For the record
What happens to business software when AI can do more of the work?
When people ask me for startup tips, I keep coming back to one question: if AI takes over more of what SaaS products do today, what will customers still pay for?
I don’t think any category of software is guaranteed to be safe. But I do think it helps to ask where a business keeps the information it relies on, and who is responsible for changing it.
An AI can draft a page, suggest product categories, or work through a series of steps in an application. As it gets better at those tasks, I expect people to spend less time in the forms and dashboards we’ve built for them.
The business still needs to know which products it can sell in each market, who approved a discount, which policy was published, and whether an invoice was paid. Someone has to keep that information up to date, with a history that other people can check.
That’s why I’m interested in systems of record: the software a business relies on for its accepted version of those facts. An agent needs somewhere reliable to read from and somewhere to save changes so they’re available to everyone else.
Who gets to change the record?
Take a product description. An AI might write a good one from a spreadsheet, but that doesn’t tell us whether the product is in stock or whether someone has approved its claims for the German market. Before that description goes live, we need to know where the information came from and who is allowed to publish it.
Payments make the need for a history especially clear. A payment may be authorized, captured, refunded, or disputed. Finance needs to reconcile it. Support needs to explain it to a customer. If an agent starts a payment or a refund, those teams need a record they can check later.
When the conversation with the agent is over, the business still needs to know what changed and who approved it.
Take the CMS
I can imagine much of a content management system’s interface moving into a conversation. Ask for a campaign page, review a draft, then ask for versions for other channels. That could save a lot of time spent filling in fields and moving between screens.
Underneath that, the CMS still needs to track the content’s structure, language, publishing status, approvals, and version history. If someone asks what was on the site at 09:17 last Tuesday, there should be a way to find out. If an agent proposes an update, we need to distinguish that draft from the version approved for publication.
What do customers trust you to keep?
That’s the question I’d ask if I were building a SaaS business today. Adding AI might make a product easier to use, but it’s also worth asking why a customer would keep using it if an agent could do the same tasks elsewhere.
If customers rely on you to keep an accurate record, there’s plenty of useful work for AI around it: suggesting changes, sorting information, preparing reviews, and helping people understand what happened. The record and the rules for changing it still matter.
I’m especially interested in how this plays out in fintech and insurance, where businesses need to trace decisions and explain them later. I think there’s an opportunity for companies that can remove repetitive work while keeping approvals, access controls, and history easy to inspect.
That doesn’t guarantee the existing vendors will win. A new AI product could become the place customers keep their records. An established product could lose them because its data is hard to use or its workflows are painful.
My bet is that customers will still pay for software they can trust with the history of their business, even when an agent does much of the day-to-day work. That’s where I’d look for something worth building.
Robin Westerlundh
← back to the index