Writing this at the start of 2025, I can say with a high degree of certainty that this piece is unlikely to age well, but at least we'll be able to look back and laugh about just how wrong we were.
Jumping right in - why might SaaS be 'dead'?
Death of the Backend
For context, let's restate Microsoft's Satya Nadella's December 2024 assertion that SaaS as we know it is dead:
I think the notion that business applications exist, that's probably where they'll all collapse right in the agent era, because if you think about it, they are essentially CRUD databases with a bunch of business logic. The business logic is all going to these agents and these agents are going to be multi-repo CRUD right, so they're not going to discriminate between what the backend is - they're going to update multiple databases and all the logic will be in the AI tier so to speak. And once the AI tier becomes the place where all the logic is, then people will start replacing the backends.
It's fair to say this interview sent shockwaves through the industry - not because it was the first time anyone had floated the idea that SaaS was moribund, but perhaps because it was such a concrete formulation of how it was going to happen coming from a senior industry player. But combing through the comments on YouTube, reaction videos and subsequent blog posts, it was clear just how polarised opinion was on how this might play out. A lot of experienced engineers refused to countenance the possibility that backend logic could be subsumed by LLMs, exemplified by comments like this:
I'm a SaaS enterprise architect and have been coding for 40 years. This is 100% absolute nonsense because you will never predict the detailed requirements of the customer, who themselves (particularly, no single individual) can't define them accurately. Agents will only facilitate execution of pre-defined business logic on behalf of the user, and might express results in some fancy way, but the business logic itself is usually highly optimized and specialized, and usually discrete and purposely immutable, which AI is not.
On the other hand prominent YouTuber Matthew Berman has this to say in his reaction video:
"If you've been been watching this Channel at all, you know I've been talking about this for a long time. I truly believe the entire application stack is going to disappear - there just really is no need for it when you have artificial intelligence interacting with the core grounded data that sits within a database and again what does that actually mean for the entire SaaS industry? What does that mean for application developers? I don't know exactly but I have a feeling it's going to look vastly different than it looks today and I've been in SAS my entire career I have built multiple SaaS companies. I've worked for SaaS companies and if I'm being transparent I probably wouldn't start another SaaS company right now and I probably wouldn't invest in any SaaS companies right now either."
Whilst black and white thinking is attractive, it doesn't help much strategically because the final outcome, as always, will probably lie somewhere on a spectrum and it will depend very much on the SaaS in question. My own feeling is that logic-heavy, complex, transactional, domain-expertise-laden platforms (think ERP, accounting and the like) are probably 'safer' than simpler systems with less ingrained business logic (to-do applications come to mind). In other words, the closer you are to CRUD, the more your platform looks like a thin layer over an SQL database with a dollop of custom UI, the more at risk your software. I'm by no means sure about this, nor could I say for how long that relative margin of safety might last. Developments in LLMs over the past month show that making any kind of predictions about their abilities is likely to make one look foolish, at best.
Service-as-a-Software (or SaaS Inverted)
Before Nadella's bombshell comments in December 2024, the venture capitalists had already cottoned on to the fact that SaaS as we know it was heading out the door, to be replaced with a new software-like industry: vertical AI. The story goes something like this: modern SaaS is built for human teams to use. Humans cost money and need to be fed and watered. AI can replace many human workflows, especially those that are language heavy. So rather than building software for human agents, let's build software that comes with agents built in. In other words, as a business owner or manager you won't buy software for your team, the software will be your team. Listen to Jared Friedman from Y Combinator weigh on why vertical AI agents could be 10x bigger than SaaS:
"Here’s my pitch for 300 vertical AI agent unicorns. Literally every company that is a SaaS unicorn you could imagine, there’s a vertical AI unicorn equivalent in some new universe. Because most of these SaaS unicorns, beforehand, there were some box software company that was making the same thing that got disrupted by a SaaS company. And you could easily imagine the same thing happening again, where now, basically, every SaaS company builds some software that some group of people use. The vertical AI equivalent is just going to be the software plus the people in one product."
Of course this doesn't mean the SaaS incumbents are going away overnight, just that the industry as whole will gradually shift from SaaS to Vertical AI, just as it shifted from 'box software' to SaaS a generation ago.
The reason why this shift is just so tantalising to VCs comes down to where the increased 'software' spend is going to come from. Listen to Caty Rea from Bessemer Venture Partners explain it in a video entitled Vertical AI shows potential to dwarf legacy SaaS:
"They’re spending almost as much on this little kind of one-off service as they are on their core system of record. So, I know you know the answer to this, but, why does that make sense? Why aren’t they balking "wait a minute, we can’t double our technology budget in a year" — how do they think about it? Yeah, I think, you know - it’s a leading question, and the reason is because they’re really not. They’re just replacing existing services spend, and so they don’t need to create entirely new budget. Instead, it’s a reallocation of services spend to something that is productized thanks to AI."
Put simply: replacing humans with AI agents. Beyond the obviously seismic socioeconomic earthquake that's likely to cause, there's a fundamental nugget of wisdom that can be extracted from this assertion. In the example they are talking about, which concerns the 'legacy' SaaS platform Litify (a system-of-record workflow software for personal injury law firms) and the new vertical AI startup EvenUp (sells a service to those personal injury law firms drafting legal demands to insurance companies), the incumbent SaaS product and incoming vertical AI product live side by side. I think this idea is worth dwelling on a bit longer. Returning to the Bessemer roundtable, they see the new opportunity as split two ways between software entrepreneurs who are quick to identify AIable workflows and professional service companies essentially AIising their offering:
"If you can capture some of that human spend, that’s a really powerful opportunity. And, and I think it’ll take two forms. One is the EvenUp-style of a denovo venture-backed company: a really talented entrepreneur identifies a workflow, automates it, and, you know, it’s lightning in a bottle. I think we may also start to see traditional services businesses—you know, BPOs, law firms, accounting firms, experts, adopt these types of technologies to drive automation inside of their business. You ask for a spicy take: I think the demise of professional services firms is a bit overblown. If anything, some of the value associated with AI may end up accruing to these professional services firms that can harness it."
That's an interesting, and much more savory, option for us SaaS founders and one we'll explore when thinking about AI strategy.
But before we do that, let's briefly address the elephant in the room.
Software is Worthless
A parallel but somewhat related debate is currently raging amongst software developers in the comments sections and on Reddit. It concerns the answer to the following question in different guises: "I'm just about to start a 5 year Computer Science degree in the hope of becoming a software engineer/developer, should I bail?". As you'd expect, opinion is polarised, not least because software development tools have been making engineering 'easier' for decades now. Higher levels of language abstraction, frameworks, superb developer tooling - all have increased developer productivity. So there's an argument that the new AI-powered development paradigm is just an extension of a trend that perhaps started with 'no code' or 'low code' builder platforms and now extends to LLMs doing most of the coding for developers.
My own feeling is that this time it's different (I know, famous last words). I've tried Lovable, one of the new breed of AI full-stack development tools and the results are astounding. They are limited in their scope, I get that, but where they are a good fit, which is quite honestly a lot of use cases, they are undoubtedly an industry-shifting tool that is only going to improve. I watched a non-developer founder friend build a full SaaS platform MVP in about 5 days - a process which he'd previously been through only 3 years early at a cost of hundreds of thousands in seed capital and probably about a year in time.
He and I don't necessarily see eye to eye on whether this is a good thing for SaaS founders though. For non-technical or semi-technical founders you can see this type of tool as either a godsend or a death knell. Yes, you can build a prototype, an MVP or even a somewhat functioning SaaS platform (remember, today's AI is the worst you'll ever use) in essentially zero time at essentially zero cost, but why bother? SaaS, which was once gold, is now sand.
At least that's one, not particularly nuanced way of seeing it. We'll see how the 'software is worthless' argument fits in with the different strategic approaches to AI that I've identified and will now outline.
7 Strategies
So, as a SaaS founder, what are your options right now? After a pretty extensive round up of current thinking, these are what I think represent the main strategic paths - a conceptual framework for SaaS operators to think about the way forward with AI. It goes without saying that they are not mutually exclusive, you could adopt one, two or all of them. I've tried to list them in rough order of how deep they involve jumping in, starting from 'don't go anywhere near the pool', through 'sit on the edge with your feet in' to 'nose dive straight into the deep end'.
Strategy #0: Milk the Cow
It might sound like sticking your head in the sand, but paradoxically I think 'do nothing' should be the default option for many SaaS platforms, particularly those that fit some or all of the following criteria:
- Mature, been in the market for a long time
- Lots of existing customers
- Large existing revenue base
- Little capital or resources to build AI
Something that forward thinking technologists often overlook is that adoption is sloooooow and legacy software often lasts a really, really long time. A bet you know examples of really big businesses still using spreadsheets to run their entire organisation, and maybe even signifcantly sized businesses using pen and paper. I have a friend that works as a well paid Fortran programmer for one of the biggest banks in the world. Fortran for God's sake.
So riding the long adoption curve of AI and milking the cash cow as long as possible whilst perhaps not building too much 'worthless software' could be a decent strategy for many companies. Just don't expect it to last forever.
Strategy #1: Be the Database
Of course, I don't literally mean just the database. Postgres is the database. But if your bet is that at least some business logic layer, however thin, will always be necessary, then this is your play. Essentially, you're predicting that core system-of-record type platforms will continue to exist for a very long time and won't be displaced by AI-only LLM-powered logic layers.
Here's who I think this applies to:
- Mature, been in the market for a long time
- Enterprise
- Platforms with very complex, transactional business logic
- 'Wide verticals' (eCommerce, ERP, CRM etc)
To 'be the database' in the forthcoming age of AI does mean adapting to AI-centric workflows and UI - it just means you won't be the one building them. You will watch as the battle for supremacy in the 'AI Platform' wars unfolds between the really big players like OpenAI, Microsoft, Anthropic, and Google as well as the current and future agent building platforms like Crew and Langchain. You'll expose your core business logic via well-documented and well-structured APIs that can be connected and consumed by agents, allowing your product to remixed into and combined with an ever evolving suite of other backend systems-of-reference, but you're probably doing this anyway, right?
Strategy #2: Sprinkle
This was the first and most obvious strategy to emerge in the early days of generative AI, before we all started getting obsessed with agents. The idea was to give the admin user simple, one-shot tools to ease annoying, otherwise manual tasks like:
- Removing a background from an image
- Drafting a reply to a customer ticket
- Create cookie-cutter content for marketing
- Creating templates for marketing emails
- 'Reading' a paper receipt and creating an expense record
I think for a while it seemed like this would be where AI's role in SaaS would start and end. It felt like sprinkling a little bit of magic on top of existing admin panels, but definitely not a step change, much less the 'death of SaaS'. Oh to go back to those days...
Strategy #3: Customer Facing
Possibly a variant of the 'Sprinkle' strategy, this approach applies to SaaS platforms that have both 'admin' and 'end' users - the canonical example being eCommerce (where 'admins' run the merchant businesses, and the 'end' users are the customers that shop on their stores). Examples of adding AI to customer-facing features include:
- Custom product recommendations and personalisation
- Customer-service bots
- Powerful search and help
The danger I see here for incumbent SaaS platforms is that a lot of these functions can and will be built better by third-party vertical specialists. Data integration will continue to be a problem, as it always has been, but there are thousands of plugins and widgets in the 'customer facing' category that you can add onto your WooCommerce, Shopify, Wix or Magento store, and I don't see that changing - in fact I think its a category that's set to explode in terms of third-party development.
Strategy #4: Copilot
Now we're really jumping in, perhaps not right at the deep end, but definitely not the kids pool. Copilots are starting to pop up everywhere in the SaaS world (and the software world in general). You can have a copilot write a function for you in Excel, reword a response to a customer in a ticket system, or pull up a report. Whilst some might already consider this 'agentic AI', I don't class it as such in my strategic thinking because Copilots are generally:
- Not autonomous
- Limited to small sets of well defined tasks
- Not able to interact with each other
As always, there's a spectrum here between a simple copilot, more complex copilots, and fully agentic AI agents (SAP's Joule appears headed that way) and there's crossover with the 'Sprinkle' strategy (Shopify's Magic appears to be a good example).
From a strategic viewpoint, its a pretty safe, if vanilla, bet, pretty much independently of what your platform is and does - they work well almost anywhere. Time will tell whether the copilot UI and use paradigm really beds in and becomes an expected and accepted interaction pattern, in which case most platforms will probably be forced to add some kind of copilot - in other words, copilots become as ubiquitous as hamburger menus, dialogue boxes and popup notifications.
Strategy #5: Fully Agentic AI
Firstly, let me be clear about what I mean here. I'm talking about adding a fully built out suite of agents, their orchestration and interaction, and the UI to build, manage and communicate with them directly into your platform. This means that users of your platform can basically put their feet up and have the software do the work for them.
Now obviously this applies less the thinner your vertical is - if your SaaS is a one trick pony then you can just do 'vertical AI' (see below) by either launching a parallel offering or completely replacing your traditional SaaS product with a vertical AI product.
No, this strategy is going to be primarily for large, somewhat multi-purpose SaaS - ERP, eCommmerce, CRM etc - platforms that border on the general use or 'wide vertical' or even 'horizontal' definition.
Right now we're only seeing this in a handful of cases, and only with the real enterprise heavyweights like Salesforce's AgentForce.
My feeling is that this approach is going to be out of reach for small and mid-market SaaS platforms who would be better served focusing down on their core functionality and integrating into the wider agent ecosystem.
Strategy #6: Vertical AI and/or Service-as-a-Software
I was in two minds whether to split this into two separate strategies but really it's two sides of the same coin and depends more on how the product or service is projected. Vertical AI is both software and a service and I suspect the distinction is more marketing than anything else.
Should incumbents go after slices of the vertical AI market? I think it depends on whether there is really high-value core expertise (or even straight code) that can be ripped out of its comfy bed in your existing SaaS platform and transplanted to an AI-powered workflow which can be packaged up and sold as 'Service-as-a-Software'. That, of course, is fraught with risks - does it devalue your existing offering? Do you have the expertise and resources to build a whole new product? What is the exposure to competition for this particular vertical?
Ultimately, if the VCs are right, this is where trillions of dollars worth of value will be created in the new AI 'software' market, so it might well be too tempting an opportunity for many entrepreneurial SaaS operators to ignore.
What I do suspect we'll see are SaaS companies leaning heavily on AI for the professional services they already offer - think support, implementation, integration, customisation and training. There are quick wins to be had here for any SaaS vendor, not only for cost cutting but also for streamlining and speeding up onboarding pipelines, the bane of any non-trivial SaaS platform.
Conclusion
SaaS decision makers are somewhat caught between a rock and a hard place right now. With the AI landscape changing at breakneck speed, it's almost impossible to make a big decision with lasting repercussions. For what it's worth, here's how I'm thinking about this in my own company:
- What is our core value proposition? (Business logic, UI, workflows)
- How far from our core value proposition should we stray?
- How vertical is our vertical? (Wide, general purpose vs narrow well-defined vertical)
- What are our short, medium and long term business goals? (Customers, revenue, business value)
- Should we look outwards (integration into ecosystem) or inwards (try to capture AI value internally)?
- What is the appetite for AI adoption in our segment? (Conservative vs forward-thinking)
- What resources do we have available?
- How can me improve internal operational processes with AI?