Artificial Intelligence has moved quickly from science fiction into everyday life and if you’re using Microsoft Office, one of the most visible examples is Copilot. It’s suggesting it can improve you email, summarize your meeting or align you Powerpoint slides.
Copilot seems to be everywhere and is becoming a digital assistant for both office workers and everyday users. But not everyone is happy with Copilot, calling it AI slop and preferring other AI tools like Chatgpt, Gemini or Claude. Spoiler: if you use Copilot your actually using both of them. So, what exactly is Copilot and how does it work?
At its heart, Copilot is not the Artificial Intelligence itself. Instead, it acts as a bridge between people, powerful AI models known as Large Language Models (LLMs) and the information found inside organizations.
An LLM can ‘understand’ language, answer questions write text and perform complex reasoning. Think of it as a remarkably knowledgeable conversational engine. However, despite its impressive abilities, an LLM does not automatically know anything about Topigs Norsvin, our company policies or feed manuals, your current projects or the meetings you had regarding those projects. This is where Copilot comes in. Rather than simply sending a user’s question to a LLM, Copilot gathers relevant information from the user’s environment and provides it to the model. This process is known as grounding.
Imagine asking, "What decisions were made during yesterday's project meeting?" A standard AI model would have no way of knowing. Copilot, however, can search authorized sources in our Topigs Norsvin environment, such as meeting notes, emails, Teams conversations, SharePoint documents, or business applications. It then provides that information to the AI model, allowing the response to be based on actual company data rather than general knowledge. This grounding process is one of the reasons why AI is becoming useful is business settings. The AI is no longer working in isolation; it is working with context.
Copilot also becomes aware of its surroundings through the user’s identity and permissions. It understands who is asking the question, what your position is in the company and only retrieves information that person is already allowed to access. It can even access business applications like Dynamics or Anaplan and combine that information with information it retrieved from Pigbase. Copilot is able to answer questions using information stored across many different systems rather than relying on a single source.
As impressive as AI may seem, Copilot is only as good as the information it can access. Providing it with the wrong information can lead to digital inbreeding. Organizations sometimes assume that implementing AI will automatically solve information management problems. In reality, the opposite is often true: AI quickly exposes weaknesses in data quality. If documents are poorly organized, outdated or stored without context, Copilot may struggle to determine which information is trustworthy and relevant. The result can be incomplete, inconsistent, misleading and wrong answers.
I hope I made clear that Copilot is not just an AI chatbot. It acts as the orchestrator that combines information about you and your function in Topigs Norsvin with relevant data it gathers from Outlook Onedrive, Sharepoint and business applications, feeds that to one or more language models and presents you with the results.