AI Copilot — AI Glossary

Reference details

Related terms

• AI Coding Assistant
• AI Content Generation
• Tool-Using Agent

How it works

1. The assistant reads the state of the surface it lives in — the open document, the selected code, the record on screen — rather than starting from a blank prompt.
2. That context is assembled per request, which is why two copilots on the same model produce visibly different results.
3. Suggestions are rendered in place as reviewable diffs or inline completions, so accepting is the user's action rather than the system's.
4. Permissions are inherited from the host application, so the assistant can only see and change what the user already could.
5. Interaction stays turn-by-turn, which is the defining constraint: the user remains in the loop on every change.

Example: Why in-product context beats a better prompt

Asked to summarise a deal, a general chat assistant needs the user to paste the account history, the open opportunities and the last three emails — several minutes of copying, and whatever they forget is simply absent. An in-product copilot reads those records directly because it runs where they live. Same model, same question; the difference is that one had the data and the other had a description of it. Context access, not prompt quality, is what the product is selling.

Commonly confused with

Used interchangeably with agent, and the difference is who holds the loop. A copilot suggests and the user accepts or rejects each change, so a wrong suggestion costs a rejection. An agent runs many steps autonomously and returns an outcome, so a wrong decision compounds before anyone sees it. Most products now ship both modes under one name, which means 'is it any good' has to be asked per mode. It is also not a chatbot: the defining feature is reading the host application's state, not the conversational interface.

When to use it

Reach for it when:

• The work happens inside an application whose state is the context the task needs.
• The user should review each change rather than receive a finished outcome.
• Permissions should follow the user's existing access rather than a separate grant.

Reach for something else when:

• Multi-step work with a machine-checkable goal, which agent mode handles better.
• The needed context lives outside the host application and cannot be read.
• As a substitute for understanding work the user is accountable for.

Continue your research

Use these connected ModelRefs sections to compare alternatives, inspect implementation paths, and review the evidence and governance boundaries relevant to AI Copilot — AI Glossary.

• Browse models
• Review providers
• Inspect benchmarks
• Explore workflows
• Learn AI concepts
• Read decision guides
• Review trust and methodology
• Start a decision path