Picked from the software, topic, and level of “Agentic AI Without Coding Class #21: The Connector Concept: Linking AI to Gmail, Drive, Calendar, Sheets, and Notion”.
At most one email a week. Unfollow anytime with one click.
Transcript
The full narration of this video. The current sentence is highlighted while the video plays; click any sentence to jump there.
Show transcript
Introduction
Welcome to Session 5 of Agentic AI Without Coding. This session is called Connecting AI to Work Apps. In the first episode of this session, we look at the connector concept: how AI can read your own emails, files, schedules, tables, and notes, how that differs from copy-paste, and what decides whether this feature is available to you.
Step 1: From the internet to your own work data
In Session 4 we saw AI that can search the internet and act in the browser. But most of our work isn't on the public internet. Our work lives in our inbox, our folders, our calendar, and our spreadsheets. This is where connectors come in. A connector is a bridge between AI and your work apps. With your permission, the AI can read your emails, files, schedules, and tables, and then answer based on that data. There are two key words here: real data and permission. The AI's answers are no longer general guesses. They're based on what's actually in your inbox and your files. And all of this only happens as far as the permissions you give. Let's look at how this has grown. In Session 3, the AI assistant read files you uploaded, like SOPs, catalogs, and FAQs. That's what made the Sunset Coffee customer service assistant know the prices. In Session 4, the AI searched the public internet and websites on its own, for example to research competitor prices and fill in forms in the browser. In Session 5, the source moves to your own work data. The emails that arrived this morning, next week's schedule, and yesterday's sales table. This data changes every day, so uploading it once isn't enough. Picture Dewi's morning. She's the operations manager at Maju Bersama Co. To get ready for a meeting, she opens Gmail, then Drive, then Calendar, then the stock spreadsheet. Her data is spread across many apps. If she wants help from AI, Dewi has to select text, copy it, paste it into the chat, and repeat that for every source. Ten minutes gone just moving data around. And those copies go stale quickly. The stock number she pasted this morning may have changed by the afternoon. Connectors solve all three problems at once.
Step 2: How a connector works: ask, fetch, answer
A connector always works in four stages. First, you ask in everyday language, just like asking a coworker. Second, the AI works out what data it needs to answer, then picks the relevant connector. It's like an agent choosing a tool, which we covered in Session 1. Third, the connector fetches data from its app, for example searching your email or opening a file, and only as far as the permissions you gave. Fourth, the AI reads the result and answers. A good answer also names its sources: which email, which file. You don't need to name the connector, because the AI decides which data it needs. Let's see an example. Dewi types a simple question: I have a meeting with Sinar Paper Ltd. tomorrow, what do I need to prepare? She doesn't paste in any data. Notice the gray lines with the gear icon. That's the sign a connector is working. First, the AI searches Google Calendar and finds tomorrow's review meeting at ten. Then the AI searches Gmail for Sinar Paper, and finds three emails from this week. Because one of the emails talks about prices, the AI also opens the contract file in Google Drive. Nobody told it to open the contract. The AI decided on its own that the contract was relevant. The result: three things to prepare. A response to the paper price increase, the late-delivery data, and clause five of the contract, which requires thirty days' notice before prices change. At the end of the answer, the AI names its sources: three emails and one contract file. That way Dewi can open them and check for herself before the meeting. A job that usually takes twenty minutes is done with one question. One thing that often confuses people: the name is different in every app. In ChatGPT, this feature is called connectors or apps, and it's managed in the settings menu. In Claude, it's called connectors. Many connectors use an open standard called MCP, short for Model Context Protocol. You don't need the technical details. Just know that MCP is a shared language between AI and apps. In Gemini, they're called apps or extensions, and because Gemini is made by Google, it's tightly linked to Gmail, Drive, and Calendar. In n8n and Make, the same concept is called a node or a module, which we'll use from Session 6. Since menu names change often, just look for the words connectors, apps, extensions, or integrations in your app's settings.
Step 3: The five most used connectors
There are hundreds of connectors, but for office work and small businesses, these five are used the most. First, Gmail. The AI can search your email, summarize your inbox, and prepare reply drafts. Second, Google Drive. The AI can read documents, PDFs, and presentations, and help you find a file you forgot where you saved. Third, Google Calendar. The AI can look at your schedule, find free slots, and with some connectors, create new events. Fourth, Google Sheets. The AI can read tables and recap numbers. In some apps, Sheets files are opened through the Google Drive connector, so there isn't always a separate Sheets connector. Fifth, Notion, for teams that keep meeting notes, SOPs, and task lists there. My advice: start with the one connector you need most often, not all of them at once. To make it more concrete, here's an example question for each connector. Dewi asks Gmail: which emails do I need to answer today? Dewi also asks Google Drive: what does the payment clause in the Sinar Paper contract say? The AI opens the contract file and quotes the clause. Mr. Hendra from Cerdas Tutoring, a tutoring center, asks Google Calendar: when am I free for an hour in the afternoon this week, for an extra class? Laras, the owner of Sunset Coffee, asks Google Sheets: which product sold best this week? And Dimas, the Sunset Coffee admin, asks Notion: what's our SOP when a customer asks for a return? Notice that all these questions are written in everyday language. Let's look at a Notion connector example. Dimas gets a complaint: a customer named Budi says the one-liter iced coffee he received was leaking. Dimas has only worked there two weeks and doesn't know the SOP by heart. The AI searches Notion and finds the Customer Service SOP page, specifically the Returns and Damage section. The answer follows the shop's SOP, not something made up: ask for a photo within twenty-four hours, offer a free resend or a full refund, then log it in the complaints database. This is the biggest value of connectors for a team: new people can work by the shop's rules without asking Laras every time. And the AI offers a next step, a reply draft, but it doesn't send anything on its own.
Step 4: Connectors compared with copy-paste
So how is this different from the old way, copy-paste? In terms of steps, copy-paste means opening the app, selecting text, copying, and pasting into the chat. With a connector, you just ask. In terms of what the AI sees, copy-paste actually wins on control: the AI only sees what you choose. With a connector, the AI searches on its own, as far as your permissions allow. In terms of freshness, a copy only shows things as they were when you copied them. A connector reads the latest data right when you ask. In terms of the source trail, pasted text often loses where it came from. A connector can name the source email or file, so it's easy to double-check. And the risks are different. Copy-paste can lead to pasting the wrong thing, and data scattered across your chat history. A connector gives wider access, so permissions must be set with care. That's the full topic of the next episode. How big is the difference? Take Dewi preparing meeting materials as an example. Doing it by hand, opening email, the contract file, and the calendar one by one, can take about twenty-five minutes. With copy-paste into AI, reading and summarizing is faster, but copying from three sources still takes time. Maybe around twelve minutes. With a connector, it's one question, plus a few minutes to check the sources the AI named. These numbers are just an illustration, so time yourself before and after using a connector. Even so, a connector isn't always the answer. Copy-paste is better when the data is small and used only once, for example one paragraph of an email you want to polish. Copy-paste is also better when you don't want the AI to be able to search your entire inbox or all of your Drive. And of course, when the connector simply isn't available in your app, plan, or account yet. But notice the red cross. For highly sensitive data, like customers' ID numbers or bank account numbers, the answer isn't choosing copy-paste or a connector. Don't hand it to AI by any method, unless your company policy clearly allows it.
Step 5: Availability: app, plan, account, and region
Connectors aren't available to everyone in the same way. There are four things to check. First, the AI app itself. Which connectors does it offer, and can they only read, or can they also write, like creating drafts or events? Second, your subscription plan. Some connectors are only available on certain plans. I won't quote prices here because they change often, so check the latest pricing page in your app. Third, your account type. If your work email uses Google Workspace, an admin or the IT team can limit which apps are allowed to connect. So if a connector is blocked, it doesn't necessarily mean you did something wrong. Fourth, region and version. New features are often rolled out gradually, and they can differ between the web version and the mobile app. If a friend already has a feature and you don't, this is often why. What does connecting look like? This screen is a simulation, and menu names may differ slightly in your app. Step one, open your AI app's settings. Step two, find the menu called connectors, apps, or integrations. A list of apps you can connect appears there, like the one on the left of the screen. Step three, choose the app, for example Gmail, and press the button to connect it. Step four, you're asked to sign in to your Google account. This is where many people slip up: make sure it's the right account, work or personal. Step five, marked in red, the permission screen appears. Don't press approve right away. We'll cover how to read this screen in full in Episode 22. Step six, once connected, test it with one simple question. What does a good test question look like? Let's see on the next screen. A good test question is one whose answer you can check yourself in ten seconds. For example: list my last three emails, with sender, subject, and time. Don't summarize. The Gmail connector gets to work and fetches the three latest emails. The AI shows them in a table: an email from Sinar Paper Ltd., one from Budi at Pelita Bookstore, and one from Rudi in the warehouse. Dewi opens Gmail in the next tab and compares. It matches. That means the connector is linked to the right account and can read the inbox. Only after this simple test passes does Dewi start asking more complex questions.
Step 6: Asking well and recognizing the limits
A connector lets the AI fetch data, but the quality of the answer still depends on your question. The pattern from Session 2 still applies. Start with the task: summarize emails from a certain sender since a certain date, and flag the ones that need a reply. Then the data block. This is a new and important block when using connectors: say where the AI should search, for example only in Gmail under a certain label, and ignore promotions. The narrower it is, the more accurate the answer. Then the format: a table with the date, the main point, and the action you need to take. And the limits: name the source of every point, and don't send, delete, or change anything. This limit should always be there, especially if your connector can write. Finally, know the limits. First, connector searches can miss data. Old emails, attachments, or files in unusual formats sometimes can't be read. So if the AI says it found nothing, that doesn't mean the data isn't there. Second, the AI can misread numbers, dates, or names. For important decisions, open the source the AI named and check it yourself. Third, being connected is not the same as being free to act. Sending emails, inviting people, and changing data still need your approval, the principle we covered in Session 4. Fourth, access needs to be managed. The more the AI can read, the more important it is to choose permissions carefully. That's the topic of Episode 22. Let's see the first limit in practice. Dewi asks: what was the total billed by Sinar Paper Ltd. over last year? The Gmail connector searches for Sinar Paper invoices, and gets zero results. Maybe the invoices used a different subject, like bill or tax invoice. A good answer looks like this. The AI honestly says it found nothing, explains the possible reasons, offers to search somewhere else, and makes clear it won't guess the amount. If your AI gives a number with no source instead, that's a warning sign. Add this to the limits in your prompt: if you can't find the data, say so, and don't guess. Before moving on, do this small task. First, open your AI app's settings and note which connectors are available for your account and plan. Second, write three work questions whose answers are in your email, schedule, or tables. Like the questions from Dewi, Laras, and Mr. Hendra earlier. Third, mark which data is sensitive and shouldn't be opened to AI, such as customers' personal data or financial documents. Fourth, hold off on connecting your work account until you've watched Episode 22, so you know exactly what permissions you're giving.
Done!
Let's sum up. A connector is a permission-based bridge between AI and your work apps. The flow: you ask, the AI picks a connector, the connector reads the data, and the AI answers while naming its sources. Availability depends on the app, the plan, the account, and the region. In the next episode we cover an equally important side: security and access permissions. We'll read the consent screen line by line, tell read permissions from write permissions, and learn how to revoke access. See you in Episode 22.
Practice exercises
Try these on your own after watching to lock in what you learned.
Open your AI app's settings and note which connectors are available for your current account and plan.
Write three work questions whose answers are in your email, schedule, or tables, along with the connector each one needs.
Mark which data is sensitive and should not be opened to AI. You will use this note again in Episode 22.
Ask the class material
Type a question or keyword; we find the part of this episode that explains it and jump the video there.
My notes
Notes are saved with the current video time and are only visible to you.
Questions & answers
Stuck on a step? Ask here; the instructor and other learners can answer.
No questions yet.
Your instructor
Cerdas Inovasi Digital Team
Software developers, BIM & 3D engineers, and AI practitioners
53 video classes1,283 episodes
We build software, BIM models, games and AI automations for clients every day, and turn that hands-on work into step-by-step classes: every step is shown on screen and comes with a PDF guide you can keep.
