Before the widespread connectivity of computers on the internet, software often came in boxes with physical disks. I’m not kidding about boxes:

My very first work experience started with this software – I was paid to do some programming using dBase, an early database system.
This is obviously a bygone era, but understanding why is as relevant today as it was twenty years ago.
Software is rarely free, and the disruption caused by generative AI could upend one of the most popular ways software is currently sold. As someone who made a living writing code for 15 years during the last period of tumultuous change, I expect the coming years to be similarly chaotic!
Software business models may sound mundane, but the stakes couldn’t be higher — and the opportunity for bold entrepreneurs couldn't be greater.
(If you’re familiar with the origins of SaaS and its distribution and financial models, feel free to jump ahead to the emerging threat from generative AI.)
Whether it’s an app on your phone, the program used by your bank to produce your account statements, or the accounting software used by a small business you buy from, someone is paying for it. It costs money to run servers, maintain the software, and handle customer support. Often the way that payment is made is through a subscription, paying monthly or annually for ongoing access to the software – but it wasn’t always that way.
A customer bought some software, received the software physically, and could use it as long as they wanted and following the legal agreement of the license. Back in those days, it was common to have a pile of cardboard boxes like the one above – they even had manuals! – and stacks of floppy disks for installing the software on your computer.
Everything changed around the year 2000. A new idea, software-as-a-service or “SaaS” for short, was popularized by Salesforce. This was only possible because the internet connected previously isolated computers, and because web browsers became powerful enough to deliver a software experience over the internet.
As a developer at this time, this blew my mind. Web pages were no longer static – they became canvases not just for written content and images, but for an unlimited range of applications.
Instead of an IT person needing to install software from CDs on individual PCs or expensive servers in a frigid data center, users could simply point a browser at a web page. Buying servers, upgrading software, and being cold in a data center was now someone else’s problem!
That innovation was also enormously valuable: McKinsey valued the global SaaS market at $3T (yes, trillions) in 2022.
When people talk about SaaS, they are often talking about two different, but related, ideas. To understand the opportunity created by generative AI, we need to tease these two concepts apart.
On one hand, SaaS describes how software is delivered to the customer. Rather than shipping software with physical media, the software is delivered over the internet. This model is everywhere: many people use a word processor in the browser, like Google Docs. App stores can also deliver software in a SaaS model using a similar technique, allowing software to be downloaded and then providing the services that power that app over the internet. This is the SaaS distribution model.
On the other hand, SaaS refers to how that software is paid for. Rather than paying once for a license, customers in a SaaS model typically pay on a subscription basis. This model is also everywhere: consumer smartphone apps like Tinder and digital add-ons like Snapchat+ charge for a subscription. In the business world, everything from small business apps like QuickBooks Online to enterprise apps like Salesforce and Microsoft 365 also use this recurring subscription model. This is the SaaS financial model.
If you’ve ever gone through your app subscriptions and wondered, “Why did I buy this?” or worse yet, “What is this subscription?”, congratulations! You are not alone, and this is precisely what finance or procurement teams at every big company go through with their software, too. No innovation comes without some negative consequences.
For people who have only bought or sold software in the SaaS-dominated era, it’s impossible to overstate how significant a change the SaaS financial model was. It transformed the software industry. The success of entire categories of venture capital funds that invest in SaaS software businesses is based on the premise of recurring revenue with high margins.
The emerging threat from generative AI
Software wasn’t always sold as a subscription, though, and the opportunity again exists to disrupt what is now the status quo, as generative AI offers dramatic improvements to efficiency and enables new ways to monetize services. The flip side of this opportunity is a threat to the traditional SaaS model, particularly for software sold to businesses.
The opportunity and threat emerge from two complementary trends: generative AI is simultaneously reducing the cost to build and maintain software, and so-called “agentic” AI use cases are changing how the value of software is priced. (Agentic AI isn’t as fancy as it sounds - bear with me here!)
Historically, one of the largest costs of creating and maintaining software was human labor. While the technology and tools used to create software have gotten much better over the last two decades, the complexity of modern software has also grown. As a result, the economics just haven’t changed much – until now.
In the last year, AI models capable of writing software and fixing bugs have become significantly more efficient. According to one widely accepted industry benchmark, the success rate of the most capable AI models for resolving software bugs went from 3% last October to an astonishing 43% one year later.
My experience squares with this: just two months ago, an AI chatbot rewrote several hundred lines of code for me. Did it need a little help along the way? Yes. But was the lazy developer in me thrilled? Also, yes!
Of course, 43% isn’t high enough quality to remove the need for human software developers, and this is just one benchmark that measures one particular type of coding problem. But the trend is significant and here to stay: the models will only continue to improve and are already good enough that entrepreneurs need to take advantage of what is likely to be rapidly improving cost efficiency for building software.
This could very well be another “Salesforce moment” – a defining period of transition that remakes how billion-dollar businesses are built.
While these new AI models make building software cheaper for entrepreneurs, this technology is available to everyone. Prospective customers of software startups are realizing that they, too, can take advantage of these efficiencies and build software themselves.
Due to the expense and complexity, just ten years ago it would’ve been unthinkable for most large companies to build any “commodity” software themselves unless it created a significant competitive advantage. That cost and complexity are being reduced by AI. And, however expensive or complicated they remain today, they will be less so with the more capable models available in a year.
This will inevitably compress margins in SaaS businesses. The bold, scrappy founder of today will benefit from adopting the AI tools that make software businesses more efficient. There is money to be made by what is effectively an arbitrage of efficiency: delivering equivalent products at cheaper cost enabled by lower expenses. The natural conclusion of this is an efficiency arms race, which is ultimately deleterious to software businesses as it reduces the cost, and ultimately the value to be extracted, from building software.
Luckily, the agentic AI use cases provide an antidote to these lower-margin headwinds for founders.
Ask five AI entrepreneurs for the definition of “agentic AI” and you’ll get seven answers. The specifics vary, but the core idea is this: AI “agents” carry out a task to achieve a specific goal, and each time that goal is successfully achieved, a fee is charged.
It’s helpful to use a specific example: Companies often use software to manage business processes, like handling a customer request for a refund. The traditional, non-agentic approach to this would be a software tool that allows a call center agent to document the request, route it to the correct department, and have an employee follow a series of steps to complete the request. This traditional approach often uses a recurring, subscription SaaS financial model. The business pays annually for a subscription to the software that handles the process and is used by the business’s employees.
Alternatively, using the agentic AI approach, the customer would submit a request, and the AI, working autonomously, would perform all the steps the human employees would follow to resolve the request, either approving the refund and processing the payment, or denying the refund and communicating the outcome to the customer. Agentic AI has the opportunity to present a new model: the business pays only for the requests the AI handles successfully.
This “outcome-based” pricing model aligns the success of business providing the AI agent with the business buying the service.
The business knows that they will pay a fixed fee, say $5, per successful agentic AI interaction. And the business selling the service is highly motivated to make as many interactions as possible successful.
Approaching pricing this way requires some creativity: the old habits associated with recurring subscription revenue business models, and the margins they generate, are deeply ingrained. I get it – I love high-margin businesses too!
That attachment also creates vulnerability to disruption. While legacy businesses dependent on those subscription revenue profit margins cling to the old way of pricing their products, scrappy startups see an opportunity to try new approaches to pricing. As Jeff Bezos once said, “Your margin is my opportunity.”
Entrepreneurs who embrace periods of dramatic change become the leaders of the next wave of innovation. While generative AI is one area of innovation, the innovation in more efficient and different business models may be equally disruptive. The technology industry is now at an inflection point, and the entrepreneurs starting companies today have a choice to make.
Fortune favors the bold.
Really interesting read Chris! Find the opportunity of the salesforce moment extremely interesting!
Chris, great read — "they even had manuals!" really made me laugh. This is such a wild, wild west time for generative AI, and I’m excited to see what other angles you tackle next.