How two simple Copilot agents changed the way I work — and why the hardest part wasn't the technology.
I want to be upfront about something before you read any further. I am not an AI early adopter. I am not going to tell you about complex automation pipelines or show you under the hood of a large language model. There are plenty of people further along this road than me.
But here is what I do know: the majority of people are not where I am yet. And where I am is not that far. That is kind of the point.
This is about how I used Microsoft Copilot to build two simple agents that genuinely changed my working week. What I did, how I did it, and what it actually saved me.
The Problem
My organisation had a leadership directive to start making use of AI tools. Microsoft Copilot was available across the business and everyone had been given access to Copilot Studio. I had a clear problem in front of me, so I decided to use it.
The First Agent
Our team had collaboratively authored a document that walked the wider organisation through a set of processes they needed to follow. It was thorough. It included workflow descriptions and flowcharts. It covered the steps, the decisions, the right routes to take. It was, by any reasonable measure, a useful document.
Most people did not read it.
Whether that was down to time, inclination, or the fact that it is always easier to ask someone than to go looking for an answer, the result was the same: I was fielding the same questions repeatedly. Questions that were already answered, in writing, in a document they had access to.
So I built a Copilot agent. I took the workflows and processes from that document and translated them into a set of instructions and questions the agent could use to guide people in the right direction. I published it through Copilot Studio — no IT approval required, just a URL shared with colleagues who had standard Office 365 logins — and pointed people to it when the questions came in.
The technical lift was minimal. Writing the agent was not the hard part. The harder part, honestly, was identifying the right need and working out how to compartmentalise it into something a single agent could handle well. Once I had that clarity, building it was straightforward.
The impact was immediate. Those repeated questions stopped landing in my inbox. I got that time back. And with that time, I could be more forward-thinking and more strategic — which is what I should actually be doing.
The Second Agent
The second agent came from a different frustration. Documents were coming through with inconsistent formatting, gaps in content, unclear language, and occasional ambiguities that had the potential to create real problems further down the line. Reviewing and QA-checking these was falling to me, and I was giving the same feedback, over and over again.
I built an agent to do the initial sweep. Its job was to check that the right sections were present, that formatting was consistent, that the tone was clear and professional, and that there were no obvious ambiguities in the language. It flagged what needed attention before the document got to me.
I was always going to be the second pair of eyes. That was never in question. The agent did not replace my judgement — it cleared the path for it. What it removed was the bulk of the repetitive, mechanical review work. By the time a document reached me, the obvious issues had already been surfaced. My review became faster, sharper, and more focused on the things that actually required a human call.
The way I think about it: AI does the grunt work, I do the judgement. That split feels right to me, and it is sustainable.
What This Actually Cost Me
Building both agents required no coding, no IT involvement, and no specialist knowledge. Copilot Studio is a straightforward interface and the access was already there. What it required was the willingness to stop and think clearly about the problem before reaching for the tool.
Across both agents, I estimate I was saving roughly half a day per week. That is not nothing. Over the course of a month, that is time I was spending on work that actually moved things forward rather than answering the same questions on repeat.
The Bit Most People Miss
When people talk about AI productivity gains, they tend to focus on the technology. The tool, the model, the capability. What I have found is that the real work is upstream of all that.
It is not hard to create an agent. What is harder is finding the right need — the specific, recurring friction in your work — and then being disciplined enough to scope it down to something a single agent can do well. That thinking is the skill. The build is almost incidental once you have it.
If you are wondering where to start with AI tools in your own work, I would suggest starting there. Not with the technology. With the problem.
If any of this resonates with where you are, I would love to hear from you. Whether you are just starting to think about this or already have some ideas you want to pressure-test, get in touch through the contact form. That is the kind of conversation I am here for.