Are We Going Post-Software?

Hey y’all — it’s scary times.

No, this isn’t about politics. It’s about software.

Yesterday I read the best framing I’ve seen yet of the idea that AI is going to commoditize software development.

Believers say that AI will be able to create software on demand and that anyone will (theoretically) be able to instruct an AI to build custom software.

Skeptics, notably like Garry Tan and Aaron Levie, make good points about whether software that’s created by hobbyists rather than engineers will meet the quality bar that’s required for widespread, frequent use.

This is an existential question for founders — if AI eventually makes it possible for anyone to create software, is it still a good idea to start a software company or learn to code in 2024?

This week’s issue explores the arguments from both sides:

Are We Going Post-Software?

I’ve been thinking about this for over a year.

I remember sitting at brunch with my friend Carlos in NYC last March talking about how the rise of ChatGPT implied the need for a new modality in mobile computing.

We raised a few questions:

1. Would I need to type anything? I can record voice commands.

2. Would I need to see anything? The AI could speak to me in my AirPods.

3. Would I need to do anything? The AI could perform tasks on my behalf.

I went home and wrote this about the types of moats that would still exist for founders.

I now believe both #1 and #3 will end up being legitimate questions, but I favor glasses as a modality because #2 seems unlikely to pan out. It’s faster to see than to listen.

But just how far will #3 go? It brings up new questions:

1. How complex are the tasks that the AI will be able to perform?

2. Will you be able to guide it to build software on my laptop first, basically to create custom software?

3. Will it be able to build software entirely on its own, on demand whenever I need it?

The Argument In Favor

The argument made in the piece is that software now is like journalism in the 90’s.

At the time, producing journalism was hard and quality journalists were expensive.

Then the internet and social media let anyone be a journalist. The number of people creating content spiked, and some of it was actually pretty good.

Journalism is now cheaply accessible, and the social media platforms that control distribution own the most important node in the value chain.

Right now, producing software is hard and quality engineers are expensive.

Now, AI has come along. If it succeeds in letting anyone be a engineer, the number of people creating software will spike, and some of what’s created will actually be pretty good.

Software would then be cheaply accessible, and platforms that can control distribution will own the most important node in the value chain.

Sound familiar?

Ok But… Will It Happen?

“The End of Software” is a hyperbolic title, but there’s truth to it.

Most of the smartest people pushing back say it won’t happen because of the quality of the newly created software.

This is interesting.

It sounds similar to people in the 90’s saying why would consumers want news from random people rather than professional journalists who have their shit together.

It also reminds me of disruption theory. Here are the 3 types of innovation, according to Clayton Christensen:

In a world where foundational models don’t get better, yes, they will likely be a sustaining innovation that helps incumbents lock in their advantages in the market.

But Sam Altman has now said many times that betting that the models won’t continue to improve dramatically is a mistake. This is worth a quick watch:

It’s logical to believe then that future models will produce higher quality outputs and be able to do more complicated tasks.

That means the software AI can build for you will be higher quality as well, and some of those quality barriers will begin to fall.cIt reads to me like potentially a classic example of starting with a low-end market and expanding upmarket as cost decreases (allowing quality to increase).

In the meantime, there are many markets where being able to quickly create something custom that does a specific job is more valuable than hyper-polished enterprise-grade software and AI-generated software will find homes there.

Smart people like Garry and Aaron understand disruption theory and have seen it play out in many markets — so you may be asking where’s the disconnect?

Software is so fundamental to the rise of Silicon Valley, and has been famously “eating the world” for 3 decades now, that imagining a future where it’s a commodity is something that, I believe, will cause many to recoil from and reject outright.

And it would challenge existing Sillicon Valley power structures, much like X is a threat to traditional media.

It’s comfortable to let it seem impossible.

What This Means For Engineers

If everyone will be able to build software with AI, should you learn to code?

Unequivocally yes.

I see a future where you learn the principles of coding specifically to understand how your AI thinks when you instruct it to build software.

Engineers will be able to more expertly instruct the AI and craft more complicated software… and they’ll be able to create significantly more pieces of software in less time (meaning more opportunities to find product-market fit).

So, yes, learn to code (or hire engineers) and build software. But be on the lookout for ways to do it faster. They’re coming.

💡 How I Can Help

🚀 Advertise in my newsletter to get in front of 75,000+ founders.