Good morning everyone, happy Friday, and I read this week that Ozempic might actually slow aging?

TLDR, there’s published evidence that semaglutide, the molecule in Ozempic and her friends like Wegovy, slowed markers of biological aging in a trial of adults with HIV. It’s a small trial, it’s early days, and “aging markers” don’t mean living longer. Still interesting to me though because this drug built to manage blood sugar keeps picking up new jobs—weight the big obvious one, then heart, kidneys, maybe addiction, now possibly even time itself? Diabetes was it’s wedge, and the wedge is seldom the final market.

In the next 5 minutes over your coffee: the 50-year-old math problem that fell in under an hour, the inventory practice that just became illegal in Europe, the 99¢ decision that built a $100B business, the Fable 5 playbook most people are getting wrong, and why everything you own keeps getting worse.

There’s a famous math problem that’s been around since 1973, and nobody has been able to crack it. Earlier this month it got cracked in just 58 minutes.

OpenAI’s Sol Ultra proved the thing using 64 AIs working in parallel. Each was told to explore a different attack and was kept from copying each other, with a boss “critic squad” agent whose only job was tearing every candidate proof apart.

Hard problems are hard for different reasons. Some are hard because they need a genius. Others are hard because they need tremendous time and patience—and no human has the patience. Computers don’t run out of patience though. Nobody thinks this breaks Bitcoin or your bank’s encryption tomorrow (those are thought to be hard in a deeper way and I really hope so!), but there’s a huge backlog of unsolved problems in math, materials, and drug design (is this why Ozempic is becoming so damn powerful?!) that were only ever just waiting for someone tireless enough. That backlog is now getting close to being cleared up.

The multi agent attack process is the structure you can copy for anything you’re stuck on. If it can solve a math problem like this, it can solve your problem. For example, you could take a complex bug that’s been hard to reproduce or fix. Open Claude Code, kick off five attempts with five different strategies and don’t let them see each other’s work. Then run a final pass that attacks all five answers. And do what OpenAI’s prompt did (see below)—tell the model a solution definitely exists, and refuse to let it give up.

If today’s edition feels new, that’s because I rebuilt the formatting and UX this week—new section headers I geeked out on, new interactive buttons on web, the whole feel. I’m obsessive about tinkering with this thing, and beehiiv’s editor is the only place I’ve found that lets me customize every block until the thing is fully mine. I simply couldn’t do this type of building on Substack.

And if you’re starting from zero, you don’t need my obsession either—their AI editor will build your website or newsletter for you, one of many neat features. beehiiv is the best place to build your brand and audience, and keep everything you earn. You can start for free.

My code THEDIFF30 gets you 30% off your first 3 months when you upgrade.

Clothing companies in the EU, as of Sunday, can’t destroy unsold product anymore. I had no idea, but apparently Burberry alone was burning £28M of their own product in a single year just to protect its pricing (i.e controlling supply).

This means they have to either resell, donate, repair, or recycle. Good. Before the ban, an estimated 4–9% of unsold stuff in Europe was just being incinerated before ever being worn, which for the eco conscious folks here is around 5.6 million tonnes of CO₂ a year, close to Sweden’s entire footprint.

Rules and bans are opportunity that can mint new categories with buyers who have no choice. GDPR did this too—OneTrust went from founding to a multi-billion-dollar valuation in about five years riding one deadline. Now every large apparel company selling into Europe suddenly needs better demand forecasting, resale and recommerce infrastructure, returns processing, and inventory analytics. None of that is optional anymore and there’s $$$ on the table.

Maybe you’re going to go build something for that right now, but if not, more broadly I thing the takeaway is it’s good to keep an eye on regulation calendars the way you watch platform shifts. They’re product roadmaps to the entrepreneurial eye with dates attached and desperate customers on the other side. Find the bans coming to your industry and get there first.

Last newsletter, I wanted expert eyes on my draft before it went out. I have Claude help edit anyway, so I just typed that I wanted a pro marketer to review the copy for click-through improvements (because that’s my ad business). Tendem’s MCP picked the task up in the same chat, quoted me a price, I approved it, and a vetted conversion expert sent back a feedback doc. I shipped the suggestions and clicks came in about 30% higher than my recent average.

Tendem (by Toloka) is one place to send the work AI can’t finish alone. Describe the task in plain words, see the price before anything starts, and 10,000+ vetted experts plus multi-layer QA hand it back finished with a result you can use. The MCP connector plugs into any AI tool in about 2 minutes, then you just add “Use Tendem…” to any prompt.

My link starts you with $60 in credit, and your first 3 tasks are half price.

When Apple launched the iTunes store they thought they’d sell a million songs in six months. They did it in six days. My friend Tom broke the 2003 launch down with details I’d never seen. Jobs spent 18 months lobbying the five major labels and won by making Apple look smaller—iTunes as a Mac-only experiment (Mac had ~3% of the PC market), on one-year contracts the labels could easily walk from. The 99¢ price wasn’t even his idea. Warner proposed it, most labels wanted $3.49, and one exec recalled Jobs looking at them “like we just gave him a gift.”

Apple on purpose made essentially nothing per song. Songs sold iPods (31% → 74% US share in 18 months), and iPods built the real asset—400 million credit cards on file, more than Amazon and PayPal combined, attached to people trained to one-tap buy. Seven years later the App Store launched onto those rails with zero payment friction, the same thing that had killed every mobile app marketplace before it.

Have a think about what your product’s implicit asset is. It could be data or habit or stored trust. Often there’s something under the hood. Then ask what happens if you build towards it on purpose—what could that unlock for your business?

The best model in the world gets worse if you try use it like Opus or any other model. That’s not me telling you guys, Anthropic is, saying prompts built for prior Claude models are often too prescriptive for Fable and degrade its output. Most people see Fable, think new and better, and just keep doing what they’ve been doing before because that’s what we did when we got new models before.

Send the same thing, watch it think for five minutes, and decide it’s overhyped. I mentioned the idea of goals before and I want to rehash from a different angle with a linked asset for you.

Doing a few things different can change the output entirely. Tell it to act once it has enough information instead of surveying options. Make it check every progress claim against an actual tool result before reporting, which in Anthropic’s testing nearly killed fabricated status updates. Give every long run a hard stop—a budget cap, a time cap, a done condition.

I compiled everything into one cheatsheet doc you can steal from—the 12 prompting patterns, the orchestrator setup where Fable plans while Sonnet implements and Haiku tests, the markdown memory system, and the migration checklist. Start with the simplest one. Write a long prompt describing your business in real detail, then ask Fable what it would do.

I was in Europe last week and could not stop wondering why buildings are just so much uglier now than they used to be. Like the quality and craft once was so high. Obviously the answer is money, which has led to degradation of quality. This is true in a lot of products…if you can make the same thing for less, why wouldn’t you?

What I learnt this week was how this is a slide down problem though vs a step down problem. The way cost-cutting programs work actually test every downgrade (the thinner steel, the plastic gear where the brass one was) against the most recent version, and each one is approved because it comes back “no detectable difference.” The comparison isn’t run against the product from a decade ago. Each slide down passes. But if you zoom out, worse is obvious.

The essay builds on Robert Pirsig, who spent his life on why quality can be recognized but not defined—his students independently ranked papers the same way every semester without ever agreeing on criteria (“you know what Quality is,” as he put it). Spec sheets measure the object. Star ratings measure opinion, in week one, while the failures arrive in year three. Nothing measures how a thing holds up in a person’s hands across years which is where quality actually lives.

I don’t know to be honest, I just think it’s interesting. We’re all builders though, so maybe it’s just a PSA to keep holding up the quality and craft of our products as best we can. Users notice shortcuts because they compound. Maybe they can’t say why it feels worse because we slide there slowly, but selfish business decisions eventually get noticed. Its a good essay check it out 👇

🛠️ A tool I found — Paradigm. Give it any learning goal and it builds an adaptive course that updates as you go. The founder’s favorite demo is its tutor texting you at 11pm before a final, offering a cheat sheet built from your weakest topics. Free tier forever, built by a 22-year-old who won world robotics titles before finishing three degrees in three years. → Turn any goal into a course

📺 A video I loved — “How To Get Customers So Fast It Feels ILLEGAL,” by Alex Hormozi. The author of $100M Offers on the fastest routes to customers when nobody knows you exist. → Watch it

📬 A read I enjoyed — The Atlantic on how The Odyssey broke the box office. Like can anyone here actually get tickets?! Nolan’s epic sold out IMAX screenings a full year in advance and resellers are selling seats for $1k+. Scarcity, format, and event-making as a distribution strategy. → Read it