Vignettes from the Takeoff

Excerpts from a future history memoir

I remember the first one-person billion-dollar company. It wasn’t mine, I wasn’t working yet and was only an observer, and a distant one at that. But it felt exhilarating. A breath of fresh possibility, like any of us could do anything. A milestone in what humanity is capable of.

It lasted for a month.

The founder did very well for himself obviously, but within a matter of weeks someone else beat the record. One-person-unicorn became the 4 minute mile of company building, another rubicon crossed. Once the world knew it was possible it became inevitable. Because a world where one person can create a unicorn is also a world where another person can also create a unicorn. Maybe a day after, maybe a week, but pretty soon and it’s inevitable. And we saw the inevitable happen, four more in the next few weeks, till it became somewhat normal.

Entrepreneurship had already become a game in the 2010s as the saas boom made building big companies in short periods of time easily possible. The result was an incredible boom, many of them competitive with each other, with extreme dispersion in outcomes.

And now, when building became even easier, the equivalent to telling someone else to build things, it predictably got crazier. Not quite as easy as “make me a unicorn” but closer to it than what we’d had. Can you imagine if it was that easy? Everyone and their grandma would do it.

As the amount of effort we needed to put in to show the minimum of traction was reducing something had to shift to move us to the new equilibrium. If all people needed to do was be faster than others to ask a question, that’s a speed race to the bottom.

Once upon a time it was actually executing that was the bottleneck, soon it was project managing the thing you were executing. Then it was choosing the directions and making editorial choices about what thing you should create or run experiments on. By this layer of abstraction it was less about what could be made and more about what needed to be made. Because everyone could get AI to make almost anything it felt like but no one knew for sure what everyone else wanted.

My career started in earnest a bit after this. We all had eight monitors with running information streams from all over the company, and outside. I was called an analyst, because even though analyses had become cheap but accuracy hadn’t. Someone had to monitor the drones.

This was fine, actually. It’s not really what I thought I’d be doing but then it required me to think super fast and make a lot of decisions and keep on top of them, and try to automate some parts of those too. I liked it, sometimes it was even fun, though a lot of it was quite rote. I worked on the shipping industry side, accidentally if I’m being honest, that’s what was allocated to me, but turned out this was a pretty good window into the world. I had to keep on top of things from did the tanker break an engine part to like crude oil prices to atmospheric conditions in some strait.

Quite a lot of it was also dealing with competitors. I mean the normal stuff the AI could do, but the fun part was to confuse their AIs. Ships seemingly going the wrong way, or water displacement made to look fake, all sorts of tricks, some legal and some not. We all had the same machines but adversarial games are more fun, you know?

The rest of it, to look at the machines themselves and react quickly when necessary, that was okay. A hard job, much harder to pay continuous attention than to actively do things, at least for me. A lot of it was also reactive, and not just because of the adversarial problems. Like, even though the ability to analyse and communicate anything became instantaneous, it hadn’t necessarily helped in making the right decisions all that often yet. What it did mean is that if you were making a mistake, you got to make it faster now. There was no escape from Hayek. Every part of every company became more efficient in doing things even as knowing if you were being efficient in the right direction remained a mystery.

It felt like playing a videogame, highly stressful. You were always on call, always trying to figure out what broke and fix it, or find ways to game around what someone else might do. It was hard.

One thing that eventually helped me a year or two in was that corporate secrets stopped existing. Or at least they didn’t survive for long. Anything anyone did could be reverse engineered pretty quickly. As soon as things turned more adversarial this was probably inevitable. Who knows maybe it might have been just at the same rate as AI in the 2020s or software in the late 2010s or the entirety of Chinese manufacturing knowledge before that but it didn’t feel like it. Living through it felt like sailing the high seas, pirates and privateers at all sides.

Despite the respite brought by the new world, I ended up quitting that job another year after this. Just being alert for hours on end every day was hard, and no amount of laying around or drinking cured it. It had also meant that long undivided time to think and come up with ideas on your own was a dying art, and I had dreams of contributing to the world this way.

I understand why it was hard though. It’s hard to spend a decade coming up with a new idea for a car when you could just steal your competitor’s ideas that already worked. Why take a risk. The world became much less divergent. Sure, people did try to do things that were unique but like the Hollywood of yore everyone just copied from everyone else while occasionally a great piece of cinema broke out from nowhere.

I did feel though the size of individual companies were shrinking on average while the top exploded. When I was looking for jobs I kept seeing this. The one I ended up working for was tiny, maybe about 30 people. It was either this or just go independent contractor route. The Coasean bargain that made some companies larger broke apart, there ended up being a much larger number of individual contractors and smaller companies than were feasible before. Even I thought about joining their ranks, which would’ve been a bit more work.

Identifying and capturing those people is the most incredibly important piece of leverage. Some of the largest companies ended up being the conglomerates made up of these people who individually wanted to go and help them figure out answers to problems that they could not answer otherwise. There were other options too I suppose, like the original AI lab model which by now had disappeared, they had many fewer employees than those old companies of their size, but did run a large network of arrangements that would make the economically dependent population number in the hundreds of thousands.

As the AIs got better firms soon started calling for a new type of role, an “analyst”. They would get brought in to do a particular task once, whatever it took. I started out doing this for worldwide logistics networks. Deciding when AIs started going in a loop against the others it negotiated with for prices and goods or routing decisions. Which factories needed to be built, and which types of models to analyse for those. What pieces of data we were collecting were actually trustworthy, when the world had changed enough that our very model had to shift.

We all had something installed that could read and analyse everything that was done on the machine, to help us do the job better. But pretty soon, at the end of it, the AI just learnt from what I’d done, every part of it, and be able to just do it from then on.

Every job was the last job. What is done once got done for all time. The progress bar would go from 0 to 100 as you did it, and once done it remained done.

I remember getting paid for one of these jobs, about shipping logistics; it took a week and I made as much as I’d made in a year before. The value was high, and I was too stupid to think about “terminal value”.

These gigs themselves were also better as the AIs got better. It was much less stressful than frenetically monitoring it yourself like before. Mostly supervising other AIs, sometimes other people, sometimes other people supervising AIs. I hear of the days when people used to have the same job for 40 years and it sounds like a fairy tale because people today have jobs for 4 months. If they’re lucky with that they get to own a piece of the machines.

Some friends who were smarter started to ask, what even is a “job”? And I too worried, things of all my projects, would this disincentivise deep thinking? In the end it did, a bit, but the market corrected as time went on, because capital had to find ways to protect ideas, especially since many of them could be now reverse engineered. A lot more secrecy, for a short time could be monetised, because soon after you knew it would be known. I wasn’t at the cutting edge of anything enough that I could ask for a billion dollars and quiet time, but some were, and they prospered. Even the whiff of a good idea was enough.

This was the hardest part, because until this point all jobs people did throughout their lives relied on the jobs themselves being somewhat predictable day to day. Nobody except maybe some CEOs during a particularly tumultuous phase had to do completely different things hour after hour, day after day. What it meant to get paid for a few minutes of your time, a form of knowledge transfer, instead of getting paid for your actual labour, was enormously complicated, and societally destabilising.

Nobody quite figured it out but much of it ended up similar to contract work, where the work was timebound and sporadic and you got paid a premium for this gig work. These companies aren’t really companies, they mainly “collect” many of us to save us the trouble of searching. A thin wrapper between my agents and those that want my efforts.

The goal of doing all this, of your career, what constituted actual success, was to own capital. Most of my work has been in turning my personal labour into capital. And it was still good to own capital. It always is. You could deploy it and see people line up to take it and build things that would change the world in months. After all, building physical things remained a problem. Logistics remained a problem.

Anyway I don’t know if this is worth it to be honest anymore. What’s the point of having cash that you could give to an entrepreneur to build something, when others with capital could also do the same, and make damn sure that neither one of you would make much money without getting lucky? If the true skill of my labour is not differentiated enough, then what’s the point of just pouring more? Won’t everything just become highly competitive but undifferentiated, like in the commodity markets?

Those markets, despite the product being literal commodities and the process being the only differentiable part, mostly survived because different places have different regulatory structures and codified preferences. Which in turn determined who ends up being the marginal producer that can then be refined or transported or used. And so on and on.

The only choice was the robots, which were plentiful, everywhere. Robots gave leverage, a person could use it to help teach it how to do certain pieces of work and then supervise it thereafter. This held just as true for those who manufactured the robots as those who used it. The idiot index might have been a useful target to aim at after all. And with robots it’s no longer the case that you need hundreds of thousands of people in these industries. Energy and land remained bottlenecks, because you could always use more and they could always be cheaper, but the world didn’t oblige to the exception of everything else.

Don’t get me wrong, there was innovation to speed these up, but ultimately the decisions of what to invest in, what to create and what to make faster all turned out to be market problems as opposed to analysis problems. And market problems are wicked, and you cannot solve it just by running fast. It requires actually traversing the demandscape and banging your ideas against the real world, there are no shortcuts.

Even for those who had abundant capital, figuring out what portfolio of bets makes the most sense remained difficult because the response required information gathered from all over the world. And compared to how long a ship takes to build and sail, the decision on what type of ship to build didn’t take that long at all, even though it mattered the most. Those who claimed to have some insight in how to help folks figure this out lived quite well.

When I look around these days though, food is cheap, goods are cheap, learning is cheap, health is cheap, and if you want something more the amount of labour you need to provide those basics is miniscule. It all seems pretty nice. The biggest surprise from the heady days when the future was utopia might be that the pace of scientific discoveries changed, but not too much. I’m no scientist so I couldn’t tell you why this was the case, but it’s true. We did get better food and medicines, but string theory remains a theory. There are flying cars, but nobody’s riding rockets to the moon. I think maybe the discoveries just maybe weren’t bottlenecked by our inability to do analyses in the first place? We could run a ton more tests now but there are only so many problems we could brute force our way through. And once the low hanging fruit got picked over in the early 2030s, we sort of all got stuck again. Like how fundamental physics was in the late 20th century I’ve heard, stuck needing new ways of conceptualizing the world.

Attention is still capped because there are only so many humans. There are only so many hours in the day. One person’s gain is another’s loss. If you’re reading an essay it means you’re not reading another essay. Zero sum. The most drastic change was what happened when the only signalling that was costly was individual presence, since everything else could be faked.

For most of us who are at least somewhat young, in the last few years the world took a turn and became a lot more analog. Many of us don’t remember a life unmediated by the digital realm, but that was changing. When nothing you see or hear could be easily trusted then what remained were small enclaves functioning like private clubs. If you couldn’t be trusted you couldn’t enter. But even there, the rules had to become draconian because our daemons, our digital twins, our agents, could penetrate it if we had permission. Hence, physical presence.

This physical network also meant agglomeration, which is why I moved cities. Not for commerce, or work, but for my social life. I mean, it was either that or live a nomadic existence, traveling the world and seeing others wherever they are.

That’s mostly what I do now, while doing the occasional decision support job in areas I had learnt quite a bit about over the years. I have to keep spending some time every day making sure I keep up with the latest, but it’s fine. The jobs are sporadic, but it pays a good living even though you always feel like the other shoe’s just about to drop. The remaining time I have, which is most of it, I spend making entertainment for others in ways that are, for now, hard to imitate. There are physical plays people put on now that I go to sometimes, participate in sometimes. It feels good. This is life.