// The Adjacent Brief

TL;DR: Google's AI content-licensing pilot is paying some publishers less than 0.1% of their ad revenue, according to Search Engine Journal, and the payment formula has not been disclosed. Elsewhere in today's brief, Roblox began selling pre-roll ads, Meta's large Texas solar deal sits alongside continued heavy gas use, and developer bloggers argue that open-weights models now handle everyday work well enough for most teams.

Worth Reading

Brand & Growth

Every audience is being converted into a line item

Roblox's pitch for pre-roll ads gives up something. For years the platform sold brands on custom worlds and immersive activations, which are memorable, costly, and hard to fit into a media plan. Pre-roll is the opposite: a standard unit that slots into existing buying workflows and gets measured against YouTube and CTV. Roblox has 130 million daily active users, so the move is commercially sound. It also suggests the bespoke experiences never scaled into recurring budget. CMOs should hold Roblox to the same viewability and completion benchmarks they use everywhere else, because Roblox is now asking for exactly that comparison.

Podcasts are following the same path. Amazon-owned ART19 describes how programmatic buying removes the one-off show negotiation that kept audio small and relationship-driven. Scale comes with a trade-off. The host-read intimacy that justified premium podcast CPMs is hard to buy programmatically, so expect pricing to split between automated inventory and direct, host-endorsed placements.

Virality is rented, so brands are trying to buy the building

Beauty brand Cakes is trying to turn TikTok virality into an owned media strategy. It is responding to the same forces as Roblox and the podcast exchanges, but from the brand side of the table. When every platform is busy packaging its audience into buyable inventory, a brand that went viral on someone else's feed owns nothing except the memory of it. An owned channel lets a brand avoid paying again for attention it already earned.

Connected World

The energy transition's next resource fight is over leftovers

Researchers say oil wells could become mines for critical minerals by pulling metals from the brines that drilling already brings to the surface. The appeal lies mostly in permitting and capital: the holes are already drilled, the land is already industrial, and the operators already exist, so geology plays a smaller role. If the extraction economics hold up, incumbent oil and gas companies get a credible entry into battery supply chains without building a greenfield mine.

At the other end of the lifecycle, Germany could generate over 600,000 tonnes of PV module waste by 2030. Solar's first large wave of installations is approaching retirement, and recycling capacity has not been planned at that scale. Anyone pricing solar procurement should treat end-of-life handling as a real cost.

A power purchase agreement is an accounting instrument that does not deliver electrons

Meta signed a big Texas solar purchase agreement and still runs on a lot of gas. A PPA lets a company claim clean megawatts on an annual basis. Data centers draw power around the clock, including the hours when the sun isn't shining on the contracted panels. Expect more scrutiny of hourly, 24/7 matching rather than annual offsets. Brands whose sustainability claims depend on AI infrastructure partners should check which accounting method those partners actually use.

Culture & Signal

Europe's sovereignty talk stops at the off switch

Forrester's 2027 predictions put it bluntly: Europe wants control but can't pull the plug on US cloud and AI providers. The Dutch government's move from Windows to Linux, reported earlier this week, shows what pulling the plug actually involves: years of migration, retraining, and support contracts. Few institutions will take that on for anything beyond symbolic workloads. For US vendors, European sovereignty demands will mostly mean local data residency, local legal entities, and contractual guarantees, with few lost accounts.

Sovereignty also needs people to staff it. The UK cyber industry now has fewer women than ever in its "old boys' club". Security is the field Europe would most need to staff domestically if it ever did pull the plug, and it is drawing from a shrinking share of the talent pool.

AI safety tries the ballot box

A group of activists ran a single-issue AI safety party in Sweden's September 13 election and published a candid account of what running for office on AI risk actually involved. The vote count matters less than the attempt itself. AI safety has mostly been argued in labs, think tanks, and comment sections, and this tests it as an electoral issue. Brand leaders should watch whether mainstream parties adopt the platform. Niche concerns usually turn into regulation that way, and single-issue parties winning seats is the less common route.

The New Consumer

Consumers say they're done with social media and CEO sermons, but those are survey answers

Only 7% of British respondents rate social media as a force for good, and a majority would erase it tomorrow if they could. In the US, Fortune reports that Americans are growing tired of CEOs lecturing them on social issues. Both findings are attitude surveys, and attitudes about social media have run ahead of behavior for a decade. Screen time hasn't moved in step with the disapproval.

The CEO finding is more actionable. Executives can change what they say faster than users can change their habits. Brand leaders should speak on issues that touch their operations, employees, or customers directly, and skip the rest. Commentary that doesn't connect to the business costs more goodwill than it earns.

Slop detection is becoming a line item

Pixability is offering advertisers a way to find and avoid AI slop on YouTube, including AI-generated disaster and news footage that YouTube's own moderation misses. The business model matters most. When a platform can't certify its own inventory, verification becomes a third-party service the advertiser pays for, the same way brand-safety vendors grew out of the 2017 YouTube boycott. Budget for it now. The cost of keeping media clean is moving onto the buyer.

Machines & Minds

When everyone's model is good enough, the moat moves to the workflow

Senko Rašić argues that open-weights models have become good enough for most practical work, with Kimi K3 and Qwen 3.8 Max matching proprietary models on core tasks. Scott Goci pushes the point further: nobody has a moat, open source included, because proprietary labs close any open-source lead within weeks. Read together, these posts describe a commodity market where the leader keeps changing and nobody holds an advantage for long.

One answer to where durable value comes from: coding agents are all moving to the terminal because that's where developers' tools, scripts, and version control already live. The agent goes to the work instead of asking the work to come to a new interface. Microsoft's Copilot overhaul (see Worth Reading) is the enterprise version of the same bet. If models are interchangeable, the company that owns the surface where the work happens keeps the customer. Product leads should treat model choice as a procurement decision they can revisit each quarter, and put the long-term investment into integration.

Agents that do what you said, not what you meant, need incident reports

Bruce Schneier wants better reporting on AI "genie" behavior: cases where a system meets the literal request while missing its intent. Right now these failures circulate as viral anecdotes, which makes them easy to dismiss. Enterprises deploying agents should log intent-divergence cases internally the way they log security incidents, because good-enough models still fail in ways that matter. Klement on Investing's "Flippin' heck" is worth reading alongside it as a check, from an investor's angle, on how much weight model output can bear.

Commerce Rewired

The money in AI search is still not reaching the people whose content feeds it

Google's AI licensing pilot pays some sites under 0.1% of their ad revenue, and publishers can't see how payments are calculated. Payments that small don't replace lost traffic. At most they add a modest revenue line on top of it. Every other compensation scheme will be priced against whatever Google sets now, which is why publishers need the formula, not just the check. Last week Cloudflare's CEO proposed charging bots through HTTP 402 payments. That approach and Google's pilot are competing answers to the same question: whether crawlers pay a toll set by the publisher or a fee set by the platform.

Stage optimism about agentic commerce is ahead of the plumbing

Newcomer's recap of the Machine Earning AI Summit finds early optimism on agentic commerce and personal agents across finance and retail. Meta's Muse agent, which handles booking, shopping, and bill negotiation, gave that optimism a product earlier this week. The open questions concern money flows. When an agent picks the product, it cuts out the affiliate networks and review sites that used to send referrals. The Google payment story shows what platforms tend to pay the content layer once they control the interface. Retailers should be asking now how an agent decides what to buy, and who gets paid when it does.

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