AI agents
So far, many of us use AI for basic input/output tasks. You input some text, image or video, and it gives you a new text, image or video back. You will then take that new AI-generated content and use it, say, in a document, email, design or film, for example. But what if AI could take that newly generated content and do something with it by itself? That is where AI agents come in.
AI agents essentially stitch a series of AI tools together. Instead of giving it a task, you give it a goal. Say your job is to solve customer X number of customer service queries per week and generate a report on your findings. (This is a boring example but easy for many to get their head around. Consider similar mundane tasks you might do in your job while reading this). Currently, you might use AI to help you collate all your customer queries, then use another AI tool to help find the solutions from your data bank, another tool to help write professional responses, then another to send all your responses out on mass, and then you gather all your data together to manually create a written report. With AI agents, you would teach it all the steps needed to perform this task, and it would automate all of it for you. It will find all the queries for you, resolve all their issues themselves from a database of responses, email them all back or chat with them via a chatbot, then gather all the data and create a presentation for you. These arduous weekly tasks suddenly become much easier, freeing you up to do other, more interesting work.
You will, of course, have the option to review each stage, but once people trust the AI is doing a good job, it is likely you will reduce how much you check-in. It is akin to having an intern doing the work. You will need to teach it and correct it in the same way; it will improve over time by learning. That is the essence of AI - it learns and improves over time, much like a human. Which is what is so different about it to conventional software tools you might have used before.
For those of you who know AI well, you will be familiar with agents, but for those of you who are not and don’t understand them, this will sound daunting. Microsoft Co-Pilot is launching AI Agents in 2025, making them much more accessible to the many, which will increase their use over the coming years. Where before you would need developers to set up this kind of automation, Microsoft Co-Pilot Studio claims anyone can do it, and no coding is required.
The above customer service example is a boring but simple example that many people can get their head around, but you will be able to set it up with an increasing number of tasks from accounting to designing. The one thing you will definitely hear as its use grows in the media is agents get things wrong! This will put many off, but the mistakes will be no different to errors an intern or new employee might make because it is AI making the mistakes; the media and people as a whole will over-emphasise their errors. Trust is likely to take time for people to adjust, but those companies using AI well will flourish, and those who don’t will fall behind as their AI-using competitors enjoy reduced costs and increased productivity.
While there are a few existing agentic tools and software out there already and other big ones coming out in 2025, Microsoft Co-Pilot is already used by millions so is likely to be the first time the masses use it.
An example of Microsoft Co-Pilot AI agents in use
