The Solution Space

Magical Chasms

Today, Nvidia became the fourth largest company in the world, surpassing both Google and Amazon. What’s driving this hype? Why do the public markets - and the masses in general - look at artificial intelligence in a way they never have before? I had a shower thought that might offer a simple explanation.

But first I have to talk about airplanes.

Between 1903 and 1905, the Wright brothers demonstrated controlled flight in a man-made, fixed-wing aircraft for the first time. To the public, it was an incredible feat; to them it was a distillation of a few simple laws of physics and some fine craftsmanship.

Over the next decade or so, aviation was only available to those with homemade contraptions, keeping the general public at a safe distance. In the decades following, aircraft manufacturers began to spring up, and it was then a small contingent of brave pilots who were introduced to manned flight (among them was my grandfather, holding the 5-thousandth some pilot's license in the 1920's, and enjoying some nine or so deadstick landings in his flying career). While they may not have had as much intricate knowledge of the engineering behind the machines - these pilots were reasonably technical themselves, and had to have a solid grasp on the mechanics of flight and various components of the airplane to safely operate them. They too had an understanding of the fact that there was no magic behind an airplane.

It wasn’t until roughly fifty years after the Wright Flyer that aviation had its watershed moment. This was when the jet airliner was introduced, the “jet age” began, and global air travel started to become more readily available to the masses. But this time, the consumer of aviation was the passenger. For the price of a ticket, they could utilize aviation as a service - with all of its technical details entirely abstracted away. For the air traveler who has no knowledge of, or interest in how an airplane works, and who can be transported halfway around the globe in less than a day by simply sitting in a seat - it might as well be magic to them.

I think we’re seeing the same thing in artificial intelligence. Machine learning has been around for more than fifty years since the invention of the perceptron. First it was only available to those who built their own machines or implemented machine learning algorithms from scratch. Then, the machines and algorithms became democratized and it became available to practitioners who could operate these algorithms by training them with their own data. But the transformer architecture truly blew things open, and the general public is able to utilize pre-trained models that perform general tasks without the need for any technical understanding. Like the airplane, people can now just in their seat and enjoy the magic.

What does this mean for the present, and future? Well for one thing, we may actually expect to see some similarities to the aviation industry. Large, high CapEx hardware providers may accrue the most value (that is, as Boeing is to Nvidia), and model providers as well as inference companies may fight for margins and resemble airlines over time. While the largest and most well-funded companies may own a lot of market share with their great volumes of data and model size (as OpenAI/Anthropic is to American/United), there will always be room for more specialized providers offering lower cost or more niche offerings. There will also likely be more logistics focused inference companies, solving complex and/or high-volume business problems - the DHL’s and FedEx’s of the AI world.

As for all of the other startups around the space - there may be room for some - but not all. Just as air travel is a tool for personal and business use - it’s just that, a tool. Individuals and businesses will adopt AI, and happily pay for it, but it will not take over every aspect of everything. And just like there isn’t an airline that only flys from Cincinnati to Bora Bora (or at least, I think there isn’t), there may not be a lot of room for AI companies that solve really bespoke problems - they’ll be run over by the larger, more generalized players.

I also think we can also expect to see greater regulation that prohibits any one company from becoming too powerful in this space, much like in aerospace and aviation. And for recreational pilots like myself - I think we can expect a bountiful future of fun, practitioner-focused AI offerings for the small but tight community of those who can’t help but to nerd out and get behind the controls themselves.

Order from Chaos: The Subtle Superpowers of Transformer Models

In late 2023, I still don’t think we’ve seen the peak of AI-mania. It will likely come at some point soon when more data reveals realized ROI from enterprise SaaS companies integrating the first wave of generative-AI experiences into their existing products. I’m not a techno-pessimist by any means, but I like to view technology through the lens of utility, and I just haven’t seen all that much yet from large-language models.

That being said, I hold the opinion that the real power of transformer models will lie behind the scenes, in ways that most consumers and even many product/engineering folks won’t immediately see - like chatbots and generative art. Rather, it will primarily be in the way machines utilize and manipulate data in new ways. I’ll go over a few changes I’m particularly excited about:

1) Unstructured to Structured Data

Over the last decade we’ve developed incredible cloud-based tools for distributed computation over relational data. It took a while for the promise of “big data” to actually start to become realized, and I think it’s still in its infancy. In fact, I’d go as far as to say that the most enterprise value to be captured over the next several decades is the “moneyballization” of nearly every company and industry. In other words, the “long-tail” of data-driven transformation is one very long, very tall tail.

However, you can only really start to do useful things with data once it’s cleaned, modeled in a sensible way, and loaded into a system that can process it. This is the first hidden superpower of transformer models. They can do these steps incredibly well without much human intervention, which will enable us to get to the real meat of that data - the “insights” steps - much faster. In many cases, it will allow us to process data that might have previously required teams of mechanical turks to sift through, and in some cases will allow us to process data that was economically never feasible to extract. Think about the petabytes of data trapped in invoices, pdf’s, medical records, government documents, aerial imagery, and so on that can be autonomously read, modeled, cleaned, loaded, and prepared for downstream processing by autonomous AI.

The more sci-fi manifestation of this comes in multi-modal models capable of processing audio, video, depth, and more - and these aren’t far off either. Almost all of the “observable” world around us can be modeled in data, and soon enough we’ll be autonomously capturing and modeling nearly everything, preserving it in ultra-cheap storage with perfect recall, and mining it for insights in the background.

2) Fuzzy API Calling

Now that language models can be forced to produce deterministically-formatted outputs (e.g. JSON or otherwise), you can call an API in countless ways. This means a GET/POST request might carry the same weight as saying something aloud, taking a picture, or just about any other type of prompt - direct or indirect. Some of these were possible before, but developers will have a dramatically easier time building applications that don’t enforce strict programmatic calling of methods.

Patrick Collison has a great quote, saying that “end-used software increasingly operates behind bulletproof glass … [and] end-user computing is becoming less of a bicycle and more of a monorail for the mind.” But I suspect the consequence of this new paradigm shift is an increase in generalization of software capabilities. This may or may not be a good thing for startups, who were previously able to niche down to find a less-competitive market to get a foothold in. When the companies with the biggest models offer true Swiss-army knives to users - where will opportunities still exist to create new software companies?

3) Reconciling Fragmented Data

I’ve recently been working with Shaper Capital, started by Travis May. He was a co-founder of LiveRamp and the founder of Datavant, and has a belief that there are data fragmentation companies to be built in nearly every industry. If you can provide the common data language for everyone to speak - you can become a big, sticky, and important business.

But if you combine ideas 1) and 2) together, you might notice that it’s easier than ever to allow people to “speak the same language” regardless of their direct cooperation. Consider an LLM living on top of a centralized database, with access to low-level CRUD (create, read, update, delete) operations, maybe some higher-level API routes with specific business logic, and general freedom to mix and match such methods as needed. It can receive all kinds of multi-modal inputs, including natural language queries, miscellaneous types of documents, images, and programmatic input. In this way, the underlying database can become a living, breathing transactional system of record (and system of action if desired). It no longer relies on a specific set or type of inputs in order to make modifications, allowing it to receive updates from participants who may not even know they’re modifying it.

A canonical example of this might be a new version of Wikipedia, where users can upload miscellaneous historical data, and have conversations with the system about topics they know about. The system can intelligently insert, update, parse, purge, and re-weight its records as it receives new input, and constantly change the information it displays on each page based on it’s latest understanding of ground truths. There are more vertical specific, industry focused examples of this - and I'd encourage you to reach out to me or Travis if you have any strong convictions here.

These are just a few of the net new, extremely powerful tools that I foresee emerging in the second wave of generative-AI. I suspect we’ll see even more interesting, powerful tooling developed in the third wave and beyond. Stay tuned!

If Aliens

There’s been a lot of interest in UFO’s (UAP’s) in the last several years. A recent whistleblower claimed the U.S. has multiple advanced “technical vehicles” in its possession from extraterrestrial species. He also expressed that we may possess the bodies of their pilots. The people are outraged - “we’ve been lied to! We deserve to know the truth!” But do we?

I’ve had an interest in UFO’s and aliens since I was young. As an engineer and pilot I could only dream of a world where we have access to vehicles and power sources orders of magnitude beyond out current capabilities. My personal conclusion is that there is likely to be an advanced species living on and off Earth whose origins predate human evolution. Given the surface area of unexplored ocean floors, our earliest records going back only some five thousand years, and our limited understanding of how life even formed in the first place, it seems more likely than not that a more intelligent species came before us.

If that’s true, it’s fair to assume that some subset of the 8 billion people on Earth have privileged knowledge of the situation. It’s also likely that the countries with the most advanced weapons systems, black ops programs, and area of land mass have had the opportunity to down or recover some of the vehicles of this species in the last century.

If these assumptions are true and the U.S. government does in fact possess a handful of these vehicles, do the people deserve to know this and furthermore have access to the knowledge we’ve acquired from them?

As much as I’d like to say yes - it’s a definite no from me for several reasons:

1. National security - if the U.S. even affirms it has recovered engineered materials from a more advanced species, every hostile nation without access to similar materials will make it their number one priority to recover some for themselves. This could mean immediate land war or cold war the likes of which we’ve never seen. There is also the risk of domestic lunatics committing terroristic acts in the name of freedom.

2. Technological discontinuity - in the history of mankind, we’ve never been handed a technology from a more advanced race. The gap in the capabilities of these technologies is not only likely to be many orders of magnitude beyond anything we’ve ever created, they likely doesn’t obey our fundamental understandings of the natural world.

As much as we like to believe we’re a highly innovative species, every ounce of technological progress happens incrementally. Every technology we’ve created had a worse predecessor that helps society adopt and cope with the next thing. If anything I’d say the biggest step change we’ve ever seen was the Manhattan project.

The vehicles that fighter pilots have observed are capable of accelerating from zero velocity to a point 40 miles away in less than a minute, meaning the power they draw alone is likely far greater than anything we’ve come close to engineering - if they even use power in a way we conventionally understand. I’d imagine that even for our greatest physicists the task of understanding these systems like fitting a linear, 2-dimensional model to a high-dimensional, highly non-linear curve. In other words, we may be able to understand a little bit of what’s going on and make use of it, but it’s very unlikely that we can understand the whole picture.

3. Economic discontinuity - let’s assume for a moment that we have successfully reverse engineered the propulsion and power systems of one of these vehicles. It’s likely that this technology renders all current forms of air, land, and sea transportation as well as power generation relatively useless. This immediately devastates enormous sectors of the economy. But can’t we just give all of the private companies the technical know-how to produce these systems and bolster our economy instead? Which leads to…

4. Greed - if anything, the high-tech sector of our economy only continues to display new levels of greed, irresponsibility, and corruption. As unfortunate as it is, only a small subset of capital allocators have the sophistication and moral integrity to deploy capital into responsible innovation. I can only imagine the fierce land grab to bring alien technology to market regardless of the impacts on human life.

So what’s the alternative? I would suggest a similar strategy to that of the Imitation Game. Once Alan Turing’s team cracked Enigma, they had the choice to open the floodgates and stop all of the imminent German attacks. However, this would have alerted the German’s to their knowledge of having solved Enigma, rendering their efforts useless. Instead, they used statistics to strategically choose points in the course of the war where they could have significant influence and over time allow the Allies to win the war.

As painful as it is to say, I likewise think we should leave UFO/UAP vehicle research and knowledge transfer to the federal government. Over time the learnings should be quietly transferred to research institutions and private companies to make use of. It may sound a bit corrupt, but the government should be responsible for allocating these understandings in a way that is fair to competitive dynamics and to institutions who can render them useful for society with protected downside risk. I wouldn’t be the least bit surprised if this has been happening for decades already, and frankly think this would make for a very cool movie plot.


The other day I ran into an acquaintance at a coffee shop. He’s a composer, deeply embedded in the community of musicians in the city. I originally met him when I subleased his room nearly two years ago upon arrival in San Francisco.

We caught up for a bit, and near the end of the conversation he invited me to a “piano beach bonfire.” It sounded odd, but not unlike other hijinks my musician friends might organize. I half-heartedly agreed to stop by if I could find the time. I texted another pianist friend and asked if he wanted to go with me, figuring it would be a good excuse for another overdue hangout.

When we arrived at the event (which had by now moved to Golden Gate Park), what we found was awesome. About twenty or so friends, acquaintances, and strangers were all gathered around a grand piano - taking turns jumping on, playing duets, adding vocals, guitar, tambourine, and whatever else the moment called for. It was a really authentic experience, and people were enjoying themselves to an extent that could only be felt in such a strange, spontaneous event.

Now you may ask, how was someone transporting a grand piano around San Francisco for such random gatherings? As it turns out - the owner had imported an electric piano mover (which resembled a small tank) from France, and was renting a moving truck to drive it around. I asked her how much it cost, which she politely declined to answer but countered with a recommendation to donate.

But what stuck with me more than the how was the why. Why had this woman decided to go so far out of her way to make this happen? There appeared to be no business model, intent to commercialize, or even expectation to recuperate the money or time she was spending on this. There was only one reasonable conclusion I could draw: this was important to her. The magnitude of that importance was reflected in her effort, and that effort resulted in a collective enjoyment that was almost incalculable.

In general, life is composed of things we have to do, should do, and want to do. Things we have to do include eating, brushing our teeth, taking out the trash, and emptying the dishwasher. I call a lot of this overhead “adult admin stuff.” Then there are the things we should do - which include going to school, getting a job, and saving money. “Responsibility” is the word that comes to mind for this category. The things we want to do fill up the rest: hobbies, passions, travel, romance, and the like.

For many of us, we let that which we have to do, and that which we should do fill up nearly the entirety of our lives. We push the things we want to do to the side until the “time is right.” But as we all know, our time isn’t infinite, and that right time may never show up. Our short time on Earth is even further shortened by other periods of unavoidable obligations. And as I’m coming to realize, we’re all collectively limited by the things that each one us wanted to do, but decided wasn’t important enough in the moment.

I challenge you to ask yourself: “what’s my piano beach bonfire?” What’s the thing you love - that if all obstacles were taken away - you’d choose to pursue? This can be as small or large as you’d like: the only important thing is that it’s important to you. Our society is shaped by those who felt something was important enough to chase with vigor, even when it seemed crazy to do so. But after all, isn’t it crazy to live your whole life ignoring what feels important?

The Code Less Stupid

I’ve been reading a book called The Road Less Stupid. The thesis is that one’s goal in decision making should not be to focus on making more smart choices, but rather fewer stupid ones.

This has been especially true in my experience as a software engineer. I was mostly self-taught, so I always assumed that what I was doing was dumb, all the time. When I became a professional software engineer I realized this was entirely correct. As it turns out, most code is written stupidly (so I had nothing to feel bad about). It’s just a matter of how stupid.

But what is stupid code? In my opinion, stupid code necessitates re-engineering. That is, if you build something in a way that requires you to backpedal, change, or delete a lot of what has been written when you need to make a modification, add a feature, or scale the system - it’s stupidly written.

How do we minimize stupid? We think! It’s hard for non-coders to understand, but someone can be a very productive software engineer when they’re not in front of a computer. Often times when I reach a crossroad in my work that can be solved in a number of different ways, I step away from the problem for a bit. I go for a walk, whiteboard something out, or sleep on it. Often times, the right solution comes to me when I give myself time to think about it. And what is right in this case? It’s the solution that maximizes extensibility and minimizes backtracking and refactoring in the long-term.

But sometimes it’s stupid to not be stupid. Facebook’s famous motto “move fast and break things” comes to mind here. Essentially, software engineering often comes down to a resource planning problem where the primary constraint is engineering hours. In some cases, the objective is to move as fast as possible to beat competition or release something before a deadline - which will earn you future capital for engineering hours that can fix your stupid code. I believe most startups operate this way, so the good news is you can have a long, steady career of fixing stupid code if you join later-stage companies that don’t die from stupidity.

There’s a term for this kind of deferred payment on engineering stupidity; it’s called technical debt. Much like real debt, if you’re not carrying some amount of it, you may not be using it appropriately. But if you carry far too much, it will weigh you down for the rest of your life. I think this explains much of the bloat and tendency for modern software companies to become marginally far less efficient with scale.

My perspective is that companies should be really careful with tech debt, and seriously question when speed should be the priority (naturally it’s usually not the engineers who think it should be). If you can take a little bit more time to build things in a less stupid way, you won’t be spending the later stages of your life paying down the tech debt and fighting fires; you’ll be innovating and spending your time adding more value for your customers than your hastier competitors can afford. You’ll also have a leaner payroll since your marginal efficiency will be far higher; so you’ll likely also be operating a much more capital efficient business.

It probably goes without saying, but your company’s people systems work similarly to your codebases. Spend your time hastily hiring the wrong people or treating them poorly in your quest for success - and you’ll be paying down that debt for a long, long time.

So I’ll introduce my own, hopefully one day winning motto: “move fast and minimize stupid.”

The Best Build Tools

My grandfather was an expert wood carver, metalworker, machinist, pilot, and angler - as well as a teacher and school principal. When I began to tinker with engines and electronics as a kid, he would stress the importance of using the right tools. He had a saying along the lines of “you’re only as good as the tools you use.” As any arrogant kid would, I largely blew this off and didn’t think much of it at the time. At least, I certainly didn’t think about it at the individual, organizational, and societal scale that I do now.

Now that I work in the software world where magic can be made without leaving your chair, it’s stark to see how much of a difference the right tools can make. You see the impact of proper tooling in ideation, scope of capabilities, speed, cost, team/company size, competition, and morale. It’s incredible to witness the leverage a great engineer has with the right tools, and how much less capable they are without them. My genuine hope for the next several decades is that the development of tooling in the physical world begins to accelerate as fast as it has in the world of software and computing.

The very best engineers are so good that they recognize clear gaps that exist between where their capabilities are and where they could be, and then go build the right tools to compensate for this gap. Likewise, the best companies recognize common problems in their organization and build internal teams and tools to solve these problems. Sometimes these companies are so good - and the tools they build are so generally useful - that they overflow out of their parent companies and became useful to others. A few examples include Airflow which originated from Airbnb, React and GraphQL which came out of Facebook, and AWS which spun out when Amazon started building impeccable developer tooling in house.

Likewise, the very worst companies skimp on time and budget dedicated to building and procuring tools. Because technological progress is compounding and hence exponential, the opportunity cost of operating without the best tools can be far larger than point-in-time cost saving considerations. They say that companies die by suicide - not homicide - but I believe that repeatedly cheaping out on running shoes is not an ideal way to win a race. But inventing a bicycle while you’re running seems to be a reasonable approach to never having to worry about competition.

Let’s take a step back - what is a tool? I’ll skip the dictionary definition and offer my own: A tool is any mechanism that can provide new capabilities or leverage to a user without the tool creator’s intervention. In other words, the tool’s net utility to society is not bounded by the time the creator spends working on it. Perhaps the original creator of the wrench spent a day designing and smithing the first prototype. Since then, wrenches have independently been used trillions or quadrillions of times. On the other hand, the utility of a mechanic who uses a wrench is bounded by the amount of time they spend using it.

That being said, I believe that almost anyone can build useful tools. A tool can be as simple as an instruction manual or how-to guide, or as complex as an orbital rocket or new mathematical theorem.

I’d also argue from a moral standpoint that everyone should try to build tools at some point in life. Building a tool answers the question of “how can I be useful to others even when I’m not present?” In some ways, I believe the pinnacle of society might be when we’re all just builders and users of tools, enabling greater leverage amongst ourselves, endowing ourselves with richer capabilities, enhancing our constructs for learning, and outsourcing non-creative work to the tools we’ve built. I’d also argue that from a capitalistic standpoint, everyone should try to build and sell tools so that, as Warren Buffet says, “you can make money while you sleep.”

Regardless of what you do or where you are in life, I think it’s worth taking a pause and thinking deeply about what tools you currently use, and what tools be could improved upon or invented that would enable you to do what you do even better. Ask your friends and others in the field the same question. Look for such differences in tools used across schools you’ve attended, companies you’re worked for, or cities and countries you’ve visited. Finally, ask yourself what you know a lot about that others might not. Is there a way to share that knowledge or skill with others even when you’re not around?

The next time you ask yourself how you can become better - remember that among the best, the best build tools.

The (Non-)Utility of College

On a semi-regular basis, I interact with people who have dropped out of school, are on leave, or are considering dropping out. This is partly due to the industry I'm in (tech), the city I live in (San Francisco), and partly because of where I work (Contrary).

Contrary is a newer venture capital firm with a truly people-first thesis. We identify great up-and-comers in tech, and then invest in and support them relentlessly through their careers - often well before they even decide to start a company. We have a network of 500+ really sharp young people in our community. Many are in or have attended elite universities, and many dropped out of school. Within the firm itself, we have some college dropouts working for us full time, several advanced degree holders, and one PhD.

Now that I'm almost two years out of school, I'm starting to take a wider view of my college experience, and reflect on how it was useful so far in my career. My hope is that anyone who is teetering on dropping out can read this and absorb it into the constant stream of information that is often encouraging them to leave school.

I came close dropping out of college - not once, but twice. Not only did I finish, but I stayed to earn a master's degree as well. I believe these decisions were ultimately the right ones for me, but maybe not for others. I think there are a number of reasons why it may be worthwhile to leave school, but this post isn't meant to convince you one way or another. I want the reader, above all, to use this information in the context of their own situation - and as always - to think for themselves.

With that in mind: Why did I find college useful?

I first want to get the point across that college became most useful when I stopped expecting it to be. I came to college with the expectation that all of the content would have practical utility, as many do. This was a large source of my frustration while in college and only ceased when I was willing to stop concerning myself with this aspect of it. Once this happened, it paradoxically became much more "useful". I'll explain how:

I learned to truly love learning.

I started to enjoy solving problems simply because they were hard or interesting, collaborating with or learning from others because they creative or brilliant, and going deep or wide on a new subject simply because it was new.

College is a unique playground for the mind. If you're curious by nature, it'll be one of the best times for you to go and study something interesting for the sake of it. A well-structured course is like climbing a mountain of knowledge, and a good professor is like a sherpa. It's well worth it to climb such mountains with a guide who knows the route.

I learned how to "wrestle problems to the ground."

I was handed lots of super hard and ambiguous problems during my studies. Collaboration was usually encouraged and support was regularly available. However, it was ultimately on you as an individual to find the right mechanisms to solve problems. Cheating was rare because of the honor code at my university, or impossible because there were no resources containing solutions to the problems. So you had no choice but to find a way to solve them. I remember working on math proofs that would individually take upwards of twenty hours to solve. This was an absolutely invaluable process to learn/skill to obtain.

The harsh truth is that most problems you'll encounter in professional life are actually not that hard. I'm a software engineer, and most code I've come across requires far less brainpower to write compared to solving a modestly hard math or physics problem. If you want to increase your overall problem-solving grit - or just like to be challenged - I recommend studying hard subjects and being held accountable for achieving a baseline proficiency in them.

I attained new mental models that changed how I look at the world and process information.

As you might know if you've taken computer science classes, there are algorithms that are far more efficient than others at sorting and processing data. So if you can hardwire new algorithms into your brain that improve your ability to process information and make decisions, you effectively become smarter. It wasn't the actual algorithms I learned in classes which enhanced my perception of the world, but rather the constructs required to understand new subjects.

For example, graph theory taught me to look at the world as a series of connected entities, forming and clustering in predictable ways. My differential equations classes taught me to see functional relationships in my day-to-day, and how varying inputs or rates-of-change of inputs can effect outputs. Economics and game theory classes taught me view the human nature response to incentive systems. Artificial intelligence and machine learning classes taught me about models of learning themselves, among humans and computers alike. Art and architecture classes taught me to see how you can design spaces to encourage or discourage behavior. And so on with dozens of more classes.

"Standing on the shoulders of giants" - and seeing the world through their constructs - allows you to see more, understand things in new and unique ways, and begin to distill or create your own mental models for interacting with the world and making decisions. Ultimately, life is just a series of connected decisions, so it's worth doing what you can to learn to make better ones.

The Dunning-Kruger effect is real.

If you're not familiar with it, the Dunning-Kruger effect is a rough model of the relationship between confidence and competence over time for most people. Early on - as you wade into a new subject - your confidence will be super high and then quickly drop off into the "Valley of Despair" as you realize how little you know. If you're undeterred, you'll slowly climb out of the valley, and eventually back up to some point where you feel you understand the subject, but know where your limitations lie. This is an incredibly important process to go through over and over again - and worth it each time.

College is a uniquely great place to experience this in short and medium-term bursts, without consequence. I failed my first graph theory exam miserably. However, I knew that I had a passion for the subject, so I fought through passing the class. I then went on the next semester to do research in the subject with the postdoc who taught the class and made it a cornerstone of my senior thesis. But had I been unwilling to enter the valley this outcome would not have been possible.

Every single semester I made sure to take at least one class that I felt was stretching my capabilities in some form or another. In doing so, I dove headfirst into the valley, knowing that discomfort was actually a friend. I'd trudge through the course, and climb out of the valley once again. I rarely cared what my grade was in such courses - it simply wasn't the point.

While watching Hulu's show about Elizabeth Holmes (the disgraced founder of Theranos), my girlfriend astutely pointed out that Elizabeth Holmes simply never trudged into the valley of despair. She pretended to be a genius, lied when she couldn't produce a working product, and pretended to sail right over the valley to greatness. Of course, she'll go down as one of the biggest frauds of all time, simply because she never was willing to have the humility to accept that she had more to learn.

Being a finisher is critical.

Learning to complete something is an absolutely critical skill to learn and practice routinely. It should be obvious that this skill is mostly for yourself.

But externally there are even more reasons. I remember when I was considering leaving school I read a quote from David Rose, a venture capitalist, that effectively said he was unwilling to invest in dropouts, simply because he had no data to indicate they could commit themselves to completing something of meaningful duration. Fighting your way through hard courses and completing degrees signals to others that if they hire or invest their time and money into working with you that you'll deliver.

College is a great place to "make yourself valuable to yourself."

You may be noticing a trend by now that most of these points have little to do with landing jobs or gaining credentials. There is more and more evidence pointing to the fact that you don't need a college degree to do reasonably well in your career, especially in tech.

I took the approach in school - and continue to in my career - to continue to make myself increasingly valuable to myself, not others. My goal has never been to become a tool for someone else. I want to set myself up to achieve my goals, pursue my passions, and live a fulfilling life. However, doing great work in challenging environments where you aren't in full control can make you far more valuable to yourself than the alternative. College is one of those environments, and only by the end did I realize I had also accidentally made myself useful to others in the process.

College opens doors.

Whether you like it or not, graduating from a great university still means something to most people, and they'll hold doors wide open for those who did. Sometimes, it seems almost to a strange or unfair extent. Without the degree, some of these doors are not held as wide open (or are shut entirely).

You might be excluded from things that otherwise seem trivial but require a college degree. I don't want to spell out too many anecdotes here, but I've already seen a surprsing number of these situations.

The network is useful (friends are good too).

Frankly, I think this aspect is a little over-emphasized, and becoming less important as there are more communities you can become part of during and after school. However, you can still certainly build a strong network of people in college that might help you along the way. In fact, I became involved in Contrary through their fellowship (one such alternative community), but I was referred by a friend from college.

To wrap this up, it may surprise you when I say that I never really liked school, and still don't. I didn't get great grades growing up, was fairly rebellious, and found most of K-12 excruciatingly boring. But I always loved learning about things I was genuinely interested in and wanted to find powerful ways to advance myself. I found a way to have college satiate both of these desires.

It was ultimately an imperfect place for someone like me, and I often felt like a square peg fitting into a round hole. If you've read thus far, there's a good chance you feel this way too. But if you relax your prior assumptions about what college should be and how it's failing to live up to your expectations, perhaps you'll start to discover some unexpected benefits of your own.

Doors that Open Doors - What is (and isn't) a Technology Company?

In case you haven't noticed, there's been an explosion in "tech" entrepreneurship recently. This is probably a good thing, but also requires some more careful thought on behalf of founders, VC's, and public market/retail investors to avoid another dotcom bubble scenario. I put "tech" in quotes here because this is the exact word I think we should be more careful about using liberally.

Being technical myself, it's more or less straightforward to make the distinction between a technology company and a tech-enabled company. It may not be so clear to others, so I'll make it clearer here:

- A technology company builds technology (i.e. digital and/or physical tools - something that you can use to build something else), sells it to other businesses (sometimes directly to consumers), and/or uses the technology directly in-house to accomplish its goals.
- A tech-enabled company uses technology (often purchased from a series of technology companies) to sell products or services that would not be considered technology itself.

Some examples of technology companies include (in no particular order):

- Snowflake
- Stripe
- Anduril
- Figma
- SpaceX
- Joby Aviation
- Boston Dynamics
- Tesla
- Waymo
- Two Sigma
- Apple

Snowflake builds and sells its advanced data-warehouse technology to other companies, whereas Joby plans to use its advanced eVTOL aircraft to build an air taxi service, but will not sell the aircraft to other businesses.

I'm actually not going to list out any specific tech-enabled companies, because I don't want to point to a finger at anyone and call them "less valuable" (but they realistically are, and I'll explain why). Categories include but are not limited to:

- E-commerce
- Digital marketplaces
- Social Media Apps
- Most B2B SaaS
- Consulting or Agencies

These types of businesses often have many more sales, marketing, and other non-technical employees than engineers or developers (which might be a good rule of thumb to use when drawing the distinction between the two).

Technology companies do much harder technical things than tech-enabled companies. So hard, in fact, that each are likely more constrained by available talent in the world to work on such problems than by capitalization concerns. If you want to create a competitor to some of these businesses, you may need to poach talent from them directly to even get started. Not to mention, you should expect to spend years working on the problem to even create an MVP. Go ahead - try taking a weekend and building a cloud hosting provider, orbital rocket, or payments processor.

Businesses that are tech-enabled do not suffer such ills. You can not only find a highly commoditized pool of talent to work on your application, but these days you can probably outsource the whole job to contractors. You can also likely build an MVP for many tech-enabled businesses in a weekend.

Simple economics says that when there is abundant competition, businesses will continue to fight a "race to the bottom" to drop the price of their product or service below that of their competitors. In turn, margins will continue to get eaten away, and profits will stagnate or drop over time. In the case of tech-enabled businesses, there are way more competitors - and new ones can show up in a matter of months. Not to mention - there may be free, open-source alternatives. As technology companies democratize access to creating tech-enabled companies - and knowledge of how to use their tools becomes commonplace - this trend will only continue.

A good way to think about this is the analogy of each company holding a key to a door. In earlier-stage companies, you can't really see what's behind the door, if the founder(s) have chosen the right door, or have the right key to it. As the door cracks open, you can get a better sense of what treasures lie inside, and you can keep funding it until the door is pushed all the way open. However, the difference between a technology company and a tech-enabled company is that inside the room of a technology company there are more doors. These can be opened by other founders or by the business itself. For a tech-enabled business, there may be a bunch of treasure but usually no more doors.

It's obviously a better scenario when there is a cascading set of doors behind the first one, because a whole bunch of treasure will have to pass through the first room to be obtained. In most cases, you know in advance which doors could have more doors behind them, and which have only treasure behind them (their are exceptions, of course).

Therefore, I would expect that over time public/private market valuations and revenue multiples will reflect the results of this economic principle. That being said, creating a successful technology company typically takes much longer, will often face steep adoption and regulatory hurdles, and underlying the technology being pursued can turn out to be flat out wrong in principle/made irrelevant over time.

So if you're a founder jumping on the bandwagon of "now is a great time to start a tech company" - think carefully about which category you fall into and how that squares with your intended time/wealth outcome for the endeavor. If you're an investor - this might be an interesting thing to consider when evaluating your portfolio or deal flow.


- This is not necessarily a new argument, but a re-hashing and reminder of ones we've seen before (see Peter Thiel's "Technology vs. Globalization" and "Competition is For Losers" arguments in Zero to One).
- Technology is not the only way to produce anti-competitive effects for a business. Take Airbnb for example, which builts its moat by cultivating a strong double-sided marketplace and community.
- Some tech-enabled businesses can build technology in-house that eventually make them technology businesses. Take Amazon for example. It started as an online bookstore, but later built state of the art optimization and logistics technology that other retailers can't compete with. Similar stories with Uber, Facebook, and many more.

The Metaverse Will Be Paved With New Communication, Not Payments Layers

I'm about to share some pretty unpopular opinions among the crypto/NFT/web3 crowd, so if you're deep in that game and are too uncomfortable with new ideas, here's your exit.

For those who are interested in sticking around, I'd like to start with a brief history of another technology. In 2011, three years after Satoshi Nakamoto introduced Bitcoin and blockchain technology, the first WebRTC spec was introduced. For those unfamiliar with WebRTC, it's a free, open-source set of protocols that enables peer-to-peer, low-latency real-time communication over the web. Using it, you can link two users via web browser and allow them to share bi-directional streams of audio, video, and other arbitrary data in real time.

WebRTC (and related technology) has enabled a slew of new applications that have subtly become a part of our everyday lives. I'd go as far as to argue that without WebRTC, the global economy would have not been able to weather the Covid-19 pandemic with the (relative) success it did with the transition to remote work/learning.

In the last year and a half, I started building in this space. Most recently I launched Mantis. Before that, I built From it, I learned this technology is non-trivial. For certain applications which involve multiparty communication, it requires very sophisticated engineering. There are lifetimes of work to be done here. I'll take the opportunity to shout out the open-source Pion project, working to democratize and extend WebRTC technology, on which much of Mantis and are built.

I believe that the next version of the internet (or at least part of it) will be the "real-time internet." That is, since static websites in the 90's, and more complex web apps through the 00's and 10's, bandwidth has only kept increasing and our lives have become only more coupled to the web. One would only expect that the packets will continue to fly faster and in greater volume. Pair this with the frightening improvements in video game graphics plus powerful cloud computing, and the ultimate end of this probably looks something like the "Ready Player One" fantasy metaverse that has launched into the spotlight recently.

This is where my path intersects with the crypto folks. I'm still not exactly sure what these people mean by web3, but I think that's okay because I don't think they really know what they mean by it either. A common answer to someone asking for a definition is "it's still too early to tell."

But from what I can tell, web3 is about:
- users of web apps "owning" the data they produce on the apps
- payments (in the form of cryptocurrency) and asset ownership (in the form of NFT's) being central to the apps
- virtual spaces for live work, play, and study where such data will be produced and payments/asset transfers will be conducted

I'm not here to argue against the value or truth of these individual points (although I'm tempted), but really about the backwards mindset of which point is clearly the most important with respect to the metaverse.

Discounting the strong possibility web3 will be a short-lived hype, the truth is that the internet of tomorrow will look very different than the internet of today. But if you look at the history of new means of communication, they came to fruition before people started thinking, or even caring about using them for transactional purposes. I actually think there is a great SNL skit in the making where Mark Zuckerberg steps into the metaverse only to find a giant NFT flea-market where an AI avatar is hounding him to buy an original jpeg of an ape for a trillion dogecoin. (He then naturally takes off his headset, throws it in the trash, and deletes all the code ever written at Facebook/Meta.)

This payments-second history will likely be as true of the metaverse as it was the internet, television, phones, mail, and most real-life places of gathering. If they were designed first and foremost as means of buying and selling, I'd suspect they'd have never seen mass adoption; they'd be absolutely repulsive.

On the other hand, if I can step into a virtual world that enables more productive and enjoyable work, learning, and play - that's a beautiful future. I not only think it can be, but has the potential to be perhaps the greatest, most fulfilling medium of all time for builders, engineers, scientists, artists, and more due to its unconstrained nature. And I most certainly believe people will be able to make immense amounts of money within it by providing real value to its users - and that's a great thing.

So if you're trying to decide what the future looks like - and consequently how to spend your time and money - I urge you to look at the spaces where people are building new things that enable other new things. I think by now it may be clear that I'm less than bullish on blockchain-based technologies, and that's because I've seen close to zero successful applications enabled by it that weren't cryptocurrencies, a derivative or cryptocurrencies, or means of exchanging cryptocurrencies.

I’m actually hopeful to be proven wrong in this regard, and somehow see the emergence of a bunch of decentralized applications that lead to better economic incentives and outcomes. But as a comparison, in a shorter amount of time WebRTC and related technologies has quietly enabled more change in your life than you probably care to know about, and may have been the only reason you were able to afford your NFT's (by keeping a paycheck, customers, or investors) in the first place. Yet for some reason few seem to care about this, and other more fruitful technologies relative to blockchain.

As always, I'd love to hear the adversarial views on my ideas. Tell me directly using Mantis, drop a comment, or send me a Tweet. And if you're reading this in 20 years, come find me in the metaverse and let me know how I did.

Automation != Leverage

There's a subtle but dangerous misconception about technology that seems to lead many business decisions astray. Worse, it paints a bleak outlook of the future. That misconception is that somehow automation always produces additional leverage in a business context. This is false.

First, a couple quick primers: A lever is a simple machine which, when you apply force, outputs a magnified force on the other end. You can think of the magnitude of leverage as how many units of output you get for every unit of input in a given system. For example, a walking human burns about 50 units of energy (kCal) to travel a kilometer. Give that human a bike, and they can travel roughly 2 kilometers with the same energy. Juice up a Lime scooter with that same energy, and the human can travel roughly 5.5 kilometers. That's technological leverage.

Automation is even simpler; it's the process of taking a system and removing (or minimizing) the need for human involvement for the system to function. Much of software is about automating tasks. That newspaper that used to show up at your door is now automatically delivered to your inbox in the morning. And it probably costs 1/1000th of what it used to for the newspaper company.

In many if not most cases, automation does directly produce more leverage. It's probably the main driver of the massive creation of wealth in the software and internet industry. Because of this, when presented with an opportunity to automate something, software companies jump on it. This knee-jerk reaction can come at a serious cost.

Consider an email campaign. Your business has 1,000 current users, and you want to inform them of a new product offering to generate more ARR. Easy! Create a series of campaign emails, and set your preferred email tool to blast them out one at a time to all 1,000 over the course of several weeks. You obtain a 5% conversion rate, translating to an additional $30000 in ARR for the year for your $50/month offering. Couldn't be easier - great success.

Now consider this alternative approach. Instead of blasting the campaign emails to all 1,000 users, you choose the top 100 you think are likely to buy your new offering. You craft personalized emails to them about why you think this offering can benefit them and offer them a calendar time to learn more. 80 of them convert. You net $48,000 in new ARR for the year. Besides the $18,000 additional ARR, you may have had some very useful conversations informing future product ideas.

Oops. Your automation actually produced less leverage than the alternative. Yes, your super cutting-edge software company incurred a large opportunity cost because of technology.

Now, critics will cry "you've toyed with the numbers!" and "this doesn't scale!" The most keen observer might say, "you actually may have produced more leverage in the first case depending on how long each of these took, and you simply got better results from the additional effort." Perhaps for this example they'd be right (or perhaps not).

Regardless, the basic idea is that companies are often faced with the decision of doing N units of shitty work or far fewer than N units of great work. It's not always a simple calculus to determine which will yield better results; sometimes the unintended consequences of doing great work are not always clear (and likewise of doing shitty work). And sometimes it requires doing great work manually before you should ever consider automating the task. As someone who loves automating just about every mundane task I do, this idea definitely got me thinking about when I'm doing this appropriately, and when I'm not.

E-commerce and SaaS companies love the idea of putting their offerings on autopilot, and allowing self-serve automations to take care of their customers from A-Z. Don't get me wrong, I love this idea too. But theory is different than practice, and you'll be hard pressed to find a company of this type that strictly benefits from removing humans from their sales/support process. A good company knows how to automate all of their processes. A great company knows when not to.

I've been thinking about this a lot as it pertains to Mantis. I talk to a lot of businesses who feel they are too big, too small, or too busy to have a human taking calls from customers and prospects. Sometimes when I hear this I feel bad for myself because I've built something they don't want. But usually I feel bad for the business mistaking their "AI" chat bot, FAQ page, or "we'll get back to you in 1 - 2 business days" autoresponder for "solving" one of their most critical problems: making their customers happy.

Have a different opinion? Feel free to tell me about it using Mantis in the bottom corner.

SaaS on a Hunch

A few weeks ago I launched Mantis, a live-audio chat plugin that any business can embed on their website. It allows site visitors to speak with them directly through the browser, with a single click, from anywhere in the world.

Unlike some SaaS offerings which are built around the discovery of some problem, Mantis is built on an opinion, hunch, or hypothesis about how things ought to be. Taking this approach is a bit like rolling a boulder to the top of a hill; I believe it actually requires me convincing (some) potential users to view things from a new point of view instead of just solving some pain. Let me elaborate.

There's a longer story here, but in short Mantis came from an end-user pain. I'm someone who:
a) likes to ask get answers to multiple questions before buying basically anything online, and
b) actually prefers calling businesses to get these answers when possible.

In my opinion, text-based live chats and email support are way too slow, and frankly demeaning when you're given a bot instead of a human. Even if I get a reply, the clarity and throughput typically is almost always worse than a call. But if I as a user/customer feel this pain, it clearly must translate to pain for a business, right?

Maybe, maybe not. Even for similar businesses, there's a ton of variation in approaches to user/prospect/customer interaction. Some businesses have told me they intentionally try to produce friction for a customer to speak to them; others have told me their availability for prospects (let alone paying customers) has made all the difference in their growth and success.

I was going to originally title this post "The Cost of Facelessness", or "No Business is Too Big to Be Faceless", but I'm not ready to quantify the earlier, or make a claim to the latter. These are the answers I'm looking for as I grow Mantis, and confirmation of such ideas will only aid its success. I'm also not driving towards binary answers, but rather seeking to understanding the conditions under which a business can greatly benefit from enhanced user interaction.

My hypothesis is that offering a frictionless, high-throughput means of allowing someone to contact an internet business (especially customers), has a positive ROI in almost every case. The magnitude of that ROI is highly variable.

For startups, speaking to users and prospects to drive product direction is downright necessary, so the ROI could wind up being the continued existence of the company (see this YC post if you don't know what I'm talking about).

For growth and later stage companies, letting your paying customers come to you about any issues seems like it should be a top priority, and efficiently answering questions for prospects should convert more customers, or generate more qualified leads. To me, these things seem obvious, but maybe I'm overlooking something.

Obviously such opinions will be met with adversarial points of view, and I welcome those. Any and all feedback is welcome, either via comments where I post this, or directly to me if you want to use Mantis in the bottom corner.

But if you really want to help me answer these questions, give Mantis a try, and let me know what you find. Right now I'm pulling people off the waitlist and provisioning access. I encourage you to drop your email there. If you want access even sooner, just shoot me a note at [email protected], and I'll try to serve you as fast as I can.