A New Hope for SaaS: Embracing AI Agents

Satya Nadella has claimed that business applications will collapse. But if that’s true, what will replace them?

I think the notion that business applications exist… that’s probably where they will all collapse–in the agent era. –Satya Nadella

Business software will obviously still exist, but I believe it will be very different. A lot of the software programs that are out there now are “forms over data,” and we don’t need them any more. Instead, we will have AI agents take their place.

Forms over data

In college I worked as a programmer in the IT department of an electronics supplier. My job was to implement changes to the forms used by the salespeople as they made calls or the reports used by managers. I could write another article on using C without libraries in a text editor on VMS, or about how we convinced management to let us write 5-line SQL queries instead of 100 lines of C. But the point that I want to make is this is what most of the department was doing. If you weren’t one of the lucky few working on the customer website, you were creating forms, business logic, and reports.

This category covers a lot of software! I’m typing this article data into my website CMS, and you are reading it through the same. Also open in my browser are tabs for Quickbooks, charitable donation tracking software, and a Github issue report. All three of these are more complicated versions of forms (input fields) over data (the system of record). Excel, one of my favorite programs, is itself a form over the data it contains in spreadsheet cells. Even communication software like Outlook fits into this category. An email is a form, the data is sent to another person, and they see it in a report of their most recent emails.

This doesn’t only cover business software by the way. If you’re reading this on LinkedIn, it’s the same story. Although the algorithm that decides the display order is more complicated than 100 lines of C code, it fundamentally is still a display of the data in the database.

If the software is mostly about helping you enter some properties or content, doing some business logic, and then displaying that content, it’s a form over data. It’s not long for this world. The future of Software as a Service is with dynamic and intelligent AI programs serving as our agents.

AI Agents

Instead, software will increasingly become agentic. I’ve written about agents in the technical sense before, but here I’ll use the more common meaning. The best way to think about an AI agent is to compare with real life agents, like talent agents or real estate agents. A human agent is an expert in some domain, working proactively on your behalf. Specifically, we want our agents to be:

• Proactive: they begin work on a task before we ask them to.
• Personalized: they understand our preferences and job to be done.
• Capable: they do useful tasks for us.
• Communicative: they work with others on our behalf.

Software agents are the same way, no matter how they are implemented technically. Until recently, only human agents could hope to be proactive, personalized, capable, and communicative. These features of AI agents are within reach, and they are even being used in some scenarios now.

Writing data

One of the first agentic scenarios with these qualities was in Microsoft 365 Copilot for Sales. It’s simple: if a salesperson gets an email from a new person, they see a screen like this in Outlook:

This is a classic form over data. This agent proactively calculated these fields when the email arrived. Other features of the same product are very personalized and communicative as well. This version of the feature is only partially capable, as it still shows the form. As trust of AI increases, expect to see the form disappear and the database get updated anyway. As AI agents become common, all the business data you see in any format will be accurately stored for future use.

Business logic

Software will continue to use pre-defined business logic to properly store and retrieve data. This backend code enforces the rules for valid email addresses and which users are authorized to view or edit records, among other things.

Algorithmic feeds have been the trend of business logic over the last several years. Social media algorithms prioritize items that we are more likely to click on. This is happening in business as well, although not as quickly. Perhaps you have noticed that sometimes the next document you want to open is in the recommended list. As AI gets more personalized, you will increasingly get the information you need, exactly when you need it.

Other business logic is performed by humans. Pre-defined software isn’t able to make the nuanced decisions required for business operations. Another reason humans manually perform business logic today is that simply no one has paid for an IT project to automate it yet. AI agents can handle these decisions without requiring multi-month software projects.

Reading data

“Forms over data” applications show you the data record and let you decide what to do with it. Agents can instead tell you what to do with it. Increasingly, AI can complete several steps of the process for you. If the agent is aware of the code you just checked in to a project, it can find the task in Azure DevOps, suggest a comment to add, and let you approve closing the bug out.

The user will become a reviewer and approver of business operations, just as if they had hired someone to do them. We need to rethink our current software interfaces. AI product managers talk about this as “managing a fleet of agents,” among other metaphors. I don’t think anyone has nailed this interface yet, but it is the direction software is going.

Remaining software

When AI agents manage work information, execute business logic, and show finalized tasks for approval, what type of software will still be needed? First, these role-specific agents need to be developed (I’ll describe best practices in a future article). We also still need the software to support creative, relational, and novel work.

• Novel: not every task is similar to what was in the training data. These cannot be easily automated.
• Creative: can you imagine trying to operate Adobe Illustrator via text commands? Even in the realm of words, human creativity has value because it is human.
• Relational: video meetings cannot go away, nor can any other software that supports relational communication. AI can assist—not replace—human connection.

When I think about the future of business software, agents are handling the menial labor of repetitive business operations for me. I’ll just review and edit the results. The remaining software supports my novel, creative, and relational work. That’s the future work that I want to have. Are you ready to embrace the coming AI agents?