending — it's less predictable, harder to allocate to business units, and doesn't respond to the same optimization levers. SiliconAngle's piece frames this as a tooling problem, but it's also a forecasting problem: when token consumption costs diverge from what traditional budgeting models expect, the variance hits finance teams who don't have line-of-sight into model usage. The companies figuring this out first — both vendors and internal teams — will have a meaningful advantage in how they present AI ROI to boards and investors.

14 articles across 5 themes · 9 sources · Powered by Folo + Claude