How to give your agents context in Copilot Studio and Project Sophia

We have been using AI for a year or two, all of us. One way or another, either for tech stuff or maybe for day-to-day life. Even colleagues and family that are not in this Microsoft world sometimes say that "it gives nonsense answers". But what is the problem behind this?

There are basically two main challenges:

1. Prompts

2. Context

I will assume that if you are reading this you know how to prompt (but in case you don't, please be my guest and have fun: Master prompt engineer with Microsoft Designer)

When we talk about context, Microsoft has put together a very cool definition that I'm not even going to attempt to paraphrase:

Context is the collection of information a language model can use to guide its outputs—including instructions, memory, and external data. Managing this information effectively is called context engineering, and is key to shaping how an AI behaves and responds

Taken from: Context Engineering - Multi-agent Reference Architecture

In this blog I will share two different ways on how to manage context in agents to make your life easier, and of course build nicer solutions for yourself or your company :)

---

When you create your agent through Copilot Studio there is no straight indicator of "ADD YOUR CONTEXT HERE", so it's you who has to actually be mindful of adding it, and pay a lot of attention to the specific questions that Copilot Studio is asking you. This will be your call:

Adding this point will be crucial for the good creation of the agent. As it will determine either a good or a bad solution for your clients. So while you are creating, keep this in the back of your mind.

---

When it comes to Project Sophia, adding the context is even easier, when you log, there is a small place where you can add. Go to the URL: Project Sophia.

If you select the top right button, there are a couple of options. Select Manage your context.

A new window will open with the following fields, complete them as your case require. There is a 2000 characters limit, so use it wisely.

Based on the context I gave to Sophia, now I can prompt questions that are very specific to my company. As you can see below I'm asking questions about a specific program that the organization has called LIFT for Leadership initiatives.

How many employees are currently enrolled or in progress in LIFT leadership programmes?

Which departments have the highest participation in Leadership category training this fiscal year?

What’s the total CHF spend so far on Leadership programmes in the DACH region?

---

Long story short — yes, data is important. But so is context. We spend so much time talking about data—collecting it, cleaning it, storing it—that we forget the other half of the equation: the context that makes it useful.

Next time you find yourself in a data conversation, slide context into the mix. It’s just as important if you actually want your agents to do their job well.