Rumors of the Death of Software are Greatly Exaggerated
One of the topics we talked about at our LP day in Paris a couple of weeks ago was: Is SaaS dying? This was before this post about “The End of Software” sparked a lot of discussion on Twitter the social platform formerly known as Twitter a few weeks ago, so the topic is even more relevant now. It’s not a new topic though. Here are two slides from our LP Day in 2018:
Since that LP Day in 2018, the market cap of public SaaS companies has more than doubled, growing from approximately $690 billion to over $1.7 trillion, despite the very significant correction in 2022. So, I’d argue that anyone who predicted the end of SaaS six years ago was either wrong or too early.
Is the prediction going to become true this time around? Let’s start by trying to understand what exactly people mean when they talk about the end of SaaS or the end of software.
When people talk about the end of software, oftentimes it’s because of a too narrow definition of software.
In some cases, the issue is mostly semantic. When Marc Benioff famously declared the “End of Software” in 2000 and staged a fake protest outside of the annual Siebel user conference, it was a genius marketing move, but the end of software marked the beginning of Software-as-a-Service. Software was never going to die. The old way of building and delivering software was. Similarly, AI is not going to kill software, it is software.
However, there are at least two lines of thought that aren’t merely semantic:
1) “SaaS has become a fast-food franchise”
The first alleged cause of death is that running a SaaS company is now like managing a fast-food franchise: high competition, limited differentiation, and low margins. The argument is that software development has become so easy that there’s no differentiation or defensibility in the product or the technology.
I don’t find this argument very compelling. If I think about the companies from our portfolio and other startups we’re talking to, product is a key factor. Always has been, always will be. Knowing what to build and forming a great product and tech team that can deliver on that roadmap is crucial. Sales and marketing are, of course, critically important too, but so are product and tech. This is why the vast majority of P9 portfolio companies have one or more technical co-founders.
It’s true that it has become much easier to build a basic SaaS application. What gave a company an edge 5–10 years ago is now table stakes, but that doesn’t mean that the best entrepreneurs don’t find new ways to differentiate and out-innovate the competition.
With AI (more on that below), this trend will continue and accelerate. Thanks to products like GitHub Copilot and our portfolio company Poolside, more and more code will be written by AI. If everyone is equipped with better tools, the bar will go up further. But it doesn’t follow that there’s no room for differentiation unless you assume that what can be built is somehow finite.
An adjacent argument relates to the fact that in the last 15 years, the SaaS business model has moved from obscure to obvious. The idea is that SaaS metrics have become so well understood that SaaS companies are perfectly priced. As a result, there’s no opportunity for outsized returns for investors and investing in SaaS companies will resemble investing in a broad index of stocks.
I would argue that this is wrong. If you look at most public SaaS companies and do a regression analysis of metrics like growth, FCF margin, NRR etc. vs ARR multiples, these metrics explain only 50–60% of the multiples. So the rest must be due to less obvious metrics, views on the market, quality of the leadership, or other subjective factors that smart people disagree on. If this is the case for large SaaS companies in the most efficient capital market, imagine how things look like for a Series A or B company, let alone a seed-stage startup, where not easily quantifiable factors play a much, much bigger role.
Similarly, a few years ago people argued that SaaS has become so predictable that it can be financed with debt. One of the things I love about SaaS is that recurring revenues make it comparably predictable, but as the last few years have shown, there’s still a huge amount of uncertainty. Look at how growth rates of SaaS companies have declined across the board — I think no one expected this.
2) “AI is eating software”
When you hear about the end of software these days, it’s usually related to AI. There are two flavors of the argument.
One is that thanks to AI, software development will become so cheap that every market will become hyper-competitive. This is related to the first point in the franchise argument. I very much agree that software development is being revolutionized by AI, but again, my thinking is that it won’t prevent the best teams from innovating.
A related line of thought is that software development will become so easy that the market for software products will shrink, and companies will develop more custom software based on exactly what they need. This may happen in some areas, but I doubt that non-tech companies will suddenly become awesome at building great software for their teams.
The other flavor of the argument is that all of the value will accrue to a small number of foundational model providers and hyperscalers. The idea here is that AI is going to replace software as we know it. How could this look? I guess no one knows exactly, but imagine you have a highly intelligent successor of ChatGPT that has access to all of your company’s files and documents (or let’s say data and knowledge, maybe by that time we will no longer have files) and that can do actions on your behalf. If you think about what people do when they use a business application, a significant part of it is looking up some information from a database (e.g. a customer record in a CRM), doing some action (e.g. sending an email to the customer) and updating the database (e.g. updating the CRM). With a very large context window, RAG, or other techniques to make data accessible for an LLM, you can imagine a future version of ChatGPT acting like an application based on high-level instructions that you provide in natural language.
The way you’ll interact with this “application” could be a combination of natural language (you tell the AI what to do, via text chat or voice) and a UI that the AI creates on the fly based on the input it needs from the humans. If you think this is far-fetched, ask ChatGPT to act like an address book, enter some contacts, and look up some “records” from your “database” based on different criteria. You might be surprised how well this works already today.
Now here comes the big “but”. What I described is, of course, a simple “Hello World” type example. To go from here to a serious, multi-user workflow solution with integrations, user permissions and complicated business rules, let alone to something that can replace, for example, a CRM, ERP, or HR system in an enterprise, is a huge leap. Aaron Levie, founder & CEO of Box, said it well in a response to the “End of Software” tweet:
[…] Enterprises don’t want a “constellation of things that dynamically serve the same intent and pain points” instead of Salesforce. Their jobs, supply chains, and quarterly numbers are on the line, unlike the risk/reward of reading an article from Vogue. AI will definitely change quite a bit about software, but mostly that it will make it more important and valuable, not less.
It’s also uncertain if foundational models will ever become great in highly specialized domains where deep knowledge about industry workflows and access to proprietary data to train models are necessary. Until we have truly super intelligent AU, specialized solutions for specific use cases or industry verticals will likely be better.
Here’s how Matt Turck, Partner at FirstMark, put it:
[…] What happens next:
* customers realize that “build” is a headache, not always good
* OpenAI / Azure etc can’t / doesn’t want to build hundreds of problem specific/ vertical specific apps
* Takes time, but legacy and new SaaS companies truly become AI-first (not just marketing), abstract away complexity of deploying LLMs
* macro environment eventually rebounds
* AIaaS becomes the new SaaS — what is old is new
The notion of the “death of software” is greatly exaggerated
Given the progress of AI in the last years and the speed with which AI is transforming the way software is developed and used, any prediction comes with a high risk of not aging well, but my best guess is that the fundamental need for software persists — and AI will enhance the value and importance of software rather than diminish it.
