Good morning everyone, happy Friday, and before we get down to business, I want you to take a quick guess...

Do you think you are closer in size to the biggest thing there is, or the smallest? The biggest being the entire observable universe. The smallest being a Planck length, i.e the point where distance stops meaning anything at all. Complete opposite ends of the cosmic and the quantum scale. Which are you closer to?

What do you think—are we big or small in the grand scheme of things?

Fine, I’ll spoil it...

You are 200 million times closer to the size of the universe. Going up from you to everything is a jump of 27 zeros. Going down from you to the smallest possible thing is 35 zeros. Not close. Not remotely.

So you are, speaking matter-of-fact between what’s infinite and what’s invisible, bloody enormous. Think about that the next time you compare yourself to someone. There’s a fun interactive tool to play with scale here, see how big and small things really are

Moving on, big guy…

In the next 5 minutes over your coffee: why a food delivery app went and got itself an air carrier certificate, the reason Apple would rather rent you an iPhone, the study on why your team doesn’t ship any faster than last year, the startup buying its customers instead of selling to them, and some interesting information on who is and who isn’t lucky.

Because it’s the future guys and DoorDash know you want your next burrito delivered automatically by robots. So, they got FAA approval this week and launched DoorDash Air (which is a very hard/rare certification) allowing them to expand their autonomous business. What’s interesting to me is more so the timeline in that they’ve spent the last four years running drone delivery with deep partners, and they went and built their own full stack aircraft business anyway.

Four years of partnership was school fees, and what it taught them was that the hard part isn’t flying. It’s the handoff between drone and hungry person and the logistics around that. So they bought the education, then went through the hoops to get a certification that takes years and five stages to earn. That’s a new moat for them now that the bureaucracy of the FAA maintains on their behalf. Just eight operators is the entire competitive set.

Spend a moment thinking about everything your product currently rents from a partner—which ones would you have to own for the thing to be defensible? Do any of them sit behind a license, cert, or approval. Sometimes the best moats are the ones somebody else makes and protects for you.

I have ~40K followers on Substack. But, 12K of them are rented to me by the platform. They see me when the algorithm decides they should, and I have no direct line to any of them. And algorithms are hard to make your friend.

That’s one of the big reasons I rebuilt The Diff here on beehiiv in January—I wrote the story here if you want it. They’re the creator platform that take 0% of anything I earn vs 10%, no followers just an email list, and there’s a real API and MCP so I can pull my own data out whenever I feel like it. Followers inside a feed belong to the feed.

If you’re an entrepreneur, creator, publisher or a business building, beehiiv helps you grow 2.75x faster with built-in referrals, recommendations and analytics. If you’re serious about scaling a community and monetizing it, start with beehiiv. You can use code THEDIFF30 for 30% off for 3 months.

A study shows that you can become more lucky, because lucky people are the ones who expect good things to happen.

Richard Wiseman spent a decade on 400 people who described themselves as consistently lucky or unlucky, then handed each of them a newspaper and asked how many photographs were inside. The unlucky group took about two minutes. The lucky group took seconds, because page two carried a half-page ad in type over two inches high: stop counting, there are 43 photographs in this newspaper.

Halfway through sat a second one offering £250 to anyone who told the experimenter they’d seen it. The unlucky group counted right past both. Asked afterwards if they’d noticed anything odd, they said no. His personality data explains it: the unlucky scored roughly twice as anxious, and tension narrows attention until anything unexpected gets filtered out.

The lucky people in his data also changed their routes to work, talked to more strangers and broke their routines deliberately, which just raises how much unexpected things crosses their path in the first place. So luck is an exposure rate times a noticing rate, and both are things you can change. He built a “Luck School” to test that and 80% of people reported their luck went up, by an estimated 40%.

Break some routines. Create some more serendipity. Talk to someone you might not have. Practice the skill of observation. Start thinking you’re more lucky, because remember, you’re actually enormous.

I grabbed a coffee yesterday with someone, which is exactly the type of meeting my notetaker has always been a problem for—I’ve never loved a notebook out while face to face with someone, it just feels impersonal to me.

But Granola shipped on Apple Watch this week. Now you can tap your wrist, have the chat while just being present, tap it again, and done—you have great notes. If you do in-persons, whiteboard sessions or walking 1:1s, this is a big deal.

Economists at MIT and Wharton did a study on public GitHub records for 100,000+ devs. TLDR—way more commits as you’d guess, way more code, but no increase in total usage at all and the bottle neck of releases just moved to a different point of the pipeline.

It’s the weak-link hypothesis. AI got very good at the one step, but that step was not the real bottleneck, and everything downstream still runs at human speed: PR review, integration, someone deciding this is good enough to go out. Their estimated elasticity of substitution between AI and human effort is 0.25, which just means these things are complements, not replacements. The constraint when it came to building didn’t go away it just shifted to judgment, where it’s way harder to buy your way out of it.

Put more dollars of AI budget into review and release rather than generation. Generation is not the problem. Whichever queue in your team is longest right now, that’s the one to point at. It doesn’t help to make more of something nobody can approve any faster/

I learnt about this company, Dwelly, who’ve raised $263M in less than 6 months. And it isn’t SaaS. What they do is buy UK property leasing agencies outright, then they install their own AI over the top to make them software-first. So far they’re acquired 17 agencies with 15,000 properties, which comes with £350M of rent collected a year. There are 20K more of these agencies in the UK, most running on spreadsheets, phone calls and vibes. But why this model?

EQT’s line was that the era of aggregating agencies purely for financial returns is over. What’s changed is the arbitrage. 3x throughput per leasing manager is what pays for the acquisitions, so the AI isn’t the product here, it’s the margin. And when your ICP is a six-person shop that will never buy software, buying them could be faster than selling to them. That’s their bet.

When grinding away at a fragmented, people-heavy market, perhaps price the acquisition next to the subscription and see which number is smaller. Then go look at how they structured it, because you can’t roll anything up on equity alone.

Apple Upgrade landed Tuesday. It’s a leasing program for anything Apple, like $17.99/mo for an iPhone.

What it doesn’t do though is cut the price of anything. They’re doing it because the price is just getting too high YoY for people to keep buying. In June they raised prices and claimed the AI-driven memory shortage. RAM has ballooned from $2.80/GB last year, to $12 already as chips get sent first to the big-bucks AI data centers. So, the shelf prices are going up and up and Apple knows the only way to keep us buying is brining the number you pay each month down.

When costs spike you get three doors: eat it, raise the sticker, or change the payment shape. Door three does two jobs at once. It hides the increase inside a smaller monthly figure and it turns “new phone” from a $1,099 decision into a subscription decision, which people make far more often and easily.

If your costs are climbing, think about the payment shape before you touch the price. Same revenue, smaller perceived number, and you get to reset the renewal moment on your own terms. Read Apple’s details though, it looks like early termination is described as “substantial” and the package you pick is a promise you’re stuck honoring for three years.

🛠️ A tool I found — a free YouTube clipper. Paste a URL, pull the exact segment you want, and stop sending people 40-minute videos with “watch from 23:10” in the message. → Grab your clip

📺 A video I loved — why “Ozymandias” from Breaking Bad is the only perfect episode in TV history. It sat at a flawless 10/10 on IMDb for thirteen years, and this breaks down how one hour of TV takes Walter White apart piece by piece until there’s nothing left but a broken man in the dirt. → Watch it

📬 A read I enjoyed — Ruben from How to AI on what he learnt from 1,800 hours of Claude. For example, send a screenshot instead of a description. → Read it