AI copilot

Because the explaining is the expensive part and this removes it. Describing a spreadsheet to something that cannot see it takes longer than the edit, so the same capability delivered in place gets used many times a day where the same capability in a separate window gets used occasionally, and adoption follows placement far more reliably than it follows quality.

It begins with access to what you are working on, which is the whole of the difference. The document, the sheet, the mailbox or the repository is already there, so a request can be short and specific rather than a paragraph of setup, and short specific requests are the ones people actually make.

The vendor chooses what it can do, and that boundary is firmer than it looks. Where a separate assistant will attempt anything you describe, one built into a tool does the things its maker wired up, so it may be excellent at rewriting a paragraph and unable to help with the thing next to it.

It usually inherits the permissions already in place rather than establishing its own. Whatever the product decided you may see is what it works from, which is why this arrangement passes security review more easily than a separate tool asking to connect to everything, and why the answers it gives differ between colleagues.

Quality varies enormously behind a similar-looking button, and the phrase itself tells you nothing. Some are deeply wired into the product's own data and actions; others are a general assistant in a sidebar with the current file pasted in, which is a different thing entirely wearing the same label.

The lock-in is real and worth naming. Help that only exists inside one product becomes a reason to stay with that product, so the convenience is genuine and the strategic cost is paid quietly over several years rather than at the point of purchase.

• A coding assistant embedded across every major editor, wired into the flow of repositories, branches and pull requests rather than sitting beside it.

  GitHub Copilot

• AI threaded through an office suite, working inside the documents, mail and meetings on an organisation's own tenant under the permissions already there.

  Microsoft Copilot

• A workspace layer that drafts and edits in place and answers from the pages and databases the team already keeps.

  Notion AI

How do I tell a real one from a chat box in a sidebar?

Ask it something that requires knowing your data rather than reading your screen. A genuinely wired feature can answer about records, history or state it was given access to; a general assistant with your file pasted in can only work with what is visible, and the difference shows immediately.

Is it better than a general assistant?

For work inside that tool, usually yes, because the setup cost is gone. For anything crossing tools, or anything its maker did not anticipate, it is worse or simply unavailable. Most teams end up with both, which is a reasonable outcome rather than a failure to decide.

Does it see everything in our account?

It generally works within the permissions already governing you, which is why this arrangement clears security review more easily than a separate tool asking for broad access. The practical consequence is that two colleagues can ask the same question and get different answers, which is correct behaviour rather than inconsistency.

What should we actually check before rolling one out?

Whether it does the specific things your people do repeatedly, rather than whether it demonstrates well. A convincing demonstration establishes that the capability exists somewhere in the product, and daily value comes from a handful of repeated actions, so testing those directly settles the question in an afternoon.

• Placement, not capability, is what makes this a different thing.
• Removing the explaining is what turns occasional use into daily use.
• The vendor's wiring sets a firm boundary the label does not reveal.
• It usually inherits existing permissions, which helps it clear review.
• Convenience now, switching cost later; both are real.

• 

  GitHub Copilot

  Embedded across editors and wired into the repository workflow.

• 

  Microsoft Copilot

  Inside the office suite, on the tenant's own files and permissions.

• 

  Notion AI

  Drafts in place and answers from the team's own pages.