> ## 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 — August 13, 2026
- URL: https://adjacent.media/briefs/2026-08-13/
- Published: 2026-08-13T14:15:03.000Z
- Updated: 2026-08-13T14:15:03.000Z
- Description: Google said its Gemini app passed 1 billion monthly active users, days after ChatGPT reported the same milestone.
- Author: Jonathan Greene
- Tags: #brief

**TL;DR:** Google said its Gemini app passed 1 billion monthly active users, days after ChatGPT reported the same milestone. Chainalysis put growth in crypto impersonation scams at 1,400% year over year, and researchers told the *Financial Times* that suspected Chinese operators assembled open-source AI agents into an autonomous tool that compromised Taiwanese government sites in July. DoorDash appeared twice today — once selling CPG advertisers its transaction data, once pushing harder on delivery robots.

## Worth Reading

- [AI's productivity gains come with a bill nobody is booking](https://www.machinesociety.ai/p/productivitypollution?ref=adjacent.media) — The externalized costs of AI deployment don't show up in the ROI deck, which is exactly why they'll show up later.
- [Suno has 2M paying subscribers and two major labels in court](https://www.bloomberg.com/news/features/2026-08-11/suno-stares-down-universal-and-sony-to-reshape-the-future-of-music?accessToken=eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJzb3VyY2UiOiJTdWJzY3JpYmVyR2lmdGVkQXJ0aWNsZSIsImlhdCI6MTc4NjQ4NjQxMCwiZXhwIjoxNzg3MDkxMjEwLCJhcnRpY2xlSWQiOiJUSk1JQ0hSMjRVOU8wMCIsImJjb25uZWN0SWQiOiI4OUM4OTNDMDhGOTQ0NThDQkQwQTQyREY1RDFCOTY0QyJ9.TBMIeS8NPLcnFaiLOvDCXHN0JX7ghM24kRXIhlh%5FtTY&ref=adjacent.media) (paywall) — Bloomberg's profile is the clearest read on whether generative music gets licensed or litigated into a settlement price.
- [Sony deleted 551 movies customers had already paid for](https://mjtsai.com/blog/2026/08/11/sony-removing-purchased-content/?ref=adjacent.media) — Paired with a \[Microsoft account that nearly took a decade of files and photos with it\](https://mjtsai.com/blog/2026/08/11/microsoft-account-with-files-photos-and-games-almost-lost/), the "purchase" button on a platform keeps meaning less than the word implies.
- [Two chat apps crossed a billion users in the same week — distribution did most of the work](https://www.theverge.com/ai-artificial-intelligence/978113/chatgpt-gemini-1-billion-users?ref=adjacent.media) — One got there through habit, one through preinstallation. Those are not the same asset.
- [The Galaxy Z Fold 8 found the ceiling on paying for iteration](https://www.sammobile.com/opinion/galaxy-z-fold-8-shows-customers-no-longer-pay-boring-upgrades/?ref=adjacent.media) — Smartphone shipments fell double digits in H1; the foldable premium was supposed to be the exception.

## Brand & Growth

**Your traffic report is measuring a step that buyers stopped taking**

Search Engine Journal makes the case that most SEO teams still aren't tracking [whether AI answers cite them at all](https://www.searchenginejournal.com/the-metric-most-seos-are-still-not-measuring-in-the-age-of-ai/584027/?ref=adjacent.media) — sessions, rankings, and CTR describe a journey that increasingly terminates inside the answer box. This lands the week after Microsoft Clarity started separating branded from non-branded AI citations, counting only the latter toward discovery. The distinction matters: branded citations mean someone already knew to ask for you, and reporting them as "AI visibility" flatters the number. For a CMO, the practical move is small and unglamorous — add citation share to the same dashboard as organic sessions before the next budget cycle, because the conversation about cutting content spend will happen with or without the data.

**When code generation gets cheap, the fight moves downstream**

GitHub, Vercel, and Replit have stopped competing on who writes better code and started competing on [what happens after the code exists](https://blog.bytebytego.com/p/github-vs-vercel-vs-replit-what-dev?ref=adjacent.media) — deployment, preview environments, workflow lock-in. Generation commoditized fast; the runtime didn't. Read alongside it the argument from Refactoring that [junior engineers are now the better hire, not the obvious cut](https://refactoring.fm/p/the-case-for-hiring-junior-engineers?ref=adjacent.media): if AI handles the boilerplate that used to justify a junior's first two years, the cheap conclusion is to stop hiring them. The expensive conclusion — the one that shows up in 2030 — is that senior engineers are grown, not sourced, and the companies that stopped the pipeline will be bidding against each other for a shrinking pool. Both pieces describe the same market dynamic from opposite ends: the scarce thing is no longer output, it's the judgment about what to ship and who can be trusted to ship it.

## Culture & Signal

**Heat is a payroll line before it's a climate line**

The *New York Times* reconstructed [a postal carrier's death on a route in extreme heat](https://www.nytimes.com/2026/08/12/business/heat-kills-hundreds-of-workers-each-year-a-postal-carrier-was-one.html?ref=adjacent.media) (paywall), one of the [hundreds of US workers killed by heat each year](https://www.nytimes.com/2026/08/12/business/extreme-heat-worker-deaths.html?ref=adjacent.media) (paywall) in an economy with no enforceable federal heat standard. For any business with outdoor labor — logistics, construction, agriculture, last-mile delivery — heat is now a scheduling, insurance, and liability variable, showing up in corporate climate-adaptation budgets that were voluntary two years ago and are now written against direct operational losses. Companies treating this as an ESG disclosure question are answering the wrong question. The question is what a shift looks like at 105°F and who signs off on it.

**Governance is being written at the hardware and export layer, leaving the model layer largely unregulated.**

A roundup from the AI Governance, Ethics and Leadership newsletter tracks the week's actual policy movement — smart glasses, Zuckerberg's personal-AI essay, and Chinese open-source models landing on blacklists, a number that arrives with an asterisk the size of Android. Google has default placement in search, in Workspace, and on the majority of the world's phones; ChatGPT reached the same milestone by getting people to type a URL. Both facts are real, and they buy very different things: one is a durable habit, the other is a distribution advantage that persists exactly as long as the placement does. If you're modeling assistant reach for a media plan, treat Gemini's number as inventory and ChatGPT's as intent.

**The behavioral signal is people building their own tools, not people saying they'd use AI**

The *Wall Street Journal* found fitness obsessives [wiring sleep, training, and diet data into custom AI dashboards](https://www.wsj.com/tech/ai/the-datamaxxers-feeding-their-every-health-move-to-ai-df0af9b5?st=hqiG7z&&ref=adjacent.media#x26;reflink=desktopwebshare%5Fpermalink) (paywall) they built themselves, building the analysis layer that no vendor shipped rather than downloading a wellness app. That's a small population doing real work, which makes it a far better forward indicator than any "would you use AI for health advice" survey. The commercial read for anyone in fitness, supplements, or connected hardware: the high-value customer is now assembling a data stack across your product and four competitors', and whoever exports cleanly becomes the hub.

**The same fluency that personalizes also impersonates**

Chainalysis put the growth in crypto impersonation scams at [1,400% year over year](https://thenextweb.com/news/crypto-impersonation-scams-1400-percent-ai-deepfake-fraud?ref=adjacent.media), driven by cheap voice and video synthesis that made the expensive part of a con — being convincingly someone else — nearly free. Crypto is the leading indicator, not the whole market, because it offers irreversible settlement. Every brand with a customer-service voice channel or an executive with public video footage holds the same exposure on a delay.

## Machines & Minds

**Agentic AI gets serious where a wrong answer is a legal filing**

Avalara is deploying agents into [real-time tax compliance](https://siliconangle.com/2026/08/11/avalara-brings-agentic-accuracy-ai-powered-tax-compliance-googlecloudaiagentsinaction/?ref=adjacent.media), which is the most useful stress test in the category because the accuracy target is a jurisdiction's audit standard, and the penalty for being 97% right is a notice from a state revenue department. This is where the enterprise agent thesis either earns its funding or doesn't, and it's a more informative case than any coding demo. The pricing implication runs through an argument in Indieblog that [per-seat SaaS is being unbundled by software people write for themselves](https://jamie.ideasasylum.com/2026/06/20/saas-is-dead?utm%5Fsource=indieblog.page&utm%5Fmedium=rss&utm%5Fcampaign=indieblog.page): when the seat stops being the unit of value, vendors selling outcomes with liability attached — a filed return, a cleared invoice — have a defensible position, and vendors selling a UI over a database do not.

**Open-source agents cut the cost of attack faster than the cost of defense**

Researchers told the *Financial Times* that suspected Chinese operators chained [freely available AI agents into an autonomous hacking tool](https://www.ft.com/content/7d2ab3e0-9085-48f6-b38a-d90260d58795?ref=adjacent.media) (paywall) that compromised Taiwanese government websites in July, with no frontier lab access required and no novel capability, just orchestration of components anyone can download. It arrives days after OpenAI shipped a model explicitly trained for zero-day discovery, and alongside a finding in *Wired* that a frontier model's encrypted reasoning traces can be [coaxed back into plaintext by feeding them to a weaker sibling model](https://www.wired.com/story/a-new-trick-reveals-ai-models-inner-thoughts/?ref=adjacent.media). Safety mechanisms that depend on withholding information keep failing to withhold it, and the mitigation gap is a distribution problem rather than a research one: defense requires deployment, while offense requires only a download.

**Provenance is a platform play, and Apple is positioned better than anyone to run it**

Apple is building [cryptographic verification that a photo came off an iPhone camera](https://www.theverge.com/tech/977921/apple-reference-image-iphone-metadata?ref=adjacent.media) into iOS, certifying origin at capture rather than trying to detect fakery after the fact. Detection is a losing arms race; attestation isn't, and Apple controls enough of the world's camera hardware to make an origin signature meaningful at scale. This is ecosystem strategy rather than a feature: insurers, courts, and newsrooms all need a chain of custody for images, and the company that owns the capture device gets to define the standard. The open question is whether the signature travels intact through every social platform's re-encode — a business-development challenge that cryptography alone cannot solve.

## Commerce Rewired

**The delivery margin was never the point**

DoorDash's Katie Daleo makes the case that [commerce media has reached its moment](https://www.beet.tv/2026/08/katie-daleo-of-doordash-commerce-media-finds-its-moment.html?ref=adjacent.media), which in practice means DoorDash is selling CPG brands the marketing data products derived from years of transaction history. Delivery economics have always been thin; the ledger of who buys which brand of seltzer, how often, at what price, on which night of the week is not. That puts DoorDash on the same path Amazon, Instacart, and Walmart took: the logistics business becomes the data-acquisition cost for an advertising business with software margins. For brand-side buyers, the practical caution is measurement: retail media networks grade their own homework, and the incrementality question that dogged Amazon's ad business is the same one to ask here before the budget shifts.

## Connected World

**Robots don't have to work to be useful to the P&L**

DoorDash is [pushing harder on replacing human couriers with autonomous delivery](https://boingboing.net/2026/08/11/doordash-sprints-towards-replacing-its-drivers-with-robots?ref=adjacent.media), which reads as a technology story and functions as a labor-cost story. Sidewalk delivery robots have been eighteen months away since 2017, and the unit economics still depend on dense, flat, temperate, permissive geographies, a small fraction of the addressable market. But the announcement does work regardless of whether the robots do: it reframes courier pay as a temporary line item during the exact period when gig-labor classification fights are moving through state legislatures. Watch the deployment counts rather than the demos, and watch whether the ad-data business above ends up subsidizing the automation, because that's the version where the math actually closes.

---

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