As apps and end-to-end services decline in value, the individual result of a function becomes the product. The path from "need" to "result" has never been shorter. For the last 2+ years I've been telling clients to prepare a big shift in how software & services are designed and built. This one is really important, so stick with me to the end. I promise it will be worth it 🙂
User interfaces are built for humans who need help operating complex services. Agents don't need a helpful UI. They need data and the right to act.
And this changes how software is built, distributed and priced. Because the value of software then gets granular. Not at the platform level, not at the workflow level. At the level of the smallest verifiable result. A confident answer and a clean data point. A guaranteed action with a known outcome. The individual result of a function is the product now.
As your agent orchestrates your workflow, there are several small results in that flow you should not trust the orchestrating agent to execute. These small partial results all contribute to the output of the workflow and they need to be executed by specialized trusted vendors with domain expertise.
Example: You definitely want your accounting software to calculate your monthly revenue, not ChatGPT. But you may want ChatGPT to compare your monthly revenue with your Google Analytics data over the last 3 years.
So the question every software company has to answer in the next eighteen months is tough: what results can we verify, and what are those results worth to a customer when consumed outside the walls of our system?
AI agent toll gates: the new buzzword
There are several CRM SaaS companies that have started to adapt to this shift by adding "toll gates". Ensuring that AI agents can't roam freely inside their services. (As reported by The information)
ServiceNow announced "Action Fabric" at its analyst day in Las Vegas. It sits between their applications and any external agent that wants in. They meter the traffic and they price it. COO Amit Zavery was direct about the mechanics: Standard APIs still exist, but the deeper capabilities now sit behind a metered layer. JPMorgan's Mark Murphy called it a tax on customers who use outside agents to touch the data. I guess that's one way of seeing it…
Datadog took a different approach. Their MCP server caps agent activity at 5,000 daily requests, 50,000 monthly. Past that you slow down or stop.
SAP is more blunt. They've signaled that external agents may not be allowed inside their systems at all without explicit approval. The good old walled garden approach.
HubSpot and Workday are already metering. Workday's Aneel Bhusri has been open about the upside. He sees real money in charging for agent activity, and he's probably right.
But what may look like a pricing change is a deeper structural change. The old SaaS model scaled with the number of human seats. The new one scales with machine activity. That's a different business entirely, and the companies sitting on the most trusted data are the ones who get to write the rules.
The new question for software founders
The strategic question used to be: how do we make the experience so good that customers stay?
That question is kind of not needed any more (well, I think experience still matters, but the where and how the experience takes place has changed).
The more relevant question is harder to answer: What slice of the result can we verify, and how do we price it? How small can we make that a "buyable unit", and how confident can we be in it?
For the companies who own critical data this is an opportunity. It's time to move from a McDonalds meal to a Tapas menu. From per seat pricing to per action pricing. From owning a customer's screen time to owning a piece of every transaction that runs through their domain.
For the companies who built mostly on the interface, this is a lot harder. Their usp was the experience and big parts of that are soon gone.
What I'm telling clients
What software companies need to do now is unglamorous. Audit your stack and find the granular results you can actually stand behind. Build APIs and MCP endpoints around those results. Price the result, not the dashboard. Stop trying to be a workflow company in a world where agents own the workflow.
If you make data, package it for agents. If you make a function, expose it as a callable unit with a clear guarantee. If you make a dashboard, ask whether anyone is going to be looking at it in three years.
The "one serves all" interface era was a long, beautiful run. And I will miss it, because complex and glorious UI's were a big part of what I used to pour my heart and soul into. But they are getting extinct. UIs and pixels are still needed. But they are no longer the primary home for users of SaaS platforms they once were.
In the new "result era" trust is what makes a function buyable as a product. Anyone can compute a number and present the result. Vendors with domain expertise can compute it in a way that orchestrating agents can take action on the result because it is trusted.
The companies that figure that out early build the toll gates. The ones that figure it out late pay for passing through.
Got questions? Feel free to reach out, I'm happy to talk in depth about this.
All the best, Anders