What is Microsoft Copilot?

Copilot is the umbrella name under which Microsoft has grafted generative AI capabilities (built partly on OpenAI's models, thanks to the partnership between the two companies) into its existing products: Word, Excel, Outlook and Teams inside Microsoft 365, GitHub Copilot for writing code, and a system-level Copilot built into Windows. The thesis is simple: bring AI where people already work, instead of asking them to open a new app. Copilot is not a single product but a family with different licenses and capabilities depending on context: the consumer version built into Windows, the Microsoft 365 version tied to existing enterprise licenses, and GitHub Copilot sold separately to developers. Each variant inherits the data and permissions of the environment it operates in, so evaluating one says nothing about the reliability of the others. This also means a company can roll out Copilot in one product while leaving it entirely off in another, a useful option for organizations wanting to pilot the lowest-risk surface before extending it further.

What it does

Inside Word and Outlook, Copilot drafts text, summarizes long emails and generates slides from a document. In Excel it helps with formulas and analysis. GitHub Copilot, the family's most mature product, suggests code line by line or entire functions inside the editor. The common thread is the absence of friction: no tool switch and no new interface to learn. The license, though, is its own line item: Microsoft 365 Copilot is a paid per-seat add-on, assigned person by person, so the friction moves from training to budget and to deciding who gets it.

How to evaluate it as a buyer

The critical point is not model quality, but permissions governance. Copilot for Microsoft 365 operates on data the user already has access to in SharePoint, OneDrive and Teams: if internal sharing permissions are broader than assumed (folders shared "company-wide" for convenience, never cleaned up), Copilot suddenly makes them queryable in natural language by anyone who already, in theory, has access. A serious Copilot rollout starts with a permissions audit, not with flipping on the license. Even with permissions sorted, the hardest part remains: getting the project past the first pilot into real production use, the subject of the post on why AI projects stay stuck in pilot mode.