How to develop AI Agent capable of scheduling, cancelling, and modifying calendar events and meetings.
Building your own agent with OpenAI’s SDK is actually super straightforward. You get full control, no unnecessary complexity — and most of the time, you don’t even need all that extra stuff.
Interacting with the OpenAI SDK:
Before anything else, you need to know how to communicate with the OpenAI SDK. The sample code below demonstrates a basic chat interaction.”
This is the output : Agentic AI refers to artificial intelligence systems that have some degree of autonomy and can make decisions or perform tasks independently. These systems are designed to achieve specific goals or objectives by interacting with their environment, often employing complex learning and reasoning processes.
Now you could change the behaviour of the model by changing system prompt above now look at the code below
This is the output : Agentic AI, it do things by self like a robot with own will, but not really. It kind of acts like a human with goals, but still just follow code!
At the core of your agent is this: sending the right prompt to the model and shaping its behavior with the system prompt to reliably get the output you’re aiming for.
API that powers the action:
The second key piece is understanding the API that powers the action — whether it’s a weather API or, in my case, Google Calendar.
Let’s start by walking through how to create events or meetings. I’ll skip the authentication setup and focus purely on what the API expects and how it works.
The main thing to know when creating an event is that you’ll be sending a JSON object with stuff like the summary, description, start/end times, timezone, and attendees (we’ll sort out attendees later).
To get the desired output, the LLM must return a JSON object with these fields. We’ll modify the system prompt to ensure that happens. Here’s the code:
With the response in hand, you can parse the required parameters and pass them directly into the API call to create the event.
now once you are able to extract the parameters and everything now you can create a gate check to see whether everything has created properly in the main block.
Now that our agent can create calendar events, the next step is making it smart enough to tell the difference between meetings and regular events. For meetings, we’ll also need to include a Google Meet link. Let’s explore how to do that.
Difference between meetings and regular events
Step one: get the LLM to tell the difference between meetings and regular events. Just tweak the prompt so it includes the type of event in the response.
If the LLM identifies the event as a meeting, make sure to set conferenceDataVersion=1 in the API request to generate a Google Meet link. Then, handle it using a subprocess like the one below.
With event and meeting creation in place, the next step is to handle attendees. Since the API requires a list of email addresses, you’ll need to extend the LLM’s prompt to map attendee names to their emails. In most cases, providing just first names should be sufficient.
first contacts database on a real app there would be a backend db but here we can have a list of names and mails like below:
This functions as a knowledge base to map attendee names. The updated prompt instructs the LLM to return an array of objects with the input name and the matched full name, which a helper function then uses to fetch the associated email IDs.
Attendees are added to the Calendar API request when present, and sendUpdates=’all’ is set to trigger email notifications.
And that’s how you build an AI agent — step by step, gradually increasing complexity and expanding its capabilities. Below are some screenshots of my agent in action.
