Guide · Linking AI to your tools
AI that can see your work is a different tool entirely.
For the person who spends half of every AI session copy-pasting context in.
Out of the box, AI knows exactly what you paste into it. Connectors change the deal: link it to Drive or Microsoft 365, your email, your calendar, your CRM (hundreds of platforms now), and it stops asking you to copy things in and starts working with where your work already lives. The rule that goes with them is the second brain's: point, don't duplicate.
The idea
A connector is permission with a pipe attached. Instead of you fetching documents for the AI, the AI fetches them itself: your policy from Drive, this morning's email, next week's calendar. Always the current version, never the copy you pasted in March.
The permissions are the real teaching. Every connection exposes its abilities verb by verb, and each one can be allowed, need approval, or be blocked. Some are governed for you: Gmail can draft and cannot send. Others are blunter than you expect, and Drive is all files or none, which is exactly why the honest setting for reading is needs approval.
This is where AI stops being a private experiment and starts touching organisational data. Leaders: make sure your AI policy covers connectors. Everyone else: read what your policy says before you plug anything in.
Permission, with a pipe attached. Connect the systems your work already lives in, then set what it may do, one verb at a time. Draft, yes. Send, no. Read, ask me first.
Two ways to put AI to work. Copilot and Gemini put AI inside the tools you already use. Connectors bring those tools to your AI. Plenty of organisations sensibly run both.
How it works
- Check the policy firstThe AI policy template in the Hub has a connector clause for exactly this moment. If your policy predates connectors, it does not cover them.
- Browse, connect, authenticateConnectors live behind the plus button. Pick the platform, sign in on its own site, and it joins your connections list. The connecting is the easy part.
- Set the permissions, verb by verbAllowed, needs approval, blocked: for each ability, deliberately. If any data in Drive is off-limits to AI, reading needs approval. Drafting can be allowed while deleting stays blocked.
- Tell it which connector to useAsk for a graphic and an AI with Canva connected will happily code one from scratch instead. Sometimes it won't use your connectors unless you actually tell it to. Name the tool in the prompt.
- Point, don't duplicateThe second brain rule: if it already lives in Drive or SharePoint, connect to it there. Copies go stale; sources don't. This is also what makes agents and scheduled tasks possible: they need to reach the real systems.
The check
Four statements before you plug anything in. Score honestly: this is the check to run before the first connection.
Score each one Not true = 0, Partly true = 1, True = 2, then add them up.
- Our AI policy explicitly covers connectors and integrations.Not AI in general, connectors specifically: which systems may be linked to AI tools, and who approves it.
- We know where our sensitive data actually lives.Client records, HR, finance: named systems and folders, not “somewhere in Drive”.
- Access inside those systems is already scoped.People see what their role needs. Nothing important is on an everyone-can-view link from 2021.
- We could name every connection we've enabled, and switch one off today.Somebody owns the list. Unplugging is a decision, not an archaeology project.
Pasting context in by hand is slower and safer, and for now that's right. Start with the policy: the template's connector clause is a two-hour fix.
You have enough hygiene to start where the value is obvious: documents, reading set to needs-approval, nothing that writes. Prove it there before granting more.
Connect the systems your work actually lives in, keep the access list owned, and let your schedules and agents work with the real thing.