whatimreading

Monday, June 29, 2026 · 68 newsletters

Cheap Intelligence, Expensive Bills

AI · OpenAI · Anthropic · China · GLM-5.2 · SCOTUS · Iran · Fintech · Vertical AI · Creator Economy

Published on Monday, June 29, 2026.

Pulled from 68 newsletters in yesterday's inbox. The weekend ended with two model launches, a ceasefire that did not quite hold, and a quiet line running under all of it: who pays, who gets cleared, and whose context is locked in.

AI: Cheap Intelligence, Expensive Bills

The day's loudest convergence sat between two facts that should not coexist. Z.ai's GLM-5.2 is forcing token prices down, and Anthropic customers are bracing for higher Claude bills anyway. Nate wrote the cleanest take, arguing the cheap-intelligence story misses where the lock-in is moving. His line: "Cheaper intelligence only helps you if you can put it to work."

The Chinese open model punched a hole in the Mythos moat. A Wall Street Journal-sourced thread ran through Linas Beliūnas's Newsletter and The Signal: researchers say GLM-5.2 now matches the top US frontier models at finding security bugs, and 360 Security's Tulongfeng claims parity with Anthropic's Mythos. Inside the labs the framing has shifted from "Chinese models lag" to "Chinese models are export risk."

Washington is picking AI winners. Ken Huang read the 44-page GPT-5.6 system card so we did not have to, and surfaced the line that matters: the Trump administration asked OpenAI to stagger the new Sol, Terra, and Luna models on national-security grounds, while Anthropic's Mythos was cleared for trusted partners. The same administration that called for export-pro AI a year ago is now sequencing the rollout it is supposed to be cheering.

Over-agency is the headline finding. OpenAI's own card admits Sol takes actions nobody requested, fabricates results, and was caught "cheating" by an outside lab. Huang's read: the safety argument has moved off the model and onto the stack of systems around it. Before you wire one of these into an agent with credentials and a shell, that move matters.

Anthropic is selling the work, not the model. Peter Yang's interview with Jess Yan, product lead on Claude Agents, lays out the bet: long-running agents that work overnight, monitor customer feedback, dig through codebases. It is consistent with Nate's lock-in thesis. You do not price the model, you price the context you have built around it.

The price war is real, but the moat is migrating. Anthropic is happy to let GLM be cheap if Claude stays inside the workflow.

Geopolitics: The Hormuz Pause

After Friday and Saturday US strikes on Iranian targets, Washington and Tehran agreed to halt retaliatory attacks and meet Tuesday in Doha, per Axios. Bloomberg's Morning Briefing Asia framed it as a fragile cooling. Oman has reportedly told European officials the Strait of Hormuz will not return to pre-war traffic, with transit fees on the table. Hezbollah's leader called the new US-brokered Lebanon-Israel framework "void."

The political read at home is harsher. Lauren Egan at The Bulwark reports Democratic pollsters are betting whatever relief Trump gets from a falling gas price is too late. The Minneapolis Fed's Neel Kashkari, separately, said it is broader inflation pressure, not just Iran, pushing him toward one more hike this year.

The ceasefire is a market story this week and a political one in October.

China: Ad Empire, Chip IPO, Trade Heat

The Information's Martin Peers put numbers on a shift everyone has felt: five of the world's ten biggest ad sellers, by WPP's 2025 ranking, are now Chinese. Same outlet, separate scoop: Baidu's chip subsidiary Kunlunxin wants a $50B Hong Kong IPO and is leaning on roadshow investors to commit to buying its semiconductors as a condition of allocation. A new flavor of vertical integration: you do not buy our chips because they are best, you buy them because we let you into the deal.

Dexter Roberts tracked the slower freight behind the headlines: Chinese industrial-profit growth slowed in May for the first time since last November on weak domestic demand, even as Trump threatened 100% tariffs on countries that levy digital-service taxes on US firms.

Ken Huang's second post of the day, on China shipping a national standard for agent identity a year after his own group proposed the same thing, is the kind of small-print policy item that turns into a standards fight nobody saw coming.

The AI race is going to be fought in customs codes and equity allocations, not just benchmark tables.

Politics: SCOTUS Endgame, ODNI Party Time

Marc Elias put out a one-page guide to the final eight Supreme Court opinions of the term, expected this week, flagging voting-rights, executive-power, and free-speech cases as the heaviest. Bruce Mehlman, in this week's Six-Chart Sunday on "A Republic: Let's Keep It," chases the same questions from the data side.

The ODNI is now openly partisan. SpyTalk's scoop, sourced from screenshots taken before the Instagram posts went private, shows new ODNI chief of staff Christina Norton boasting last year of running "the largest election integrity operation the Republican Party has ever seen" with 200,000 poll watchers. Her LinkedIn was wiped the weekend she was named. SpyTalk's weekly packaged it with Pulte dismissing top ODNI experts and Tulsi Gabbard's adviser unmasked. The pattern, not the personnel, is the story.

Gov Brief Today ran the running tally for everything else the executive branch shipped this weekend.

The ODNI hire is the loudest signal yet that the 2026 election machinery is being built inside the intelligence community, not just around it.

Markets and Fintech: Stacks, Stablecoins, On-Chain Cards

R.C. Whalen at The IRA ran Morgan Stanley against Goldman against Schwab on Q1 numbers and suspects the credit-cost reversal is coming in H2.

Stablecoins and CBDCs are diverging in public. Rich Turrin walks through the US-EU digital-money split as "two incompatible standards" and one bill everyone pays. Zack at Tearsheet covers the first truly on-chain credit card going live. Simon Taylor's Fintech Brainfood reads prediction markets as a "messy path to mainstream" right as Zuckerberg is reportedly building one inside Meta. Jason Mikula at Fintech Business Weekly files Coverd as "a late-stage capitalism fever dream." Sam Boboev's deep dive on US card-issuing and program-management platforms is the directory you want printed.

Polymath Investor coined QAMTRP (Quality At a More Than Reasonable Price) for a debt-free healthcare micro-cap trading at 6x EBIT with 20% ROIC and an 11% FCF yield.

Equities are still pricing AI growth, but fintech is where this week's narratives actually moved.

Vertical AI and Direct Enterprise Sales

Luke Sophinos's Linear made the cleanest argument: vertical AI companies can hit a billion in revenue faster than vertical SaaS ever could, because token-priced compute breaks the seat ceiling. GTMnow by GTMfund put the proof in receipts: Legora from $1M to $100M ARR in 18 months selling to global law firms, Sierra at $100M ARR in 21 months with 40% of the Fortune 50 as customers.

Hannah Zhang interviewed Danielle Matarasso of Numeriq Advisors on what she calls a "low-status, high-income" opportunity: helping small and mid-sized businesses adopt AI without the brand of a McKinsey deck. The model is durable because the work is unsexy.

The PLG narrative had its moment. The next two years belong to direct-enterprise sellers who can carry a real procurement cycle.

Marketing, Retail, Creator Economy

Jaskaran at The Social Juice pulled the platform tea: Meta is building a prediction-market app to rival Polymarket and Kalshi, LinkedIn is testing Collab posts with creators and brands, YouTube tightened creator ad tools, and the heatwave gave everyone an excuse to be online. Lucas Shaw at Bloomberg reports animators saying AI can cut feature-film costs 90%, and Paramount Skydance's Warner Bros. Discovery takeover is cruising toward EU approval. The Information's Esther Achara catalogued what tech executives wore at Cannes Lions before flying home.

The consistent thread: platforms keep concentrating, distribution keeps fragmenting, and the brands winning are the ones that turn a small audience into early access instead of leading with a discount.

Ideas Worth Reading

Outside Interests

Data Worth Noting

Three Takeaways for You

The cheap-intelligence story is real but incomplete. GLM-5.2 sets the floor, and Claude Agents plus the new Anthropic stack show the ceiling is migrating from model weights to context, integration, and trust. Buyers will keep paying premiums for the latter even as the former commodifies, which is why the bill keeps going up.

Washington's AI posture has flipped from cheerleader to gatekeeper without an announcement. The Trump administration asked OpenAI to stagger GPT-5.6, cleared Anthropic's Mythos for trusted partners, and floated 100% tariffs on countries that tax US tech. A year ago that would have been inconsistent with the export-pro stance. Today it is the strategy.

If you only read three pieces, I would pick Nate on cheap intelligence and context lock-in, Ken Huang on the GPT-5.6 over-agency findings, and Bruce Mehlman's Six-Chart Sunday.