Why agents are the next SaaS

When I was working on the commercial side at Happeo, the maths was simple. You find a company. You convince them your software makes their employees better. You count how many employees they have. You multiply that number by a monthly fee. You lock them into an annual contract. That was the entire SaaS business model for the last two decades. We rented out digital tools for human hands.

The model worked beautifully. Investors loved it because the recurring revenue was predictable. Founders loved it because once a company was using your tool, it was hard for them to leave. I have sold software this way, and I have bought software this way. But as I spend my days building AI infrastructure at Substrat, I am looking at how agents operate, and I realise that equation is entirely broken.

The end of the seat

For a long time, software made us faster. At Mediapilot, when Fredrik and I were growing the business, we eventually created more than a hundred jobs. Every time we hired someone, we bought more software to make them efficient. We bought CRM seats, accounting seats, project management seats. We were happy to pay the monthly fee because a human using a tool was vastly more productive than a human without one. The software was a multiplier for human effort.

Agents change the fundamental transaction. An agent does not need an interface to click on. It does not need a login screen, a dashboard, or a password reset link. It receives a goal, it makes a plan, and it executes it. If you build an AI agent that successfully handles all first-line customer support, the human support team naturally shrinks. If you are charging per seat, your revenue shrinks exactly as your product succeeds. The better your AI performs, the less money you make. That is a terminal business model.

We spent two decades renting out digital tools for human hands, but you cannot sell a seat to an algorithm.

Selling trust instead of tools

We have to figure out what we are actually selling when we cannot charge for a human sitting in front of a screen. The easy answer is that we will sell outcomes or consumption. You pay for the completed task. I think it goes deeper than that. When the software does the work instead of helping a human do it, we stop selling tools. We start selling trust.

If a machine is doing the work without a person clicking a button to approve it, the value shifts entirely to safety, reliability, and accountability. I see this sharply with Anchor. We are building AI to help alienated fathers navigate a system that is often stacked against them. The stakes are incredibly high. A hallucination in a legal document or a poorly worded communication to a lawyer can ruin a life. We are not selling a tool to help these men write faster. We are trying to provide a reliable system they can trust when they are at their lowest point, completely exhausted by their situation.

In the agentic world, trust, audit trails, and liability become the actual product. You will not pay for the user interface. You will pay for the guarantee that the agent did exactly what it was supposed to do, and absolutely nothing else. You will pay for the detailed log file that proves why the agent made a specific decision. You will pay for the peace of mind that if the agent makes a mistake, the provider takes the responsibility.

A bet, not a prophecy

I sit in Marbella, playing tennis badly and updating my written life plan every year. I score my family, friends, health, finances, work, and personal time out of ten. I have done this since around 2010. It keeps me honest about where I am failing. If I were scoring the current state of agentic AI out of ten right now, I would give it a generous three. We are early. The demos you see online look flawless, but when you try to put these things into production, they break. They get confused. They get stuck in endless loops.

I'll be honest with you. I do not have this completely figured out. When Fredrik told me I would be good at sales back when we met at that PR agency in Stockholm, a job I took on commission only against the advice of literally everyone around me, I thought I knew exactly how my career would pan out. I was wrong. I thought we would only stay in Spain for two years when we moved the family here in 2018. I was wrong about that too. I have been wrong plenty of times since my days travelling around the world in my early thirties, mostly disconnected from everything.

So treat this as a bet, not a prediction. The transition away from seat-based SaaS is going to be messy and defensive. Software companies with massive valuations based on seat counts are going to fight this aggressively. They will try to maintain the old model by charging for fake seats or wrapping basic API calls in expensive subscriptions. But eventually, the market will realise that paying for human interfaces when there are no humans makes zero sense. We are moving from a world where we bought tools to a world where we hire digital workers. I just don't know exactly what the employment contract for a line of code looks like yet.