ryanrusnak.com
RSS
~/writing/2026/ valuable-software-in-the-age-of-ai.md

Valuable Software in the Age of AI

Now that SaaS is Dead, what is valuable?

So many things! With code being virtually free, I feel like the world finally took a step towards being more rational. The idea that venture capitalists put so much value on software because it had no assets and scaled instantly makes sense, but it always sat terribly with me. As a guy who writes software professionally, those companies just never felt defensible. Their best defense was just that programmers had better things to do instead of put them out of business.

Now with AI being able to produce software so quickly, companies need to figure out how to create real value or they could be out of business over a long weekend.

So if code isn’t defensible, then what is?

There are probably a bunch of categories of things that are valuable, but in this moment, I can think of three.

Networks

This might be obvious considering how much time people spend on Instagram and TikTok, but networks are just so hard to copy. Anyone could make an Instagram clone in a weekend, but it would be empty. This is why Facebook spends money marketing to entire countries at a time to try to build density. Once they have it, it is so hard to disrupt. There are arguably better versions of different networking apps like Pixelfed and they just can’t seem to slay the giants.

Loops

Companies based on loops are incredibly valuable for a similar reason to networks. This is the category that Airspace is in. We could post all of the code for Airspace publicly online and people would still be unable to create Airspace because the value is the loops! We have over a decade of data on critical packages moving through airports and locations. This enables us to choose the best routing, best airlines, best drivers all in real time making a much safer and faster solution for our customers. And every time we do that, we feed the models that improve them further.

Mapbox is similar. Mapbox has SDKs in apps that are constantly moving through the world and feeding Mapbox servers that makes their product virtually untouchable. Someone would need to launch an SDK and harvest years of data before they could have a competitive product. In the age of AI, loops are king.

Expertise

This is maybe my favorite category because I think it is way more achievable for most people starting companies. Let’s look at the company NeuralDSP. They use neural networks to model guitar tones so instead of an amplifier, you plug in to a computer that simulates the sound of hundreds of amplifiers. So instead of buying thousands of dollars worth of gear, you can get the same sounds with just one piece of software. So cool!

The twist here is there is an open source piece of software, called NAM, that does the exact same thing and is completely free. NeuralDSP is cooked, right?

Nope. In fact, I happily pay for NeuralDSP products and eagerly await their releases. The reason is because open source software isn’t very good at productization. Open source is more… here you go! Use it for whatever you want and don’t call me if it breaks where commercial software is a bit more strategic. NeuralDSP’s secret sauce is teaming up with famous guitarists to create preset tones using their software. So I can just buy the John Mayer preset instead of hunting down what gear he uses and tuning it endlessly until I ultimately give up and decide it is the gear that makes the song sound terrible. Now I can just buy the tone and realize that I am the problem - and I happily do!

This is also the reason that an app like Waking Up by Sam Harris will never be disrupted. Because no app will ever be better at being Sam Harris than Sam Harris’ app. He is the value because he has the info, opinions, and silky smooth voice that everyone wants in their ears and that is what they are paying for.

Full Circle

This also proves my point that code does not matter and was never the valuable thing. Sam Harris could have written that app 100 different ways in 100 different frameworks and they all would have worked. Someone could copy NAM to create Super NeuralDSP over a weekend and it wouldn’t matter because they don’t have the library of celebrity blessed tones.

So the world finally makes sense again. Thanks, Claude.