Good morning everyone, happy Friday, and it turns out you don’t really choose things. You just choose them again.

A team at TU Dresden went through 15 separate datasets covering 700+ people and published the result in Communications Psychology. Simply repeating a choice makes you rate that option as better afterwards. Not more familiar. Better. And it holds up in brand new situations, with equally good or objectively superior options sitting right there in front of you.

Your brain mostly doesn’t re-run the comparison. It remembers what it picked last time, picks it again, then backfills a reason for you.

Which means the thing winning in your category might not be winning because it’s better. It’s winning because it won. We’ve all been reading incumbency as a quality signal when it’s mostly a repetition signal.

Rough news if you’re the challenger, because being better honestly isn’t enough on its own. But the flip side is the good part → get chosen once, properly, and the rating comes along for the ride afterwards.

In the next 5 minutes over your coffee: the sentence the internet can’t stop sharing, the founder who spent months choosing a market before writing any code, why one extra click beats your entire pricing page, the reason everything in this email might be worthless to you, and what’s left of your job once the agents do the building.

Gumloop first paid a street trumpet player to busk outside their office with a Gumloop sign. Then he hired some more. More buskers across the city during a conference. And then he posted the whole thing with one line: “What if we spent our marketing budget on musicians instead?”. People cheered for a workflow automation company in the comments.

Tom Orbach says that sentence is the brilliant move, and that it’s been working for decades. Sticker Mule spends 5% of their ad budget giving randoms on the street $100 bucks, which their CEO candidly says “it’s not philanthropy, it’s marketing.” Reddit “couldn’t afford” a real Super Bowl slot so they bought just five seconds and used them to say exactly that and became the second most Googled ad of the night. Domino’s gave towns $5k to fill in their potholes and got 137k nominations for local government.

In all of these examples, the brilliance is buying a story about refusing to buy reach. Why? Because reach itself is way more expensive vs the cheaper option of trying to get the audience to handle for you. It works because people are fried on being advertised at, and in this case the refusal to do that to them is the ad.

Plus, nobody’s actually spending their entire budget on this stuff. You either ask it as a question or just say “this month’s ad budget,” and it’s completely true while sounding ten times bigger than it is.

Pick the smallest real thing you can buy that isn’t an ad. Pay off every parking ticket on one street near your customers. Rename a drink at the coffee shop they go to. Something random to what you do but enjoyable to people. Then write the sentence. Tom has 13 of these with companies that pulled it off 👇

Right now everybody is spinning the wheels trying to automate everything. But automations only work once you and people have done the work to decide what needs the automating.

Tendem is still the only thing I’ve found running the You → Your AI → Expert Real Human → and back to You loop. You describe the work inside your favorite AI, see a price, approve it, and one of 10,000+ vetted experts sends it back done.

They do such a great job (honestly) of combining AI, expert human review and multi-layer QA into results you can trust. Meaning less time cleaning up sloppy or incorrect AI answers.

It beats pure AI output, and it beats it hardest on the things AI is worst at: the work that needs somebody who actually knows the domain to put their name on it.

Derrick Reimer sold Drip then moved on to build Level which was “a thoughtful alternative to Slack.” It didn’t work (maybe because everyone had already chosen Slack). Then instead of sprinting at the next idea he did what most of of us sprint through → he spent months hard auditing SaaS markets before building anything at all.

His three criteria were a) Proven demand, b) Recurring revenue and c) One painful point of differentiation.

His audit led to SavvyCal, a scheduling tool built on the small indignity that sending someone your booking link feels a bit one-sided. Bootstrapped to five-figure MRR.

If you’re shipping into a market with no urgency and no willingness to pay speed doesn’t help you win.

Which is awkward when all of us have close to infinite build capacity and roughly zero market-selection discipline. We’re lucky though—we’re living in the era where we can ship the wrong thing faster than anyone in history. The constraint has moved upstream of the code. And honestly i don’t think I’ve moved with it.

Spend time thinking about markets that are interesting to you and score rubrik them on his three. If none of them score that’s a well-executed exercise👇

Derrick Reimer’s full story (the guy above) is full of attempts on market. He makes a SaaS and swings, and fails and wins. The one thing though that I bet carried across all of them was what he learnt about who buys, why, and what they’d pay.

That knowledge has to live somewhere. I think Attio is a great place to save and put that knowledge to work. It’s my CRM of choice.

We all know products come and go. But your hard-earned record of customers and partners and connections shouldn’t.

Economists at Harvard Business School ran a study around cookie banners and the finding was that with zero design manipulation, people just accept everything more than half the time. Put an option (even a desirable one or for avoidance of an annoying thing) behind one extra click and they shift hard toward whatever’s easier to reach. The decision goes straight to whatever the default was, and when folks were surveyed afterwards they overwhelmingly where confused about what the defaults even were.

I get it, we like defaults…remember our lazy brain? Nobody’s preferences changed around cookies and woke up feeling different about being tracked. The only thing that moved was the default of the screen.

Repetition makes the repeated option feel better. So your defaults don’t just win the first decision, they do serious compound work and can go on to set up the preference that justifies every decision after it.

Open your product and find all the things toggled off that could be turned on as a default. What steps or flows can you make the standard for people? One click cut away could change a lot 👇

“Strategy is getting yourself into a superior position while at the same time making sure your opponent is in a worse one.” — Dan Koe.

In Chess, you might spend the first half of the game doing nothing that looks like anything. No attacks, no captures, just moving pieces onto squares. Then in an instant, a fork appears “out of nowhere” and that’s it. Game over. In his essay the point is that the fork wasn’t the skill. It was all the boring moves that looked like nothing. Take any one of them off the board and there was no winning tactic.

Tactics flow from a superior position, and tactics are the most freely available thing on the internet. Everyone it giving you tactic. I give you tactics, which is why the list by itself can be worth so little. A tactic that works for someone in a strong position does almost nothing for you in a weak one, yet we mostly copy the tactic and skip the position entirely.

The key is defining what you really want to happen. Once that thing is specific enough it stops being just a goal and starts working for you as a filter, because most tactics you get punted obviously don’t serve your specific goal, so you get to stop reading them. Until you’re clear on what you actually want, you don’t have a strategy at all and are allowing in streams of noise.

Where do you want to be in 18 months? Make it specific enough that it actually disqualifies things. Find your position so your tactics can flow from them. 👇

› beehiiv shipped AEO-ready sites this week because AI bots from OpenAI, Anthropic, Google and Microsoft crawled beehiiv-hosted content 490 million times last month. Stack that on their Cloudflare-built AI Crawl Control and you can see exactly which bots are reading you, how much traffic each one actually sends back, and block the ones taking without giving. Search sends fewer clicks every month so getting quoted in AI is the game now, and beehiiv is the platform helping that happen for you Start your newsletter here

Agents now write about 70% of PostHog’s code, but at least humans still skim ~80% of them. There’s a spectrum though and PostHog is on the right side of it. From AI-assisted at one end to what I naughtily do…which is called the “dark factory”, where agents write everything and no human reads any of it.

Kieran Klaassen from Every is just like me at least. He runs Cora (an AI email client with thousands of paying customers) and hasn’t written a line of code this year. Or read any.

What we both do though is called polish. We put the running app next to the agent, click everything, and say out loud what feels off until it doesn’t. Human tinkering is so so important. If you outsource that you’re cooked IMO.

Agent code reviews and testing won’t be able to tell you what just feels wrong. The coding bottleneck is very quickly sliding from can we build this to can we tell whether it’s actually any good. And the compounding bit Kieran does is neat→ he has every complaint or adjustment he mutters at the browser become a codified rule. Animations resolve toward the click. Scroll bars hide unless invoked. So the next feature starts closer to great without him having to say it again. This is not a replacement of sitting and doing the thing each time, but it helps make the first pass more right.

For the next agent-built feature you ship don’t review the diff. Use it and complain about it. Say what’s off. Be a critical manager who must give feedback. Don’t accept what AI pushes you, pull it towards what you want. Then bonus, write down the three complaints that would apply to any feature and not just this one and turn it into your taste file. This is the part of the job nobody can take from you 👇

🛠️ A tool I found—AgentCat is analytics and observability for MCPs. It replays sessions tool call by tool call so you see the request, the response and the exact error, then groups thousands of those sessions into what the agents were actually trying to do. If you’re shipping an MCP server you’re probably flying blind and this could be the instrument panel. → See how your agents work

📺 A video I loved—Gamma’s founders, Grant Lee and Jon Noronha, on getting to 3 million users in 3 months. Most of it is about how much of that came from just running experiments constantly and actually acting on what users told them, plus their case for AI being a collaborator on creative work → Watch it

📬 A read I enjoyed—A hot take by the team at Every centered around how writing isn’t the only way to think → In Defense of AI Writing