The End of Software?
This article was making the rounds on Twitter today:
The End of Softwarehttps://t.co/JWg6QYqLzO
— Chris Paik (@cpaik) May 31, 2024
It’s a good take – in the sense that a good “take” is something that half of readers will vehemently agree and the other half will vehemently disagree, and I can’t help but share my thoughts.
There were broadly speaking two effects that heralded the demise of newspapers and similar media, and they are both touched on in the article: firstly, the marginal production and distribution cost going to zero. This meant that suddenly all newspapers were competing with each other, and that wasn't great for business – if I can get the same news from multiple sources, and they all have marginal distribution costs of zero, then competition will drive the price to zero. This is what happened: undifferentiated newspaper content on the internet is free. The cost to produce that content has nothing to do with it. It's worth noting (and called out in the article as well) that software has had zero marginal production and distribution cost for a while. There are also parts of the software industry where the business models are nowhere near as attractive as those of the Salesforces of the world. Take mobile gaming, for example: a super competitive field where the vast majority of games is barely profitable, and most of the generated revenue goes to Meta and Google as Customer Acquisition Cost (CAC). There are also some categories where the cost has literally been pushed to zero, for example, open source projects.
This brings us to the second effect that affected the media industry, and that the article glosses over a bit: substitutability. Articles from different newspapers were already substitutable, and once the distribution barrier fell, it started a competitive dynamic that drove down prices. It turned out that user-generated content could also substitute media content to a large extent (but with much lower upfront cost). This is also the reason that fields like mobile gaming are so much less economically attractive than eg B2B SaaS – most games are easily substitutable. The ones that are economically attractive have achieved some sort of non-substitutability, like network effects or cultural significance. Why is it so hard to build another successful social network? It's not because building an app like TikTok is expensive, but because of the network effect (and TikTok spent a billion dollars to bootstrap its user base), which makes a TikTok clone not a good TikTok substitute. This is not new, by the way! Investors have been asking about the "unfair advantage" of companies they invest in for a while, simply because software isn't all that expensive to create!
If generative AI makes software much cheaper to create, the question that follows is not whether that will lead to more software – it undoubtedly will. The question is if it also increases the substitutability. The article implicitly argues that it will (replacing Salesforce with "a constellation of things that dynamically serve the same intents and pain points"). I think this remains to be seen. Some categories with already high substitutability will see even increased pressure (eg. gaming, what if the alternative to a game from the App Store is a game that is dynamically generated for you). Some other categories might be pulled into higher substitutability. Other pieces of software won't get substituted simply because comparable software is cheaper to build. Facebook, Instagram or TikTok won't get substituted because someone can build the same software cheaply. If they get substituted using generative AI, then more because it became cheaper to generate substitutable content.
The B2B SaaS space (which is the one mentioned by the example of Salesforce) is particularly interesting. Personally I don't think that B2B SaaS will be replaced by a constellation of cheap software tools, because standardization and integration has such advantages for software buyers. Consider the case of Slack vs Microsoft Teams: outside the startup sphere, Teams has largely won – not because it's superior software, but of its integration with the rest of the Microsoft stack. I think the only way that generative AI and cheap software makes a difference on the B2B side is if every company builds their own custom internal software stack, top to bottom. This is possible, but I don’t know that it’s likely.
That doesn’t mean that the software industry won’t change. I think that one argument in the article is a bit backward: I don’t think software business models are attractive because software is expensive to create – I think it’s partly the other way around. The reason that big tech pays newly graduated engineers six-figure salaries, and senior engineers seven figure salaries is because of the very lucrative business model, and because demand for capable engineers is high. That is something that might change with generative AI. If one engineer will be able to do the work of 10, 20, or 100 engineers with the help of generative AI, then demand for engineers will drop and therefore salaries will as well. That’s the reason why maybe being a CS graduate today is not the best position to be in – but it’s maybe not like a journalist in the 90s, but more like a farmer in the early 1900s, or a weaver during the industrial revolution, where a whole lot of your working opportunities are going to be automated away.
PS: The other way that generative AI might mean the end of software is if we see the emergence of AGI where instead of using software to complete any task, I might just ask an AI agent. However, that world might completely upend societal and economic systems (why have a company when you can just have a swarm of agents), and if that also means the end of software, that might not even be noteworthy.
PPS: It’s funny, and I am not sure if intentional or not, how the article uses Salesforce as prototypical software, when Salesforce’s first marketing campaign was to literally proclaim “the end of software”.