Some Things I Know (And Some Things I Don’t)

Was this newsletter forwarded to you? Sign up to get it in your inbox.

I have been thinking professionally for over two years. I’ve spent roughly 730 days exclusively focused on understanding which technology companies win and why. It has been the privilege of a lifetime. When I started writing Napkin Math, there were 17,000 readers. There are now over 70,000 of you reading what our team at Every is producing. 

Still, while those numbers are gratifying, they aren’t the point. What matters is truth. What matters is insight. Did my two years of effort elicit any of that? To figure it out, I reread everything I’ve written over the last two years. Then I uploaded all of the text into Claude and ChatGPT to summarize my work and pull out insights, and synthesized their outputs into the sections below. 

You can think of this experiment as a text-based mind palace. Each sentence is a carefully considered door in the hallway of that dwelling. You can open each door by clicking the link and reading my essay about that idea. At the end of this experiment, paying Every subscribers will learn what prompts and tools our team at Every uses to do this sort of research. Let me know if you like this idea—I think there is further to dig. 

Two years of work in 924 words

Which companies win and why? 

At the most fundamental level, the winning company is the one whose culture matches the market it is chasing. 

Great companies sell outcomes, not products. 

By matching the mission of the company to the market it serves, sales become dramatically easier. 

Investors can find these companies by taking a borderline spiritual approach of sensing the fit of the founder to the market. 

Even if you have an incredible product, if the public narrative is bad, it doesn’t matter. 

Corporate finance is storytelling, not math. 

The proper way to analyze a company is to recognize that every number is manipulatable under U.S. accounting laws. 

This is by design and is net beneficial for investors because it is a clear signal about what matters to a management team. 

By understanding the technical underpinnings of each line in its income statement, you know what truly matters to the company. 

What’s the most important competency for a company?

In a fight between quality product and quality distribution, distribution wins.

How you monetize will ultimately determine what you have to build, and whether you need your users to be addicts or missionaries. 

The invisible costs of distribution can sometimes matter more than the legible ones that come with each distribution decision. 

The point of distribution hacks is to cheaply acquire customers and get network effects going as quickly as possible.

Technology investing is about knowing when to be a lemming and when to be a maverick. 

The venture capital community has completely abandoned its spirit of adventure in favor of lemming behavior, doing whatever is popular with limited partners. 

This leaves a space for investors who are willing to embrace scientific risk and for founders who jump off the VC treadmill. 

Most surprisingly, there appears to be no correlation between interest rates and large technology companies being built. 

Risk is too often misunderstood in tech and is substituted for heuristics. 

Boundary-pushing platforms and publishers win. 

Whether you are a creator or a platform, the internet’s attention competition dynamics force all participants to ever-greater levels of moral promiscuity. 

Consumer addiction is the blood sacrifice required for platforms to grow, which is why every one of them is pursuing an algorithmic TikTok-influenced “For You”-style feed. 

The only way for platforms to get off that treadmill is to vertically integrate, but no company has successfully pulled that off yet. 

As an individual actor on the internet’s grand stage, you should recognize that we are still early in understanding its impact on the world, so you should take big creative swings. 

Ultimately, if you aren’t a platform company, being a media company is very, very hard. 

Devote yourself to the cause of your life. 

While it is tempting to think that the next promotion, job, or possession will fill that hole inside of you, no earthly thing ever can. 

In tech, too many people try to make themselves feel whole by working at a startup, despite the choice being deeply irrational. 

Even if you use the best software available, you are never productive enough to be happy. 

You have to live the unoptimized life. 

You cannot sacrifice your family on the altar of your ambition. 

Power is tempting but only invites bottom-feeders to suck you dry. 

True happiness is found in spiritual peace, loving family, and financial security. 

True fulfillment doesn’t come from having bigger things; it comes from being made more full by simple things. 

Utility maximization is futile. Devotion to being you is all that matters. 

Software platforms always win in the long run. 

Software companies win by becoming platforms, not point solutions. 

The best software companies progress beyond a per-seat-sold model to directly tie their revenue to the outcome they are providing. 

As AI makes creating new software cheaper, it is increasingly challenging for horizontal point solutions competitors to retain market share against full-stack vertical competitors because the former don’t hold the mission-critical data. 

AI makes creation go to zero. 

The internet made it free to send digital goods to users, and AI takes creation costs down to zero. 

Companies that are able to easily distribute AI tech to users will take the majority of the profit pool. 

Value will be found in replacing entire trains of thought, or in more corporate terms, automating away entire workflows. 

AI won’t enable one-person billion-dollar companies, but AI agents will give knowledge workers superpowers to get more done. 

Crypto’s problems are fundamental to the technology, not the culture surrounding it. 

While there are many promising, nerdy, and weird things (always three positive signals in startups) that tokens can do, this potential cannot be fulfilled. 

Because each token’s utility is compared on the basis of financial return versus social function, the cool stuff crypto can do gets abstracted away, reducing its functionality to a single metric: internal rate of return (IRR). 

If you are selling to a crypto-native audience, a crypto-native product can be a good strategy—there are just a relatively small number of use cases. 

The crypto doom loop of early adopters cycling through various tokens to juice returns is ever-present and unlikely to be solved. 

Making things cheaper is a moral necessity. 

Technology companies are an ethically and spiritually righteous pursuit because they make goods and services cheaper for everyone. 

The ability to build something beautiful and share it with the world is one of life’s highest callings. 

Face computers rule. 

The Vision Pro is a genuinely novel invention that will dramatically improve our relationship with screens and applications by allowing us to be more present. 

How to pull insights out of your own writing

This exercise was one of the more gratifying things I’ve ever done, and AI was key to making it happen. One of AI’s greatest strengths is its ability to pull insights out of large bodies of data. Even if you’re not a writer, doing a version of this with your journals, emails, or other writing can generate incredible understanding. If you want to do it yourself, here’s how: 

Claude Opus is best for writing tasks, but you could also use ChatGPT or your LLM of choice. 

Step 1. Copy and paste all the text you want into the prompt. If you are using Claude, you should be able to feed it about 250 pages of text. 

Step 2. Prompt Claude with, “I’ve given you a bunch of essays [journal posts, blogs, etc]. Can you please find the title for each? After each title please put a one-sentence summary for each idea. The format should look like this: ‘You Probably Shouldn’t Work at a Startup: The risk/reward ratio doesn’t make sense for most people because the compensation is lower than you think.’” 

Step 3. If you’re lucky, you’ll get a great result the first time, but more likely is that your LLM will be lazy and refuse the task. You may have to ask it to double-check its work. As I’ve written in the past, these systems are the world’s smartest intern. Sometimes you’ll have to cajole them.

For example, Dan had to yell at his Claude to do this right. 

Step 4. By the end, your output should be a list of titles and primary ideas. If you want to analyze more than 250 pages, repeat this process. Eventually you’ll have all of your ideas in a row, with a snappy summary for each.

Step 5. From there, input this prompt: 

“I’ve listed a variety of blog post titles and ideas. Can you please group them by subject matter? I’m writing an essay highlighting all my ideas.”

The LLM will provide a better response if you give it an example, but you should get a pretty good result the first time. I always like to throw in a “Can you please do this better?” a few times to see how the AI will better itself. You’ll be surprised by how frequently that improves the result. 

Have fun!