How do I find out what is currently connected?

Each tool keeps the list in its own settings, usually under integrations or connected accounts, and administered plans often expose it centrally for the whole organisation. Reading it is a five-minute job that almost nobody does, and the common reaction is recognising two or three entries nobody remembers approving and no longer needs.

Does connecting more make a tool better?

It makes it more capable and more consequential in the same motion. Every entry adds something it can do and something it can reach when a step goes wrong, so the useful test is whether the task actually needed that reach or whether a narrower grant would have done. Most connections are wider than the job required.

What should we review, and how often?

The list itself, on whatever rhythm you already review access, since this is the same question in a newer place. Look for entries whose original reason has passed, entries granted by people who have left, and combinations where reading outside material sits alongside the ability to send or publish. That combination deserves more attention than any single entry.

Glossary

Connected apps

Connected apps are the other products a tool has been granted access to, and that list decides both what the tool can do for you and how far your material now travels.

In plain terms

The other things you have let it into: your mail, your files, your calendar, the place your team talks. Each one was a sensible yes on its own, granted in a second because it made something work. Nobody ever sits down and approves the whole list at once, which is why almost no one can say what theirs currently contains.

Why it matters

Because the list is a permission decision that never gets made as one. Each connection is approved on its own merits in the moment it unblocks something, and the reach that results is the sum of those moments rather than anything anybody chose. What a tool can reach is what it can act on, and it is also what leaves with it.

How it works

Access is granted per connection and usually broader than the task needs, because that is what the connecting screen offers. Reading one folder often means being handed the whole drive, and sending one message often means the mailbox, since the permission scopes on offer are the ones the connected product defined rather than the ones your task requires. The gap between what was needed and what was granted is where most of the surprise lives.

The list is cumulative and nothing prunes it. Connections are added when something needs unblocking and are almost never removed when that need passes, so reach only grows, and it survives the departure of whoever set it up. Reviewing the list occasionally is the entire maintenance practice, and it is the one nobody schedules.

What is connected decides what can go wrong, not just what can go right. A tool that reads material from outside can meet instructions hidden in that material, and whether that matters depends entirely on what else is on the list: reaching a document store and a way to send mail is a different proposition from reaching a document store alone. The combination is the risk, rather than any single entry.

Connection reach is now a real difference between products rather than a footnote. Some tools live inside one workspace and see everything in it by default, some connect outward to a long list of separate services, and some run only where a browser is open, so the same task lands very differently depending on which shape you picked. Establishing the shape before the pilot avoids discovering it afterwards.

How a connection list actually gets made

Drawing this as a sequence rather than as a settings screen is what makes the shape visible, because the settings screen shows a list and the list looks like a decision. It was not one. Every entry on it was created at a moment when somebody was trying to finish something else, presented with a choice framed as allow or cancel, where cancel meant the work stopped. That is not a permission review; it is a blocker with a button on it. The consequence is a reach profile that no individual would have approved if shown the whole thing at once, assembled entirely from approvals each of which was reasonable. The remedy is unglamorous and effective: read the list on a rhythm, remove what no longer has a reason, and pay particular attention to the pairing of reading outside material with the ability to send, publish or pay, because that combination is the one where a mistake stops being contained.

Seen in the wild

A workspace assistant that answers from the pages and databases around it and searches outward into connected services as well.
Notion AI
An agentic product that works in the folders you choose and the applications you connect, extending further through plugins and sub-agents.
Claude Cowork
A platform where each agent is given the applications it needs, such as mail, calendar and a CRM, as an explicit setup step.
Lindy

Common misconceptions

People assume

Connecting something is a convenience decision.

In fact

It is an access decision wearing convenience clothing, and it is usually made in a second by whoever hit the blocker. The scope granted is generally wider than the task needed because the connected product defined the options, so the honest description is that reach was accepted rather than chosen.

People assume

The list reflects what we currently use.

In fact

It reflects everything anyone ever needed, including people who have since left and projects that ended. Nothing removes an entry when its reason expires, so the list only grows, and the gap between it and current use is invisible until somebody deliberately reads it.

Telling them apart

Connected apps vs Connector

Connected apps

The list of what a tool has been let into.

Connector

The mechanism that makes one such link possible.

One is the inventory and the other is the plumbing, so a product can offer hundreds of the second while your own list of the first stays short.

Questions

How do I find out what is currently connected?: Each tool keeps the list in its own settings, usually under integrations or connected accounts, and administered plans often expose it centrally for the whole organisation. Reading it is a five-minute job that almost nobody does, and the common reaction is recognising two or three entries nobody remembers approving and no longer needs.
Does connecting more make a tool better?: It makes it more capable and more consequential in the same motion. Every entry adds something it can do and something it can reach when a step goes wrong, so the useful test is whether the task actually needed that reach or whether a narrower grant would have done. Most connections are wider than the job required.
What should we review, and how often?: The list itself, on whatever rhythm you already review access, since this is the same question in a newer place. Look for entries whose original reason has passed, entries granted by people who have left, and combinations where reading outside material sits alongside the ability to send or publish. That combination deserves more attention than any single entry.

Key takeaways

The connection list is a permission decision nobody ever makes all at once.
Granted scope is usually wider than the task, because the options were not yours.
Nothing prunes the list, so reach only accumulates and outlives its reason.
Risk lives in the combination of entries, not in any single one.

Tools that use this

Notion AI
Answers from the workspace and searches outward into connected services.
Claude Cowork
Works in chosen folders and connected applications, extended by plugins.
Lindy
Each agent is given the applications it needs as an explicit setup step.