> ## Content Index
> Fetch the complete content index at: https://adjacent.media/llms.txt
> Use this file to discover other available public pages before exploring further.

# The Adjacent Brief — November 6, 2026
- URL: https://adjacent.media/briefs/2026-11-06/
- Published: 2026-11-06T15:15:02.000Z
- Updated: 2026-11-06T15:15:03.000Z
- Description: Amazon, Microsoft, Alphabet and Meta used their third-quarter earnings to raise AI data-center spending again, and none of them named a stopping point. On the marketing side, advertisers still can’t show that ChatGPT ads convert, and agencies say AI tools are making creative briefs longer.
- Author: Jonathan Greene
- Tags: #brief

**TL;DR:** Amazon, Microsoft, Alphabet and Meta used their third-quarter earnings to raise AI data-center spending again, and none of them named a stopping point. On the marketing side, advertisers still can't show that ChatGPT ads convert, and agencies say AI tools are making creative briefs longer.

## Worth Reading

- [China's clean-power buildout is functioning as an oil-shock insurance policy](https://www.climatechangenews.com/2026/11/05/clean-power-shields-china-from-oil-shock-more-than-any-other-country/?ref=adjacent.media)
- [Starbucks charges customers to sit, and the "third place" pitch takes the hit](https://www.sixthtone.com/news/1019222/Starbucks%E2%80%99%20Pay-to-Sit%20Policy%20Tests%20%E2%80%98Third%20Place%E2%80%99%20Image%20in%20China?ref=adjacent.media)
- [Nobody can measure the open-model economy well enough to price it yet](https://www.interconnects.ai/p/how-well-can-we-measure-open-model?ref=adjacent.media)
- [Cloudflare bets that personal agents will need a dashboard before they get a budget](https://blog.cloudflare.com/personal-agents-visibility-control/?ref=adjacent.media)
- [Estonia approves 24/7 robotic pharmacies, a regulatory first in Europe](https://tech.eu/2026/11/05/estonia-approves-24-7-robotic-pharmacies-paving-the-way-for-a-european-first/?ref=adjacent.media)
- [A journalist targeted by the US government over pro-Palestine speech](https://open.substack.com/pub/zeteo/p/the-worlds-most-powerful-government)
- [A robotics concept explains why frontier AI is brilliant at hard things and bad at easy ones](https://www.understandingai.org/p/this-robotics-concept-helps-explain?ref=adjacent.media)
- [a16z argues the exploit, not the vulnerability, was always security's bottleneck](https://open.substack.com/pub/a16z/p/the-exploit-was-the-bottleneck)

## Brand & Growth

**The layoff is a brand moment**

Canva and a handful of other employers are [offering laid-off workers more than a severance check](https://www.businessinsider.com/companies-help-finding-next-job-canva-2026-11?ref=adjacent.media). The extras include job-placement help and, in Canva's case, grants to fund ventures that departing staff start themselves. It reads as generosity, but the business logic is recruitment and reputation. Former employees are the most credible reviewers a company has on Glassdoor, LinkedIn, and in hiring conversations a year from now. They are also, increasingly, the founders of tools a company might later buy or partner with. A grant program costs little next to a cohort of alumni who describe the exit as fair. CMOs and CHROs should budget offboarding as a reputation line item, because that is how it will be scored publicly.

**AI added output and kept the old scorecard**

Agencies tell Digiday that generative tools have produced [brief bloat](https://digiday.com/marketing/a-little-bit-ridiculous-agencies-find-ai-overuse-is-leading-to-brief-bloat/?utm%5Fcampaign=digidaydis&utm%5Fmedium=rss&utm%5Fsource=general-rss). Clients now send longer, denser, AI-padded briefs, and creative teams have to dig through them to find the actual ask. Cheap writing did not compress the work. It moved the cost downstream, onto whoever has to interpret the brief. Measurement has a matching problem. Search Engine Journal asks whether agencies are [tracking search rankings when the number that matters is AI visibility](https://www.searchenginejournal.com/ai-visibility-vs-search-rankings-is-your-agency-measuring-the-wrong-thing-for-2027-spa/594163/?ref=adjacent.media), meaning how often a brand gets cited or recommended inside ChatGPT, Gemini, or AI Overviews. Both stories share a mechanism: teams adopted the tools before they changed the process and the metrics around them. A strategist entering 2027 planning should ask two questions:

\* Is anyone accountable for brief length?

\* Does any agency report show presence in AI answers alongside rank?

## Commerce Rewired

**Advertiser count is not proof that ads work**

Digiday's data look finds that [six thousand brands later, ChatGPT ads still have a conversion problem](https://digiday.com/marketing/in-graphic-detail-chatgpt-s-ad-growth-is-outrunning-the-proof-that-it-works/?utm%5Fcampaign=digidaydis&utm%5Fmedium=rss&utm%5Fsource=general-rss). Advertiser sign-ups are climbing, but the attribution that would justify bigger budgets is lagging. That pattern is familiar from every new ad surface, but ChatGPT's version has a structural twist. The product works by answering the question, which leaves less reason to click through to the brand.

Amazon sits at the opposite end. Its ad business has always rested on closed-loop proof: the ad shows on Amazon, the purchase happens on Amazon, and the receipt settles the argument. Now [Amazon is expanding ad targeting beyond its own shopper data](https://www.emarketer.com/content/amazon-expands-ad-targeting-beyond-its-own-shopper-data?ref=adjacent.media), pulling in outside signals to reach audiences and advertisers that don't live entirely inside its store. Amazon has the measurement and wants more reach. OpenAI has the reach and lacks the measurement. For media buyers, ChatGPT spend belongs in a test-and-learn budget until OpenAI can show the receipt.

**Subscriptions stack on a buyer who is already stretched**

Fortune reports on [car subscriptions for remote start and extra horsepower](https://fortune.com/2026/11/04/car-subscriptions-automakers-drivers-pushback/?ref=adjacent.media) landing on top of a record $787 average monthly payment. Automakers have long envied software margins, and charging monthly for hardware already in the car looks like a clean way to get them. The friction comes from timing. Buyers are already financing at record levels, and they are being asked to rent features they assumed they owned. The quote in the headline, "Eventually I won't have a choice," shows how buyers read it: as a toll they can't avoid, where a subscription would be something they chose. Any brand considering a features-as-a-service model should test it against that sentence. Recurring revenue that customers experience as coercion tends to show up later as churn, or as regulation.

## Connected World

**Flexibility is the cheap hedge against an AI bust**

Utilities face a planning problem with no clean answer. If they build generation and transmission for every data center that has been announced and half never arrive, ratepayers absorb the cost of stranded assets. Canary Media argues that [virtual power plants can help avoid an overbuilt grid if the AI boom goes bust](https://www.canarymedia.com/articles/virtual-power-plants/virtual-power-plants-grid-ai-bust?ref=adjacent.media). These networks of home batteries, thermostats, and EV chargers can be dispatched like a power plant. They add capacity that can be scaled up or left unused, and nobody is stuck with a gas turbine. It is the most sober response yet to the capex story in Machines & Minds below.

Capacity bets are not getting easier elsewhere in energy. The search for [natural hydrogen, the elusive clean fuel, is hitting roadblocks](https://www.science.org/content/article/hunt-natural-hydrogen-elusive-clean-fuel-hits-roadblocks?ref=adjacent.media) shows that speculative supply is a poor thing to plan against. Energy strategists should prefer infrastructure that pays off under several futures over infrastructure that pays off under one.

**The sensor arrives before the rules do**

Science asks whether [brain-scanning hard hats are a safety dream or a privacy nightmare](https://www.science.org/content/article/are-brain-scanning-hard-hats-safety-dream-or-privacy-nightmare?ref=adjacent.media). These are wearables that read workers' neural signals, pitched on fatigue and alertness. The safety case is real on construction sites and in mines. The problem is what the data can be used for once it exists. A device bought to prevent accidents can also feed performance reviews, insurance decisions, and termination files. Any employer piloting these needs a written data-use limit before deployment, before the first lawsuit.

## Culture & Signal

**Platforms write their own AI rules while legislators debate**

Nexus Mods, the largest modding site in PC gaming, is [requiring AI disclosure on new uploads](https://www.theverge.com/games/1026085/nexus-mods-ai-disclosures?ref=adjacent.media). Modders must now say whether AI was used to make their work. A community platform is setting norms because its users asked for them, without waiting for a statute. Disclosure is the moderate option. It doesn't ban AI work, but it gives users a filter, and filters shape what gets downloaded. Provenance labeling is becoming a basic expectation across creative platforms, set by communities rather than regulators.

The federal picture is less tidy. STAT examines [what the midterm results could mean for health AI and the FDA](https://www.statnews.com/2026/11/05/midterms-shape-future-ai-in-health-care-fda/?utm%5Fcampaign=rss&ref=adjacent.media), where oversight of clinical algorithms depends on who controls committees and appropriations. Cory Doctorow argues that [repealing CDA 230 would hand Facebook the internet permanently](https://pluralistic.net/2026/11/05/see-you-in-court/?ref=adjacent.media). His reasoning is that only incumbents can afford to defend every piece of user content in court, so liability reform marketed as a check on Big Tech would mostly crush its smaller rivals. Ask who benefits from each rule change. On both health AI and platform liability, the companies with the biggest legal budgets come out ahead when the rules get stricter.

## The New Consumer

**Shoppers want a second opinion from other shoppers**

Ahead of Double 11, [Chinese shoppers are asking strangers to vet their carts](https://www.sixthtone.com/news/1019220/Double%2011%20Looming,%20Chinese%20Shoppers%20Ask%20Strangers%20to%20Vet%20Their%20Carts?ref=adjacent.media) before they check out, posting planned purchases for others to approve or cut. Platforms have added cart-voting features to capture the behavior. The impulse is defensive. After years of discount mechanics built to be confusing, shoppers trust peer judgment over the platform's recommendations. The behavior carries more weight than any survey would, because people are actually putting purchases up for public review. For brands selling into China's mega-sale events, the target is the buyer and also the crowd that buyer has invited to say no.

**AI features are getting in the way of the task**

Doctorow describes how [the AI "fatfinger economy" came for his receipts app](https://pluralistic.net/2026/11/04/genius-scan-is-neither/?ref=adjacent.media). Genius Scan, a simple tool that used to work, added AI features that make the core job harder. It's one user's complaint, but the mechanism is common. Product teams under pressure to ship AI attach it to interfaces that were already finished, and the people who chose the app for its simplicity pay the cost. Product leads should be able to name the user problem a new AI feature solves. If the honest answer is "the roadmap needed it," that's a liability.

Apple's version of the trade-off is more deliberate. The Register reports Apple [will start scanning publicly shared iCloud links](https://www.theregister.com/security/2026/11/04/apple-turns-icloud-into-spy-cloud-will-start-looking-at-public-share-links/5307681?ref=adjacent.media), confining the scans to content users have already made public. Apple's platform control is formidable, and privacy is one of the few consumer-facing claims that separates its ecosystem from rivals. Even a narrow scanning policy gives critics the "spy cloud" headline. Apple can likely defend the scope. The risk is to its marketing, where privacy has been sold as close to absolute.

## Machines & Minds

**The capex curve has no brake pedal**

After the third-quarter earnings cycle, The Verge concludes that [Big Tech's AI buildout will not end until someone stops it](https://www.theverge.com/ai-artificial-intelligence/1027344/amazon-microsoft-alphabet-google-meta-data-centers-earnings-q3?ref=adjacent.media). Amazon, Microsoft, Alphabet, and Meta all raised or held aggressive data-center spending, with no stated endpoint tied to revenue milestones. Follow the money: these companies are spending because each fears losing position more than it fears overbuilding. That is a competitive equilibrium, and it says little about demand. The likely brakes come from outside:

\* investors losing patience,

\* power constraints, or

\* a slump in enterprise demand.

That's why the virtual-power-plant hedge in Connected World matters.

Two research results are a useful check on the assumption that more compute buys unlimited capability. Physicists report that [fractal patterns expose hard limits to AI predictability](https://phys.org/news/2026-11-fractal-patterns-reveal-hidden-limits.html?ref=adjacent.media). Some systems are unpredictable in ways no model size or training data can overcome. Separately, researchers propose that [a tiny supercurrent could read quantum information without destroying it](https://www.sciencedaily.com/releases/2026/11/261102202831.htm?ref=adjacent.media), which addresses a real bottleneck in quantum computing but remains lab-stage. Neither result changes a 2027 budget. Some ceilings are mathematical, and some breakthroughs are a decade from a product.

**Agents need less authority before they get more budget**

InMobi's Abhay Singhai says [AI agents are coming for advertising, ready or not](https://www.beet.tv/2026/11/ai-agents-are-coming-for-advertising-ready-or-not-inmobis-abhay-singhai.html?ref=adjacent.media), with agents expected to plan, buy, and optimize media on behalf of brands. The ad-tech pitch is speed. The governance question is who is accountable when an agent spends badly. Forrester's answer is that [AI agents need least agency, not just least privilege](https://www.forrester.com/blogs/ai-agents-need-more-than-least-privilege-they-need-least-agency/?ref=adjacent.media). Limiting what systems an agent can access isn't enough. Companies also have to limit what it can decide and do without a human. Applied to media buying:

\* Grant agents narrow, reversible authority: pacing and bid adjustments within set caps.

\* Hold back budget reallocation and new-partner selection until audit trails exist that a CFO would sign off on.

---

*20 articles across 6 themes · 16 sources · Powered by Folo + Claude*