Guns, Germs, and Code

A Simple Thought Experiment

Imagine that a biological virus is going to spread through the world. Its spread will be rapid, virtually unstoppable, and essentially inevitable. But this is a weird virus, and it will have weird effects. Weird and wonderful even, in some cases, and weird and disastrous in others. But there is good news: there is a vaccine available already, with virtually no side effects. It won’t be a perfect cure, but it will dramatically reduce the worst effects of the virus. At least for now.

Should you take the vaccine?

Now, in what follows, I will show you that this is the exact equivalent of choosing to work with Claude Code or Codex today. Off we go:

Software Bonkers

Craig Mod says he is software bonkers.

I can empathize.

Anybody who has used Claude Code, or any one of the equivalent harnesses available today, will empathize. If you don’t know what a harness means in this context, read this by Ethan Mollick.

Craig describes how easy it was to create the exact software that he needed, and while he has a computer science degree, there are plenty of people who could tell you similar stories. As with all stories, so with these: they are models of the world, and your mileage with them will vary. But don’t depend on any one particular story, try to get a sense of where the world is going by reading all of them.

Anyone who can use a computer and is somewhat familiar with the English language can do these things and more, starting right now. You are reading this blog, and that means you meet both conditions: you can use a computer, and you are at least somewhat familiar with the English language.

Here is what this means for you: you can now just create the software that you need for your workflows. You do not need to go out and search for a software that fits your needs. You reduce your search costs.

You no longer need to adjust your workflow to suit available software. The software you create fits your needs, rather than the other way around. You reduce your transaction costs.

Here’s an analogy: imagine that it was time for you to buy a new car. Instead of seeing what was available in the market, you could just make a car suited to your specific needs. Large legroom at the front, not quite so much at the rear, a way to easily get your dog into the car, and the music that the teenager plays only pipes into her headphones as opposed to the entire car. We are not there (yet¹), but the equivalent of all this for the software you need? We are very much there, we were there as of four months ago or so.

Craig’s post is about his needs, and how he built his metaphorical car in this brave new world.

Which means, naturally, talking about the time the Old World met the New World. (Spoiler alert: it didn’t go well for the New World at the start).

Guns, Germs and Steel

About five hundred years ago, Europeans reached the Americas. About thirty years ago, a guy called Jared Diamond wrote a book about this. The title of that book was “Guns, Germs and Steel”. It is a very good book, and you should read it. The broad idea is that when those Europeans landed in the New World, they were able to bring with them three things: guns, germs and modern industrialization. There is much, much more to the book, of course, but this is a blog post, and so we’re going to stop with just this summary for now.

Now, the thing that matters to us in this blog post is that this was a shock to the way the New World did things. Those guys didn’t have military defenses against guns, they didn’t have biological defenses against these new germs, and they didn’t have any economic defenses against steel. Economists would have called this an exogenous shock, and it wasn’t just ‘a’ shock, but rather a triple whammy. And of those three whammies, the introduction of smallpox in particular proved to be devastating. Europeans, hardened by centuries of living in close proximity to livestock, had acquired immunity against many of the germs that they ended up being carriers for into the new world. But for everybody in the New World, it was catastrophic. And I do mean catastrophic: by one estimate, 95% (!!!) of the population was wiped out within a few generations after the initiation of the Columbian Exchange.

One of the things I want to do with this analogy is to help you understand that you can use this same frame to think about AI today. It (AI) has landed on our shores, and it brings along with it the modern day equivalent of guns, germs and steel. And we don’t have defenses against these things, in much the same way that the residents of the New World back then didn’t have defenses against guns, germs and steel. Except it is worse this time around, but we’ll get to that later.

For now, let us talk about Matt Mullenweg.

That One Time Matt Didn’t Take the Bait

Matt Mullenweg is the guy behind WordPress, and other things besides. Look him up, it will be a worthwhile exercise.

This is a guy who knows his way around the internet. Why? Well, mostly because his firm just happens to be a significant chunk of the internet.

Matt has a hair-raising post about a very sophisticated phishing attack, which he was able to spot well in time. But don’t underestimate what just happened. Hackers were almost able to pull off an online heist against the guy whose job is to make sure that 43% of the internet doesn’t suffer the same fate. You can almost sense the reluctant admiration in Matt’s writing at the sheer chutzpah of it all. And that admiration is tinged with horror, because Matt knows we’re just getting started. Things are about to get a whole lot worse.

Let us begin with a rejoinder to the obvious counter you might make: that there is nothing new with phishing attacks; they have been around for a long time. You wouldn’t be wrong; they have indeed always existed. But the same technology that makes it possible to build highly tailored software solutions affordably... also makes it possible to build bespoke phishing “solutions” at low cost. In fact, a personalized phishing scam is nothing but a specialized software solution!

The equivalent of an invasion of Old World germs is about to hit us, and it is going to happen very, very soon. How soon? Within the next six months, and I’m being very generous with my estimates.

Hold on to that thought, though.

The Two Axes

In his post, Craig talks about this being the “dorks-only” phase. That is, he is saying that only the hobbyists and the tinkerers have caught on to what is happening in the world today. And by implication, he sees a world to come, in which more and more people will join the tribe of the tinkerers, and everybody will be whizzing up ultra-customized software solutions. Or that’s the way I understand his point, at any rate.

And while I agree with him that many more people should be tinkering, I think it is worth emphasizing that there are two different forces at play here. The second thing that will happen is that Claude Code (and its friends from other labs) will just get better at doing all this by itself. At some point in the not-too-distant future, you can just have Claude Code see your world, see what’s missing on the software side and just… cook up those solutions that you’re currently conjuring up via vibe-coding.

To stick with the analogy I used earlier on in this post: you don’t have to design that tweaked-just-so-for-you-and-you-alone car. Claude will know that you will need a new car, and it will just offer to make it for you.

Or, in economist-y terms: the production function for ultra-customized software won’t just be augmented with labor in the months to come, but also by capital that is going to be much more capable than it is today.

Which brings us back to the thought I asked you to hold on to in the previous section. Those viruses that the Europeans brought over from the old world, they had one important limitation. Their capability to improve upon their attack over generations was limited. To use a phrase that the post-Covid world will understand all too well: they didn’t have much gain of function. But even if you’ve been under that rock they keep talking about, you will have heard of the rapid advancements in AI. AI is not, let me assure you, lacking in gain of function. And that rapidly improving capability will build both things: the just-so car, but also the just-so phishing scam.

And so the good news is that we can just get what we want in the software world.

The bad news? We will also get exactly what we do not want at all. And lots of it.

So is there a way out?

Mithridization, Baby

(Or mithridatism, if you want to be all pedantic).

What is mithridization? There was a king who was fairly certain that he was going to be poisoned, and so the guy just went and made himself immune to pretty much every poison he could think of. That’s the TL;DR, but do read the entire article.

My specific ask of you? You should get a sense of what is going on in this world. Not just because you should be familiar with what is possible today from a creator standpoint. But also because it matters from the standpoint of survival itself. The more you walk around in this jungle, the more familiar you will become with the terrain. And so when you are attacked (and note that I said, when, not if), survival becomes easier if you know where you are, and what you are doing.

Remember the thought experiment we started with? I’m saying you should take the vaccine!

OK fine, you might say. But how many shots, and how often? In other words, are you saying I should aspire to Matt Mullenweg levels of proficiency when it comes to doing things online?

Here’s how I would answer this question: think of immunity like a spectrum. Most of us today are the medical equivalent of folks who are taking serious immunosuppressant drugs. That is, we have little to no immunity against what is coming. But as Matt’s story makes clear, even folks with heightened levels of immunity aren’t exactly safe.

Or here’s another analogy: if you’re going to be traveling about in India, you should know how to spot a fake Bisleri bottle of mineral water. You should be able to do the online equivalent, at the very least. But even if you are way better than that, you should aspire to be a bit better in the future.

Want to get started?

Get the twenty dollar plan from either OpenAI or Anthropic. Download either the Codex app (OpenAI) or the Claude app (Anthropic). Once you log in by following the instructions on your screen, get started by saying: “Can you make a presentation based on this blogpost”? And paste the link to this blog post. This is a perfectly fine start, but amp up your ambitions as you go along. The more you tinker, the better your immunity gets.

This is worth repeating: you develop immunity by becoming familiar with the terrain, and through repeated exposure. After you have built out that presentation, see if you can make a website. It could be about anything: perhaps your child’s hobby if you are a parent? About generating diagrams if you are a college student, say, or maybe an interactive representation of the three body problem if you are a school student? But really, it is entirely up to you. The bottom line is that you can build software for your specific requirements, just like Craig did.

How far should you go once you get started? As I said, a tad farther than wherever you currently are would be my answer. And that applies to me too!

TMKK?

1. Get tinkering.

2. The rate of improvement with these models is very, very fast. Not only are they getting better with every iteration, but they’re also getting better at a faster rate. Read that sentence again. Get tinkering.

3. Did I already say “get tinkering”?

1. And this, now that I think of it, is the Obviously Correct definition of AGI. You’re welcome.