What did Satya Nadella mean by "SaaS is Dead"?

A few weeks ago Satya Nadella was on the BG2 podcast and discussed a wide variety of topics, from leadership, to becoming the CEO at Microsoft, to AI, to Growth Mindset. I recommend listening to the podcast in its entirety as it really is worth the investment. With all that said, the headline from the podcast that struck a chord globally was "SaaS is Dead". What did Satya mean by that statement and what do we do about it?

The statement came when Satya was asked an interesting question (paraphrased): "does the development of first principles AI disrupt traditional business apps?" For example, let's say you use Excel to determine the answer to a budgeting question. However, you might not need to actually use Excel to determine that answer, since an AI agent could answer the question you need answering. The agent has access to the data and the logic to determine the answer, so there is no need to build calculations only to arrive at the answer 'manually'. In this context, the Excel tool may end up being irrelevant because it is simply a means to arrive at the answer, not the answer itself.

Expanding this to the idea of a typical application we've been building forever, each of these applications are essentially "forms over data", with some business processing behind the scenes. We perform CRUD operations against the databases for various purposes and then make decisions based on information we see or run an automated process (such as manufacturing) to keep other systems up to date (such as with financial or insurance use cases). At the end of the day, all application's share many of the same concepts. We made these more accessible by "webifying them" or making them available from an app store, increasing the ease of interacting with the same CRUD model we've been running for a while.

What Satya is arguing for is a major pivot in the way we think about application architecture, the means of interacting with humans, and the manner of accomplishing automated actions. All of the above will take the form of interaction with an AI agent, which is part of a network of agents, each of which performs a necessary and important task. An agent might interact with the human and get the request, whereas another might orchestrate the updating of multiple databases, whereas another might manage the execution of a dozen autonomous tasks. The idea is that instead of having 100 different applications we interact with, we might instead interact with an AI agent that does most of that work for us. The same being true for the automation layer of the application and how activity is executed.

Another way to think about this is by looking at your mobile device. How many unique CRUD based apps to you have on your phone. You likely are thinking way-to-many, which is likely the case for most of us... ;) That said, imagine for a second that you can interact with that entire ecosystem via text/speech/motion, etc. We've seen rudimentary shots at that goal before with Siri and Alexa, but neither has really hit critical mass. The scope of their capabilities was too weak to truly act as an actual "virtual assistant". However, what if that were true? What if we really did have a virtual assistant?

So, to make this real, imagine you have a medical record system that presently has a doctor updating information manually. That doctor has to swivel chair between the patient and the computer screen. I have this very experience with my eye doctor. I asked him "how do you like taking all these notes digitally?" He said, it's actually very distracting, because I can't focus on my patient. The next time I came to the same office he had an assistant in the room as we were doing my appointment. The assistant was capturing notes in the electronic medical record system while he was interacting with me. Now... tell me that is an efficient use of time? Of course not. This is technology creating the very definition of inefficiency.

Now, imagine a Medical Copilot that captures all the notes, makes suggestions, and answers direct treatment plan questions for the doctor. It listens to everything in the room, captures the notes, is able to interject (collaborate) with the doctor. The doctor may never need to actually even interact with the medical record system because the assistant, which in this case is an AI agent, is able to provide all the information he needs. In this case the cost goes down, the quality goes up, and the patient-doctor relationship is improved.

Second, imagine a scenario where a back-end rules engine evaluates insurance claims that are submitted through a rudimentary but predictable if-then-else set of statements. A possible future state is that those claims evaluations are replaced by a more sophisticated AI agent evaluation that is capable of applying the standard if-then-else rules but also is capable of "understanding" previously unrealized patterns that increase the quality of customer relationships. We've all had scenarios where we've had claims denied and had to argue with our insurance carrier. What if the result of a denied claim was for the AI agent to "take a second look" and seek to maximize the positive relationship between the carrier and the patient, with a goal of finding a win-win ratio. The if-then-else rules engine is inferior at that and thus has a higher rate of false negative. The addition of an AI agent can augment that conversation and create greater alignment.

So, where to go from here? What are the things that need to be prepared in order to respond to the idea of systems of agents that take action? The first thing to always remember is "successful AI adoption is about actualizing the mission of your business".

Ok, that's great... but what does that mean? It means that your mission is more than your delivery model or even your direct product. Think about the outcome your organization provides to its core customers. Now, imagine how you can create new ways of fulfilling that value that disrupt the process of finding, selling, quoting, configuring, delivering, and performing Q&A on that product. Think about how the information you know about your customer can be used for their benefit, not yours. If you ask yourself those questions, you can think about how another company might disrupt you. Now, ask yourself, how can I be the disruptor, not the one disrupted.

The key question to think about is, "what would delegating to an AI agent look like in the context of the mission of my business?" I urge you not to just think about the skin deep perspective on this, but truly take it to the next level. Then, the question is, how do I build a plan to make this true? How do I invest toward the possible future before someone else gets there first? The task before every company is to 'be the future'.

Let this be a moment to not only understand architectural changes are coming to front-end and back-end applications, but also that the very nature of how we deliver services is changing.

Nathan Lasnoski