Connecting AI to your real life

Today’s topic is connectors.

For the last couple of weeks I have been writing about ways to make AI actually useful instead of just impressive. A lot of that has been about memory: giving a conversation enough standing context that opening it does not feel like starting over from a blank stranger every time. Memory keeps your past conversations in the front of your mind, readily accessible when you bring them up again.

Connectors are the next step, and they are a different kind of useful. Memory holds onto what you have already said. A connector reaches into the other parts of your life, the ones that do not live inside the chat at all: your mail, your calendar, your files. It is the difference between an assistant that remembers what you told it and one that can go and look.

That is the whole idea. Most people meet AI as a blank box. You type, it answers, and everything it knows is whatever you managed to put in the message. Connectors are how you stop being the only bridge between the assistant and the stuff your life actually runs on.

What a connector actually is

A connector is a controlled link between the assistant and a service you already use, switched on once and switched off any time. With it on, the assistant can reach into that service instead of knowing only what you typed into the chat. Your inbox. Your calendar. Your files.

The good news is that they are usually secure and relatively easy to turn on. You are not handing over a password or copying anything into a sketchy box. You authorize one specific account through that service’s own sign-in, the permission is scoped to what the connector needs, and you can revoke it in a click. The hard part is mostly deciding to do it, not the doing.

These have gone by a few names. Plugins, extensions, connectors, and lately just apps. The label keeps changing and the marketing keeps rebranding. The goal has never moved: let the assistant reach something real in your life instead of only the words in the chat box.

There are two kinds, and the difference matters more than any other detail:

• Read-only connectors let the assistant find and summarize. It can pull the email thread you described, or tell you what is on Thursday. It cannot change anything.
• Action connectors let it do something: draft a reply, put an event on the calendar, edit a document. More useful, and more to think about, because now the tool is acting and not just looking.

Most of what I do is the read-only kind, and that is where I would start.

The different assistants each have their own version, under their own name. Claude calls them connectors. ChatGPT calls them apps. Gemini calls them connected apps, and because it lives inside Google, it can reach Gmail, Calendar, and Drive with almost no setup. What you get is roughly the same core set on each:

Connector	Claude · “connectors”	ChatGPT · “apps”	Gemini · “connected apps”
Email	Read and draft; will not send for you	Read and draft; asks before it sends	Read and summarize
Calendar	Read, create, edit, delete	Read; find and summarize	Read, create, edit
Files	Google Drive: read and save back	Drive, OneDrive, Dropbox, Box: read	Drive and Docs: read

Notice the pattern down that table. Reading is automatic. Anything that sends, books, or changes something asks you first. That is not an accident, and it is most of the reason these are safe to turn on.

One plain note on how this works underneath: a lot of these connections now run on a shared standard, which you will see called MCP, the Model Context Protocol. You do not need to know the term to use any of this. It just means the list of things you can connect keeps growing, so the tool you wish you could plug in probably already works, or will soon.

How you actually use them

The simplest use is retrieval. You describe what you want the way you remember it, and the connector goes and gets it. The skill that makes this work is dropping the keyword guessing game.

What you’d type in the search bar	What you can ask instead
quote … estimate … bid (guessing which word you used)	“Find the contractor’s email with the price for the rental work, sometime in the last couple of weeks.”
renewal (then scrolling)	“What’s the renewal date in the thread with my tenant?”
(you cannot remember the sender at all)	“The message about the leak. Who sent it, and when?”

The part that surprised me is what happens when two connectors are on at once. The value compounds. With the inbox and the calendar both connected, I can ask one question that neither could answer alone: “Given what is on my calendar this week and the threads sitting unanswered in my inbox, what am I about to drop?” That is not a search. That is the assistant doing the cross-checking I used to do in my own head on a Sunday night.

The full map of what you can connect

This is where it gets bigger than email. The directories now run past two hundred entries and grow every month, so this is not the whole list, but it is a real sense of the range. Most of these exist in some form across Claude, ChatGPT, and Gemini.

Part of your life	What you can connect
Mail and calendar	Gmail, Google Calendar, Outlook
Files and storage	Google Drive, OneDrive, Dropbox, Box
Notes and documents	Google Docs, Google Keep, Notion
Tasks and lists	Google Tasks, Todoist, Trello
Messaging	Slack, WhatsApp, Microsoft Teams
Design and creative	Canva, Figma, Adobe
Errands and everyday life	Uber, Uber Eats, Instacart, Resy, Booking.com, AllTrails, Tripadvisor, StubHub
Travel and local	Google Maps, Google Flights, Google Hotels, Expedia, Zillow
Money and admin	TurboTax, Credit Karma
Health and reference	Apple Health, PubMed
Music and media	Spotify, YouTube, YouTube Music, Audible

And if the thing you want is not on a list yet, the newer tools let you add your own connection to almost any service that offers one. The point of the long list is not that you should turn them all on. It is that the assistant can almost certainly reach the part of your life you keep doing by hand.

The ones I actually use

I run a small set, and they fall into two camps: the ones that save me time, and the ones that are just fun. Both earn their place.

For getting time back

• Gmail, Google Drive, and Google Calendar, on every assistant I use. These are the workhorses, connected in Claude, ChatGPT, and Gemini alike. Find the email by what I meant. Point at a document, the lease or the long policy, and ask it questions. See the week and catch the conflict I was about to miss. If you only ever turn on three things, make it these three.
• A personal-finance connector in ChatGPT. This is the one that should give you pause, because it is exactly the kind of data I said last week to be careful with. The reason I use it anyway is how it connects. ChatGPT links through Plaid, the same service a lot of banking and budgeting apps already use to connect your accounts. You sign in through your own institution, Plaid passes along read access, and your actual password, along with the ability to move money or change anything, never reaches the app. What I get back is a secure, real-time picture of where things stand. That is a different thing from pasting account numbers into a chat box. The connection is scoped and easy to revoke, and I decided the trade was worth it. You might decide it is not, and that is exactly the right way to make the call.
• In Claude, a set of developer tools for the software side of what I build. Most readers will never need these, so I will not belabor them. Same idea, pointed at code instead of the rest of life.

For fun

• Apple Music in ChatGPT. Building playlists, turning up music I would not have found on my own. Not everything has to save time. Some of it just makes the tool feel like yours instead of one more place to get chores done.

That mix is the point. A few connections that quietly hand back the morning, and one that just makes me happy. Both are good reasons to plug something in.

Turning them on without regretting it

A connector is a real permission. You are giving an assistant access to something that is yours. That is worth doing deliberately, especially after the last couple of posts about being careful what you connect and what you keep out. Three things I check before I turn one on:

• Connect one account, not all of them. Start with the inbox. You do not have to wire up everything to get the benefit of wiring up one thing.
• Know whether it is read-only or can act. For most of what you want, read-only is plenty. If a connector asks to send mail or change files on your behalf, that is a bigger decision than finding a thread, and worth a second of thought.
• Know where the off switch is. It is one click to revoke, in the same settings where you turned it on. Knowing that ahead of time is what makes turning it on a low-stakes choice instead of a leap.

What they don’t do

A connector finds and summarizes. It does not decide.

It will not tell me whether to reply to the contractor today or wait for the second quote. It will not judge that the renewal email matters more than the leak. It reads, scoped and read-only, and it hands the thread back. The deciding, the replying, the “is this urgent or can it wait” call: all mine, made on purpose, not made for me by whatever happened to be at the top of the unread pile.

That is the whole point, actually. I did not connect the inbox so it could run my morning. I connected it so it would stop running my morning.

The inbox stops being the thing that decides your day the moment you can ask it for one thing and get one thing back.

This is what a little breathing room looks like.