Growing Ventures
Building Software Businesses for an AI Future
Software-as-a-Service (SaaS) has been the darling of investors and founders alike: build once, charge forever, watch margins expand as revenue scales faster than cost.
But once AI has entered the market and found product-market-fit with software development, SaaS valuations have fallen off a cliff. My take is, they will not recover.
Many AI enthusiasts are proclaiming the death of software as we know it, because agents will take over everything and make SaaS obsolete.
Personally, I don't buy it. But I also don't think things stay the same. In fact, we have seen this first-hand with our portfolio company Pxl.
What Pxl taught us about hitting a ceiling
When we acquired Pxl in 2024, it positioned itself as a short link platform, QR code generator and landing page builder sitting at $340 MRR. Over the course of the following year, we grew it to $1,600 MRR mostly through performance marketing and SEO.
Then, by the end of 2025, we hit a wall.
We realized the already crowded space of short link platforms and QR code builders got even more crowded and performance marketing stopped delivering.
I talked to Nem, my business partner at Waterglass, and we decided to go back to the drawing board. But this time we were not looking to test another acquisition channel, but to rethink the very product itself.
Building for an agentic future
What we landed on is that Pxl will be rebuilt to focus on two interfaces going forward: one for humans, the way landing pages have always worked, and one for agents, for discovery and interaction through AI.
But what matters now is what forced this rethink in the first place: what will SaaS look like a few years from now? What types of software businesses will come out stronger, which ones are going to disappear?
Agents won't kill SaaS, they'll use it
The mistake I see most people make in such conversations is to see "agents will replace SaaS" as a binary outcome: (1) either software survives untouched or (2) it gets vaporized.
I think simple SaaS businesses with low switching costs are going to gradually disappear, but for complex software, neither is what's happening.
If you look at what the biggest AI labs are actually building, you'll notice something interesting: Anthropic didn't build their own PowerPoint or Excel replacement.
Instead, they built agents that use PowerPoint and Excel.
That tells me something important: the people with the most skin in this game, the AI labs with hundreds of billions at stake, don't seem to believe SaaS is going away any time soon. They believe the user layer is just changing fundamentally.
Replacing vertically integrated software from scratch is actually pretty hard. The data models, the compliance, the integrations, the complex workflows that took months to get right – none of that disappears just because a language model got good.
What's easier, however, is abstracting away the interface. The buttons, the forms, the dashboards, the onboarding flows. That layer is where agents are moving in.
So the question for anyone building or buying right now isn't "will SaaS survive." It's whether the value in your product is just the interface or whether it sits underneath, defensible and hard to replicate.
What this means for founders and operators
We updated our acquisition thesis at Waterglass to reflect this. Three categories now sit at the top of our list:
- Businesses with proprietary data. If the data isn't replicable, the product isn't replaceable. An agent can wrap a better user interface around your product, but it can't reproduce your dataset out of thin air.
- Businesses in regulated industries. Compliance is a slow moat, but a real one. HIPAA, GDPR, financial regulations – these don't get disrupted on an AI lab's quarterly roadmap, because there are people's lives and money at stake.
- Businesses with deep integrations. The kind of software that is wired into twenty other systems. This includes digital infrastructure.
The question I keep coming back to is this:
If you stripped the interface off your product tomorrow, what value would be left?