As anyone working closely with AI will know, conversations around autonomous digital workers outright replacing human employees are rife right now. And with media headlines like “Microsoft AI chief gives it 18 months—for all white-collar work to be automated by AI” popping up almost weekly, it’s understandable.
Meanwhile, businesses worldwide are continuing to pour money into AI initiatives In fact, worldwide spending on AI is forecast to total $2.52 trillion in 2026, a 44% increase year-over-year, according to Gartner, while workplace AI agent use has almost doubled, rising to 59% from 31% a year earlier, according to recent research by Stack Overflow.
With statistics and headlines like these, it’s hard to dispute that AI will very likely automate a significant share of entry-level knowledge work within time. But for me, there is a much bigger and longer-term shift at play, and the conversation as it stands is far too narrow.
In fact, most of the industry has become so obsessed with AI agents and is so laser-focused on whether AI will replace employees that organisations have largely become too distracted to notice the impact something in an entirely different AI category is already starting to have.
Enter AI applications
Before delving into how AI applications are changing things, it’s worth distinguishing them from the two categories we’re seeing take the lion’s share of today’s AI conversations: AI assistants and AI agents.
Let’s think of AI assistants as category one; tools like Microsoft 365 Copilot or ChatGPT. They all help people complete work more efficiently – whether that’s summarising information or drafting content and answering questions – but the human is in control. So they augment the user rather than act independently.
Next come AI agents, which perform a task – or several – on your behalf. They run behind the scenes and work autonomously to achieve a specific goal, deciding how to complete the work using the tools and systems available to them. As mentioned, much of the industry’s AI obsession is centred on AI agents, and we’ve already seen big players like Microsoft pushing a “human-agent teams” vision, where AI acts as a digital teammate.
I think that obsession is somewhat misplaced. Not because AI agents don’t create value – they do – but because they’re solving a different problem from the one most organisations actually have.
Which leads us onto a third, very separate category: AI applications. An AI application is not simply a workflow, nor is it an agent. It’s a highly specialised intelligence layer designed to solve a specific business problem. Instead of asking a user to navigate software, interpret information and arrive at a conclusion, the application delivers the conclusion directly. In other words, it performs a full “function” you could say, providing its own interface and eliminating the need for you to use traditional software by understanding your intent and delivering the business outcome.
It’s important to note here, however, that when it comes to AI applications versus AI agents, one does not exclude the other. In fact, I think agents will scale significantly. But as mentioned, agents are optimised for execution, and most organisations aren’t suffering from a lack of execution or action, but rather a lack of clarity and better decisions. AI applications essentially address that bottleneck more directly by delivering understanding and outcomes, and not just carrying out tasks.
Removing the need for traditional software interfaces
The above is why I believe AI applications will achieve widespread adoption faster than agents, and why they offer a much bigger near-term opportunity. But the other interesting point here is what effect this will have on traditional software applications.
Most business software we use today exists to help users find answers, navigate interfaces and move information between systems. If AI can change that – so instead of opening applications, searching menus and completing workflows, users interact with systems that generate the answer, outcome or action directly – then the question also changes, from “what jobs can AI replace?” to “what software do humans still need to interact with directly?”
Because if AI can understand intent, execute workflows and deliver outcomes, the traditional application layer starts to look very different.
Even in that world though, humans will still need to understand what is going on in their business and want to be able to monitor that. It’s not that dashboards disappear entirely, but rather that their role changes. The real question becomes what shape and format this takes.
Some believe notifications will become the primary interface layer, others expect more consolidated, automated reporting. In practice, “chat with your data” is likely to be the next major development, and we’ve already started to see this emerge over the last year or so. In other words, the need for information doesn’t go away – but the way it is surfaced moves further away from traditional dashboards, reports and software interfaces, and towards more direct, AI-driven explanations of what is happening and what needs to happen next.
The future of AI applications
With the forward-leaning companies I’ve worked with, there’s a clear line of thinking emerging – they’re either increasingly looking for AI applications rather than purely AI agents – or looking for a combination of both. And it’s this route that I think will be the next natural step. Agents will not be enough to replace the current operating model, intelligent applications will be the replacement of functions, or cross functions.
Because of this, I believe we will very quickly start to see a new wave of AI applications in sectors like sales and marketing, and finance. A sales AI application, for example, would work by continuously analysing CRM data, email threads, meeting transcripts, and product usage to tell a sales leader exactly which deals are at risk and what actions are most likely to close them – without the leader ever opening a pipeline report. A finance AI application, on the other hand, might monitor cost structures, revenue trends, and market signals and explain why margins declined this quarter and whether the pattern is likely to continue.
In both cases, the user is not navigating software, but receiving intelligence. I also see this extending beyond traditional business functions and into product experiences themselves, perhaps even moreso in consumer-facing sectors like gaming or casino-type environments.
But no matter which path wins out, one thing is increasingly hard to ignore, and that’s that the real disruption is not just in jobs or workflows anymore, but in the very idea of software as we know it.
