The Weekend Leverage
MODELS MODELS MODELS
This was one of the craziest, most hectic weeks I’ve ever seen for technology. Multibillion dollar acquisitions, multiple AI models and product releases, plus, Elon had four more kids. That last one isn’t real, but also, maybe? Who knows with that guy.
If there was one thing that this week proved, it’s that we are wholly and truly in the midst of a platform shift. Love ‘em or hate ‘em, LLMs are the next big thing. When markets are this dynamic and the future is this murky, fortunes are created and lost by the decisions you’ll make over the next 24 months. If you are agentic, aggressive, and have just a pinch of arrogance, there’s never been a better time to be a founder.
Here are the opportunities and challenges I’m seeing. But first, today’s issue is brought to you by the founder’s bank of choice, Mercury.
The best software products are the ones that care about the edges. The perfect complementary color here, the delightful automation there. Until I found Mercury, I had never experienced that in my banking. My old bank account felt like I was just waiting for an error to occur that would make me stress about money. That doesn’t happen to me anymore. Mercury’s product suite is full of fast automations and delightful touches. Plus, I can rest assured my money’s safe with up to $5 million in FDIC insurance thanks to their partner banks’ sweep networks. If you want banking software that's extraordinary to use, check out Mercury here.
(Mercury is a fintech company, not an FDIC-insured bank. Checking and savings accounts are provided through our bank partners Choice Financial Group, Column N.A., and Evolve Bank & Trust; Members FDIC. Deposit insurance covers the failure of an insured bank. Checking and savings account deposits may be held by sweep network banks. Certain conditions must be satisfied for pass-through insurance to apply. Learn more here.)
MY RESEARCH
In defense of bad businesses. Too often we think of good companies as those with judicious profit margins, or eye-popping growth numbers. Founders think that the winning condition of the game of building is “changing the world” or just building a big-ass company. This is totally backwards. Being a founder means you get to create the tasks, products, and company that you want. A good business is simply one that is an honest expression of your desires. Which is to say, when God gives you a blank canvas, why do you choose to paint B2B SaaS? More here.
OpenAI punches Apple in the throat. By acquiring Jony Ive’s io for $6.5 billion in equity, the company signalled that it wanted to be totally vertically integrated from server to consumer device. Assuming the device works, this is a remarkably good idea. ChatGPT’s memory feature will allow for an out-of-the-box personalization so magical that every other device will appear to be dumb. Still, there are multiple reasons to doubt the device, the team, and the strategy. Paying subscribers get access to my thoughts on what happens next.
THE BIG THREE STORIES
Autonomous cars, slowly, then suddenly. Waymo is one of those companies that makes me giddy to think about. Cars and firearms kill roughly the same number of people every year in the U.S.—about 40,000 each depending on how you slice the data. I don’t think technologists will ever solve the political quagmire that is gun control, but they can certainly build a profitable autonomous car service. Waymo announced this week that they have successfully completed 10 million paid trips, with 5 million of those in the last five months. They’ve started to map out the town here in Boston, where I live, and I cannot WAIT to have them on the streets. If policy makers actually cared about saving American lives, they would give Waymo the biggest tax break you’ve ever seen and scale these up nationwide. Studies have shown that Waymo's vehicles have 85% fewer accidents compared to human-driven vehicles.
Claude 4 is a bet that capability matters more than context. There are two stories in the foundation-model race that are being sold to investors. The first, exemplified by OpenAI, is that having memory and exclusive data access will make models far more performative than any other form of software. The second vision, as demonstrated by Anthropic’s launch this week, is that code is all that matters. If you can get the AI models good enough at writing code and software engineering, the models can recursively self-improve by writing code that improves themselves. From there, it should be trivial to expand to other areas.
It is still unclear to me which story is correct. (Likely it is both). That said, this quote from a CNBC article covering the launch raised my eyebrows, “Anthropic stopped investing in chatbots at the end of last year and has instead focused on improving Claude’s ability to do complex tasks like research and coding, even writing whole code bases, according to Jared Kaplan, Anthropic’s chief science officer.”
This is Anthropic bidding adieu to the consumer market and swinging for the fences. Contrast that to OpenAI trying to do everything!
OnlyFans is officially the most successful creator economy startup—ever. Reuters reported this week that the owner is looking to sell the firm for about $8 billion. The company went from $300 million in revenue in 2019 to over $6 billion today. It has done all this while having fewer than 100 full-time employees! The company has cornered the adult entertainment market for creators. It doesn’t provide demand aggregation like social media, and creators are left on their own to grow their business. Despite those limitations, OnlyFans enjoys a 20% cut of their creators’ revenue. Perhaps the correct way to view this business is as a company that takes on the risk for CSAM materials and is de-banked on behalf of creators. Since no one else in the world is willing to take that on, it has been able to profit wildly. Still, that adult entertainment is the most successful creator economy company of the last decade should make us pause and consider what we are actually doing here. Surely there are better companies that should be built?
VISUAL SIGNAL
Alarm bells should be going off—AIs are far, far more convincing than the average human. A new study out of Princeton had 900 participants paired with either a human debate partner or GPT4. Some of the study participants, both AI and human, had access to their debate partner’s demographic data like gender, employment, ethnicity, and political affiliation.
The study found that the AI was 64.4% more persuasive than humans when given that personal information. Look at the impact here!
The evidence that AI is so good at debate and changing people’s minds should frighten you (it does me). I didn’t think that people would be so easily persuaded, but it turns out that next token prediction is a remarkably effective way to hack the human brain. The obvious concerns are around nudging people toward extremism, but I think the more dangerous, more subtle threat is sycophancy. These systems are mysterious, alien things, and if we reward them for “accomplishing goal” or “make user happy” people will find talking with them far more engaging than interacting with human beings. It sounds like sci-fi, but here we have a study using outdated models that proves I’m right. Eeek.
There is an opportunity for startups to do something here. They could create bots that want to create nuanced and positive behavior from their users. I’ve been using Auren the last few weeks and have found that it is about as effective as therapy. It isn’t as smart, but because there is absolutely zero fear of judgement, I find I can be more honest with it (and myself) than I could when I was in therapy. Would love to see what the professional versions of this are. What could a business coach that has access to your health data, a startup’s P&L, or meeting transcripts do?
TASTEMAKER
Rick Rubin, vibe coded. Depending on my level of caffeination, I either hate this or love it, but either way, it is worth checking out. Anthropic partnered with music producer Rick Rubin to release an ebook titled, “The Way of Code: The Timeless Art of Vibe Coding.” I would say the possibility that Rubin actually wrote this is close to zero, but kudos to his ghostwriting team. This sort of stunt rhymes with what crypto was pulling in 2022 with Super Bowl ads and celebrity endorsements.
I am also a fan of his ghost writer’s other work. The Creative Act, Rubin’s other book, is excellent. When I reviewed it in 2023, I wrote, “If Rick Rubin started a cult, I would be its first card-carrying member. Bathe me in goat’s blood, howl at the full moon, sacrifice a virgin to the god of music, I don’t care, I’ll do it—this dude is onto something.” It is a useful companion if you are trying to increase the creativity in your life.