Plus: a recipe automation I built this week that saves my family hours, the $50K-$200K SaaS opportunity hiding inside the EU AI Act, and 9 rapid-fire stories you should know about.

That's wrong.

Anthropic won because they pointed Claude at the most boring, most lucrative use case on Earth — writing code for engineers — and they made the model better at it than anyone else. Cursor runs on it. Claude Code is the tool of choice in serious engineering shops. The big banks license it.

OpenAI spent the last 18 months trying to be Apple, Pixar, and Google all at once. Sora, Voice Mode, the rumored phone, the Musk lawsuits. Anthropic spent it building the world's best coding partner and selling it to companies that pay invoices. One of these strategies is now printing $30B in ARR.

Guess which.

The lesson for founders: the highest revenue per dollar of compute isn't AGI demos. It's developers writing CRUD apps faster. The unsexy markets keep winning, and they keep paying.

You can see this pattern in every industry, not just AI.

Costco is the most boring retail strategy on Earth. Membership warehouse. Low margins. Big boxes. No marketing. No e-commerce circus. Same store layout for 40 years. They just announced their 23rd straight year of revenue growth. Quarterly profit beats every analyst estimate. Their stock has outperformed Tesla over the last decade.

Vanguard launched the first index fund in 1976. Wall Street called Jack Bogle stupid for two decades. Hedge funds were clever. Picking stocks was sophisticated. Index funds were "guaranteed mediocrity." Today Vanguard manages $9.5 trillion. Most hedge funds have underperformed the S&P 500 for the last 15 years. Boring beat clever for half a century.

Domino's was nearly bankrupt in 2009. The CEO got on TV and said the pizza was bad. Then they made one boring decision: don't try to be a fancy chain. Be the fastest pizza delivery on Earth. Build the tech for that. Twelve years later, the stock outperformed Apple, Google, and Amazon. They did it by getting better at one boring thing.

AWS was a side project at Amazon nobody believed in. Bezos kept Amazon boring on infrastructure while Google chased Glass, Wave, Stadia, and a dozen other moonshots that died loudly. AWS now generates more profit than the entire rest of Amazon combined. Boring infrastructure beat famous moonshots.

The pattern is everywhere if you start looking. The companies that win their decade pick one boring thing, get insanely good at it, and refuse to be distracted by the trend cycle.

The hard part is that boring requires turning down a steady stream of more exciting opportunities. The phone. The movie. The keynote. The talk-show circuit. Every one of those things is a perfectly reasonable yes. Saying no to all of them year after year takes a kind of monastic discipline that's genuinely rare.

But that's the trade. Boring isn't the easy path. It's the hard path that most teams give up on.

Anthropic just demonstrated, on a public stage, that the hard path still works.

Nobody became a developer to write documentation. But the docs still need to get written — PRDs, README updates, architecture decisions, onboarding guides.

Wispr Flow lets you talk through it instead. Speak naturally about what the code does, how it works, and why you built it that way. Flow formats everything into clean, professional text you can paste into Notion, Confluence, or GitHub.

Used by engineering teams at OpenAI, Vercel, and Clay. 89% of messages sent with zero edits. Works system-wide on Mac, Windows, and iPhone.

Try Wispr Flow free

We're always scrambling for recipes. The week starts and nobody knows what we're cooking. The cart never gets ordered until somebody panics on Sunday night.

So I built an agent that solves it. Every Friday, Claude does the following:

Saturday morning we open the cart, click checkout, and the week is done. No deciding, no list-making, no Sunday-night panic.

This is the unglamorous side of agents that doesn't get talked about. Forget enterprise transformation slide decks. The biggest immediate value of AI agents is doing the dumb logistical work that lives between systems — the calendar invites, the shared docs, the carts, the things your spouse asks about that always fall through the cracks.

If you want the actual Claude Skill I built, reply to this email and I'll send it. If enough people want it I'll start sharing more of the automations I'm running for our family. Let me know.

Jurisdictional AI Compliance-as-a-Service

Goldman just blocked Claude for its Hong Kong staff. The EU AI Act high-risk deadline hits August 2, 2026 — covering AI used in hiring, screening, performance reviews, and termination. Banks, law firms, hospitals, and global consultancies all suddenly need to know: which model is allowed where, by which employee, doing what task, with what audit trail.

The play: A SaaS layer that sits in front of your AI stack and enforces jurisdictional, regulatory, and contractual rules. Think of it as a routing engine plus an audit log. User in Hong Kong asks Claude something? Blocked, logged, alternative model offered. EU candidate's resume hitting an AI screener? Auto-flagged for human review with a compliance receipt. A single source of truth for "did we use AI legally in this transaction."

What you'd actually do: Start by mapping the top 20 enterprise AI vendors against the EU AI Act, NYC's automated employment decision tool law, GDPR Article 22, and the location-restriction clauses in major model contracts. Build a Chrome extension and an API gateway. Sell it to compliance and legal teams at mid-market financial firms first — the ones too small to build it in-house and too regulated to ignore it. Charge $50K-$200K per year per company. Cold outreach the General Counsel, not the CTO.

Why it works: The August deadline is real and the fines are 7% of global revenue. Goldman just publicly demonstrated that geo-blocking is now a thing enterprises do. And there is exactly zero off-the-shelf solution today. The first 50 customers will pay anything for peace of mind. Two years out, this is acquisition bait for ServiceNow, OneTrust, or Workday.

That's the briefing. Now go build something.

Why it works: The August deadline is real and the fines are 7% of global revenue. Goldman just publicly demonstrated that geo-blocking is now a thing enterprises do. And there is exactly zero off-the-shelf solution today. The first 50 customers will pay anything for peace of mind. Two years out, this is acquisition bait for ServiceNow, OneTrust, or Workday.

According to analyst Ming-Chi Kuo, OpenAI is working with MediaTek and Qualcomm on a smartphone-class device, designed by Jony Ive's team, that flips the iPhone paradigm on its head. Instead of a grid of apps you tap, you'd just talk to an agent and it does everything for you. Booking a flight. Replying to email. Ordering dinner. No apps, no taps.

Mass production is reportedly slated for 2028, with specs locked by end of 2026. This is the AI-first device Sam Altman has been hinting at for two years finally coming into focus.

Why it matters: This is OpenAI's hedge against losing the model layer. If Anthropic and Google are going to commodify foundation models, OpenAI needs distribution. Hardware is the ultimate distribution. The question: do you really want Sam Altman running the OS that holds your contacts, your calendar, and your Face ID?

Awkward earnings call moment of the week. On Wednesday, Musk admitted that millions of Tesla owners with "Hardware 3" cars will need both a new self-driving computer AND new cameras to get unsupervised Full Self-Driving. He'd spent years promising the opposite.

The fix is so logistically painful Tesla is now exploring building micro-factories in major cities just to handle the upgrade volume. Meanwhile, the company finally taped out its AI5 chip — almost two years late.

Why it matters: "Buy the car, the software upgrade comes later" was Tesla's whole pitch for half a decade. Owners paid $10K-$15K extra on that promise. Many are about to find out the answer is "you also need to buy a new car." Class action lawyers are sharpening pencils as we speak.

Rogo, the AI agent for investment bankers, closed a $160M Series D led by Kleiner Perkins. 35,000 finance pros already use it across Rothschild, Jefferies, Lazard, Moelis and Nomura. Wall Street is officially deploying agents on the trading floor.

Aidoc, a clinical AI company for radiology, raised $150M Series E led by Goldman Sachs Growth Equity. Total funding now over $500M. NVIDIA's VC arm participated.

SpaceX Falcon Heavy launched for the first time in 18 months on Wednesday, dropping ViaSat-3 F3 into geostationary orbit. Falcon Heavy used to be the company's flagship. Starship has eaten its lunch.

Motorola is launching the Razr 2026 today — clamshell foldable with a 6.9-inch AMOLED, dual 50MP rear cameras, MediaTek Dimensity chip. Three colors: brown, green, purple.

Meta is cutting 8,000 more jobs, about 10% of its workforce, plus removing another 6,000 open roles. 2026 has now seen 100,443 tech layoffs, with 47.9% explicitly attributed to AI replacing roles.

AWS launched the Amazon Quick desktop app — an AI productivity assistant that connects to email, calendar, Slack, and your local files to flag stuff and build agents. It's basically AWS taking a swing at Microsoft Copilot's market.

Mistral launched Workflows in public preview — already being used by ASML, ABANCA, CMA-CGM, France Travail, La Banque Postale, and Moeve. Europe's AI champion is finally shipping product, not just press releases.

Japan is piloting humanoid robots to handle airport baggage at major hubs. The labor crunch over there has officially crossed into "yes, robots, please."

Google confirmed Gemini will fully replace Assistant on Android in 2026. Twelve years of "Hey Google" memories, headed for the trash bin.