Are AI Agents Actually Going to Kill SaaS?

Everyone keeps telling me agents are about to wipe out SaaS by 2028. I spent a week digging into it. The headline is wrong, but what's underneath is more interesting than the hype.

Here is the short version. Agents are already making it stupid cheap to build small internal tools. That part is real and it is happening now. What they are not doing is replacing the systems your business actually runs on. The data still lives somewhere, and whoever owns that data is in a stronger position than ever.

What I found

Switching costs are the real moat, not the software

I kept looking for the case where a company rips out its core platform and rebuilds it with agents. I could not find it at any real scale. Once your customer records, your billing, and your history live inside a system, the cost of leaving has almost nothing to do with how good the new tool is. That ceiling holds no matter how capable the agents get.

"Replacement" is really just the interface coming unbundled

What agents are good at is generating the screen you need in the moment. The button, the dashboard, the quick view. That is the interface layer. The system of record underneath, where the data actually sits, is not going anywhere. So when people say SaaS is dying, what they usually mean is the front end is getting cheaper. Not the same thing.

The incumbents are moving faster than the startups

This one surprised me. The big platforms are bolting agents onto what they already own quicker than new players can rebuild the data layer from scratch. They have the customers, the data, and the distribution. A startup with a better agent still has to win all three of those before any of it matters.

Generation was never the bottleneck. Trust is

Spinning up a tool is the easy 20 percent. Keeping it reliable, safe to run against real data, and trusted enough that someone bets their workflow on it is the hard 80. That is where adoption actually stalls, and it is the part the demos never show you.

The thread that connects it

It was never about the model. It was always about who owns the data.

Every angle I looked at came back to the same place. The moat is not model quality, and it is not who has the slickest agent. It is data gravity. Agents do not move that power away from the incumbents. If anything they hand them more of it, because the value shifts to whoever the data already trusts.

What I'm still watching

The thing nobody has priced in yet is what happens when agents start touching production data at scale and the compliance world wakes up. The moment security and regulation enter the picture, the timeline stretches and the companies with the cleanest audit trail win by default. That is the question I cannot stop thinking about: who captures the value when the screen is generated on demand but the data still lives in someone else's vault?

Sources

• a16z — The Economics of AI Agents (2025) — market and unit-economics breakdown
• Stanford HAI — AI Index 2025 — enterprise adoption data