SaaS Is Dead... Or Is It? — Will AI Actually Eclipse Software-as-a-Service as We Know It?

If you'd like to access the full archive and support my work, please consider subscribing on Substack by following this link. Thank you! ??

While this interview has set the tech industry aflame, let’s not overlook that Nadella’s move was nothing short of a genius piece of marketing!

In case you’re not up-to-date with the latest, let me give you the TLDR version of the story. In a recent podcast interview, Microsoft’s CEO Satya Nadella announced that AI will soon make SaaS as we know it obsolete. He explained that what’s behind all the SaaS services we use today is essentially just a bunch of CRUD Databases. AI agents will soon start using them precisely as such, directly performing operations on them, orchestrating interactions, and fulfilling complex tasks regardless of what domain those DBs belong to. This means that all the business logic that is currently required to interact with those repos will no longer be needed.

This transformation would be so profound that many commentators have labeled this scenario as "SaaS is dead”.

Now, before we even start debating on whether this prediction is accurate or not, let me explain why I think this is such a genius marketing move.

While discussing the potential end of cloud services as we know them, Nadella wasn’t merely promoting Microsoft's cutting-edge AI advancements—an area where the company has heavily invested with the expectation of significant returns. He was also making a clear nod to Microsoft's most lucrative customers: medium and large "non-tech" corporations.

For decades, the biggest unspoken dream of every corporation has been to eliminate their reliance on those nasty and costly system integrators. Now, the prospect that these AI "agents" might finally achieve this has everyone jumping from their seats.

But will this dream actually materialize?

Non All Companies Are “Tech” Companies

If you’re reading this, you likely work in tech. And most probably, you work for a company that considers IT a core asset. This makes our perspective quite skewed and biased as we live and breathe technology.

But let’s not forget that a fair number of non-tech big corporations, and most medium-sized ones, see IT just as something they have to deal with to get business done.

The way a mid-sized manufacturing business views its Asset Management, CRM, and Payroll systems is as a bunch of Web Apps they have to use to get work done. For them, they’re just tools. All they care about is:

1. Does it work?
2. How much does it cost?

They’re usually okay with paying some recurrent OpEx for license and maintenance costs. But whenever a slight change in the business process has to be made, hell unfolds in front of their eyes—IT folks, software vendors, system integrators, UAT, bugs, change management, people training, and the list goes on and on. That immediately means a lot of time, a lot of people, and, most importantly, a lot of money.

This is why Satya Nadella’s podcast interview was so polarizing. You have two opposite camps:

1. Executives (especially from non-tech sectors) who are thrilled at the idea of eliminating the entire IT cost center.
2. The rest of us working in IT, who are fuming and rushing to explain why this is totally wrong.

But the thing is, it’s not a matter of who’s right or wrong—that’s hardly ever how decisions are made.

The Only Decision Driver

Regardless of which camp you’re in, the lowest common denominator will always be one: cost. In business, any decision is always made while considering costs—additional revenue and/or higher margins. Ultimately, that’s the only real driver for decision-making.

So, whether the scenario described by Nadella will materialize or not, will ultimately depend on one thing: cost. So, let’s look into that.

Will AI Drive TCO Down?

In the long-term, I think yes. But not as dramatically as one might think (or hope).

Let’s make the bold assumption that the adoption cost of AI is negligible. When we all know that it’s not so because installation, configuration, prompt engineering, fine-tuning, and domain-specific adaptations take significant time and effort. But let’s focus on OpEx for a moment.

Even though the price-per-token has decreased dramatically in the last couple of years, running LLM-based workloads does not even come close to the cost of running on “traditional” infrastructure.

Imagine you need to fetch the last 5 orders placed by a customer; there is no way using AI will be cheaper than issuing a direct HTTP GET call to an API endpoint. Not to mention the performance.

And your objection would be, “Why on earth would you use AI for that!? Just call the API!”. And you’d be 100% right! You won’t use AI for that! But that’s precisely the point I want to make! This means you’ll have to support both workloads (and both infrastructures). You’ll use:

• Traditional infrastructure for simple, frequent queries.
• AI for complex, open-ended scenarios and multi-step process orchestration.

In my view, the cost reduction from using AI will not come from decreasing infrastructure or licensing costs. Those are likely to increase. The reduction will come from:

• Less ad hoc software development—you will not develop dedicated integrations and orchestrations from scratch every time a new process is required. You will rely on prompting agents and let them figure out what system to interact with.
• Increased productivity—as a consequence of the above, giving a chat-like interface to your workforce will bring the time-to-market of a new process close to zero.
• Workforce efficiency—tasks that require a few button clicks or simple triage of data retrieved from a couple of sources, currently carried our by human operators, will likely be replaced by AI agents.

Will Agents Make All “IT People” Redundant?

I am sorry to disappoint a number of executives and business people here, but the answer is no. Or, at least, not yet. There are several reasons for this.

First of all, as I said, “traditional” infrastructure will still be required, as not all workloads will require or justify using AI. So “IT People” will need to manage that.

Secondly, you’ll need “IT people” to, at least, build and maintain a “trust layer” around the CRUD Databases the AI agents interact with. You’ll need to ensure the specific agent only has a specific set of privileges, can only perform specific actions, can only see the right subset of data, and so on.

Apparently, SaaS is not dead yet!

The truth is that Microsoft and the rest of BigTech will keep selling both, old-fashioned SaaS and the new flashy agentic magic. They already do! You not only have Microsoft AI Agents, but you also have Salesforce Agentforce. ServiceNow also has its own agents. I’m sure there are many others, and more will come.

Does Anyone Remember SOA and BPM?

I don’t know about you, but as I was listening to Nadella, it felt like déjà vu.

Doesn’t all of this look a lot like AI-enabled service orchestration? Remember the good old days when SOA promised that all you had to do was hook all of your services to the orchestrator, define the process in BPMN, and let SOA do the REST (pun intended ??)? Is it just me, or does this look a lot like the same thing?

I mean, I know, Agents are much more appealing because now, in theory, you don’t even need to define a process upfront. You put together your nice query in the form of a text prompt and boom, job done!

But are we there yet? Will it work this time?

Explainability Will Be The New Observability

I spent a few years of my career working on SOA products, and believe me, when something isn’t working as expected, understanding why is a true nightmare.

Observability is one of the biggest challenges with distributed systems, and that is despite these systems being completely deterministic.

Now, have you ever tried testing and doing UAT on an agentic system, or even a simple LLM-enabled chatbot?

Will you today trust your agent enough to grant them unsupervised POST/PUT/PATCH privileges on any of your data sources?

Let’s say you do. Let’s say you grant full access to your agents and they screw up—at scale. What do you do? Who’s accountable for that? Is it the vendor, or is it “IT people” that configured the product? Was it the person putting together the system prompt?

The challenge with AI-enabled systems will be explainability. Assuming that the technical infrastructure is working properly, understanding ‘why’ these agents behave in a certain way and take certain decisions will be crucial.

For this ‘new’ paradigm to take root, you will need to have tools and processes in place to effectively perform ‘intention’ auditing and enforce strict data access governance. Not to mention that AI brings CyberSec threats to a whole new level with adversarial attacks, model inversion, and prompt injection. You need tools and processes for that too.

What’s The Takeaway?

Is SaaS dead? Not yet. Will AI profoundly impact software development, system design, and the IT industry as a whole? For sure! It already has.

What do we make of all of this? What’s the message for folks like us who live and breathe tech?

A few things:

• At least for now, “traditional” infrastructure and services will still work alongside AI workloads. So knowledge and skills required to manage “traditional” infrastructure are still very relevant.
• Mastering AI and its use is critical—we should heavily invest in understanding and becoming proficient in the use of AI.
• Along with AI, new IT needs, risks, and challenges will arise—our technical skills should reflect and adapt to that.

Conclusion

I hope you enjoyed this post. If you did, please let me know by leaving a comment or sharing it with a friend.

If there’s a topic you’d like me to cover in a future issue, feel free to send me a message. I read them all. I’m always looking for feedback, suggestions, and new ideas.

Thanks for reading. See you next week. ????

Cheers,

Fusco

?? If you'd like to know how I can support you, check out this?link.

?? Or, if you’d like to have a chat, send me a message on?LinkedIn.