I still spend the majority of my day sitting at my desk working. So when I look at what AI agents are starting to do, a lot of my interest is fairly practical: how much of that work could they take off my hands?
I’m excited about the complicated things these models can do. But I’m also interested in the stuff that simply takes up time. The tasks that keep someone at a desk when they could be making dinner for their family, going for a walk or doing something else with their day.
I probably could use more of the technology that way myself. The heat in Dubai isn’t doing much to encourage me out of the house, but I can certainly see the appeal.
A lot of the useful work is fairly ordinary
Think about having a presentation due tomorrow morning and receiving a report full of numbers that need to go into it. Previously, that might have meant spending the evening working through the report, checking the figures and putting the slides together yourself.
An agent could help work through the numbers, prepare the presentation and give you a summary of the points to cover. You still need to understand and check the work, especially if you’re going to present it. But a lot of the preparation could happen on your behalf.
That’s a meaningful improvement in efficiency. It doesn’t have to be some incredibly complicated task to make a difference to someone’s day.
OpenAI’s July 2025 introduction of ChatGPT agent already described the ability to carry out research, analyse information and produce editable slides and spreadsheets. What interests me as these products develop is how much easier it becomes to use those capabilities together.
If a system has enough context about the task and can work with the relevant tools, there’s less time spent moving information from one place to another. It can start to feel like something that works with you throughout the project, rather than a separate window you visit to ask a question.
I think, two or three years down the line, we’ll see much more work being done that way. Some of it will be complicated. A lot of it will be the ordinary work that takes up hours because someone still has to do it.
Making it useful is part of the progress
The quality of the model matters, of course. But making the technology easier to understand and use matters as well.
A business owner doesn’t necessarily want to spend the day figuring out how to connect a collection of tools. They want to know whether the system can help with the work that is already in front of them.
That’s why I pay attention to the products being built around the models. Can people bring in the information they need? Can the system remember the relevant context? Can it do something useful with that information and give back work they can actually use?
I think that’s where a lot of the next stage of adoption comes from. The technology becomes easier to fit into an ordinary working day, and more people start to understand what an agent is capable of doing.
It still takes time for a new way of working to become normal. People have habits, and there’s a learning curve. But once something becomes useful enough that people rely on it, adoption can move quickly.
I’m also not assuming that the company with the best product today keeps that position forever. The balance between the leading AI companies is going to move around. The useful question is what the tools can do for the task in front of you.
I think this is worth paying attention to as a business owner or a professional. The opportunity to get more done in less time is becoming much more tangible. I’m looking forward to seeing more people use it—and to getting a bit more work off my own desk.
