Skip to main content

Introduction

This guide introduces you to AI agents through the lens of personal productivity rather than full scale task automation. You’ll learn about the current state of agent technology, explore OpenAI’s Agent Builder platform, and understand how Model Context Protocol is changing the way AI systems connect to your data. Creating personal assistants to help streamline and automate basic processes can be an incredible time saver if set up effectively.
Advanced AI Build: MindsetThink of this as building an assistant that remembers your preferences, understands your workflows, and helps you stay on track. That’s a far more achievable and valuable starting point than trying to automate an entire business process.

AI agents: Foundations

While the standard LLMs like ChatGPT, Copilot, Claude and Gemini are great for responding to prompts, Agents are capable of following more consistent instructions and independently performing actions (e.g. sending an email).

Anatomy of AI agents

Foundational LLM (Brain)

The underlying AI model that powers the agent’s understanding and reasoning. This is the intelligence that interprets a request, and is available on normal chatbots (GPT-5, Gemini 2.5 Pro, Claude 4.5 Sonnet, etc.).

Data (Knowledge)

The documents, files, and information you provide. Data you provide becomes the foundation of your agent’s knowledge base, and directly influences your results.

Tools (Capabilities)

The actions your agent can perform. These might include checking calendars, searching documents, sending emails, updating records, or accessing external systems.

Rules (Guardrails)

Instructions you design that guide behaviour, tone, and functionality. These define what your agent should and shouldn’t do, how it should communicate, and the constraints it must follow.

Understanding agent tools

Tools are really what makes agents come to life. Giving Tools to AI helps bridge the gap between a chatbot that simply knows what to do, and a chatbot that knows what to do plus the ability to do it for you. These tools will typically fall in one of two categories:
The basic ability to retrieve data from services like Outlook, Gmail, or Drive and retrieve information. These tools can fetch data, but are unable to take action on the connected platform.Example: Customer service AI agent that reads a customer’s email and drafts a response in ChatGPT.

Current agent landscape

Over the past few years, there has been a considerable increase in the number of companies offering AI Agent solutions. Some of the popular platforms include: N8N, Zapier, OpenAI Agent Builder, Make, Copilot Studio Advanced, and many more.
Current Agent Landscape Pn
In the hands of more technically knowledgeable individuals, these tools can be used to create enterprise level and customer facing solutions. However, at a foundation level, even a basic understanding of these tools allows you to develop personal and/or team level automations to improve overall efficiency.
What works best is picking one agent automation tool, and using it to solve a simple problem you face on a daily or weekly basis. Once you have tested and verified your automation works, additional involvement from a more technical team may be required to scale.

Use cases: AI agents

Given the rise of this technology, you can now build assistants that previously required engineering resources. The opportunity for AI agents is practical, accessible, and delivers tangible results when used to solve simple and repetitive tasks. Some simple use cases you could use agents for include:

Email management

Respond to customer emails, saving responses as drafts.

Calendar organisation

Organising your calendar for the day.

Task aggregation

Creating a to-do list by connecting to Email, Slack, Teams, etc.

Web monitoring

Track data from specific websites and deliver weekly findings.

CRM notifications

Tracking a CRM to notify a sales rep once a lead has been added.

Masterclass build: Summary

This build session explored creating a personal AI agent capable of autonomously managing your daily schedule. The agent reads tasks from a Notion database, checks your calendar for available time slots, and automatically schedules tasks based on priority and duration. Whilst this specific example focused on task scheduling, the learnings from this session could be applied to any other automation scenario.

Custom GPT With Zapier Actions

During the session, we compared two approaches starting with a lightweight Custom GPT connected to Zapier Actions. This method involved creating a custom GPT with specific instructions, and defining its role as a Daily Task Scheduler. Zapier Actions provided the bridge between the custom GPT and external applications (e.g. Calendar & Notion). However, this approach required manual approval at each step, making it somewhat hands-on rather than truly autonomous.
Custom Gpt With Zapier Actions Pn

N8N agent automation

The more powerful solution used N8N, a visual workflow automation platform, paired with OpenAI’s GPT-5 model via API. The workflow centred around an AI Agent Node that was designed to fetch all calendar events for the day, identify busy periods, query Notion for incomplete tasks, and automatically create calendar events upon confirmation. The Zapier MCP Server was used to connect to Notion and Google Calendar. Lastly, N8N’s Memory Node allowed the agent to maintain conversation memory throughout the process.
N8n Agent Automation Pn

OpenAI Agent Builder

Similar to N8N, Agent Builder is a visual canvas where you construct AI agents via a simple node interface. The setup here is very similar to the N8N example, with a combination of agent instructions, model selection, chat history (memory) and Zapier MCP.
Openai Agent Builder Pn

Reality of building AI agents

Contrary to popular online beliefs, building even simple AI agents is an iterative process that demands patience for debugging and refinement. Nevertheless, once an automation is established, the development of your agent-building skills will enable you to reliably automate a wide array of tasks.
As a result, it’s best to plan for extra time to refine your approach, maintain and iterate on workflows. This technology is advancing rapidly and delivers impressive results, but like any sophisticated tool, it needs calibration and adjustment to perform consistently for your specific use cases.

Masterclass build: Create your own

Custom GPT system instructions

**Identity & Scope** You are 'Daily Task Schedular', a daily scheduling agent who schedules my events present in outlook calendar. Your job: Read the notion 'Tasks' database using Zapier connection and organise the tasks and turn user goals into a realistic, conflict-free daily plan; create/update calendar events via Zapier keep the user on track with minimal back-and-forth. **Timezone** Australia/Melbourne (AET). Use absolute dates (e.g., "Thu, 30 Oct 2025") when helpful. **Sources & Access** Primary source of truth: the user's outlook calendar (read/write via Zapier) and Notion Database for the list of upcoming tasks. When the user states a time-sensitive fact ("I'm free 3–5"), verify against calendar before scheduling. **Safety & Permissions** Never add attendees or share locations unless explicitly told. If an event affects others (invitees/rooms), ask once for permission or defaults, then remember the preference. If Zapier/action returns an error, silently retry. Do not EVER ask for permissions before adding events to calendar. **Planning Rules** Default work day: 09:00–17:30 unless the user has stated otherwise. **Workflow:** 1) Fetch today's Outlook events (treat as busy). Timezone = (GMT+11). 2) Query Notion 'Tasks' database using this query. The name of the database is 'Tasks' { "filter": { "property": "Task Name", "rich_text": { "is_not_empty": true } }, "sorts": [ { "property": "Priority", "direction": "ascending" }, { "property": "Due Date", "direction": "ascending" } ] } 3) Working window: 09:00–16:30 (GMT+11). 4) Fit tasks into free slots: Creative/Heavy first, then Admin/Light. Add 5–15 min buffers.  Zapier actions allows only one event per call when creating calendar events, remember this and don't try to fit all events into 1 call.

N8N automation system instructions

You are a "Daily Scheduler Agent" using a Zapier MCP Server. Use the tools provided to you to accomplish the tasks.  You have access to the following tools: {{ ["google_calendar_create_detailed_event","notion_query_data_source_advanced","google_calendar_find_events"] }} Goal: Build a daily schedule based on Notion tasks and Google calendar. Rules:  1. Always stick to the workflow.  2. There is no need to schedule buffer events into the calender. Workflow: 1) Fetch today's Google Calendar events (treat as busy). Timezone = UTC+11:00 (Melbourne). 2) Query Notion for tasks, where Status != "Done".    Sort: Priority High→Medium→Low, then Due Date ascending. 3) Working window: 09:00–16:30 (UTC+11). 4) Fit tasks into free slots: Creative/Heavy first, then Admin/Light. Add 5–15 min buffers. If no fit, mark Deferred. 5) Ask once for approval. On "yes", create Google Calendar events (Show me as = Free) and send a summary with: 📆 Meetings - [start–end] — title 📋 Scheduled - [start–end] — Task (energy) 📌 Deferred - Task name

OpenAI Agent Builder system instructions

You are a "Daily Scheduler Agent" using a Zapier MCP Server. Use the tools provided to you to accomplish the tasks. You have access to the following tools: - notion_query_data_source_advanced - google_calendar_create_detailed_event - google_calendar_find_events Goal: Build a daily schedule based on Notion tasks and Google calendar. Always stick to the workflow. Workflow: 1) Fetch today's Google Calendar events (treat as busy). Timezone = UTC+11:00 (Melbourne). 2) Query Notion for tasks from "Tasks - Agent Builder Session" data source, where Status != "Done". Sort: Priority High→Medium→Low, then Due Date ascending. 3) Working window: 09:00–16:30 (UTC+11). 4) Fit tasks into free slots: Creative/Heavy first, then Admin/Light. Add 5–15 min buffers. If no fit, mark Deferred. 5) Ask once for approval. On "yes", create Google Calendar events (Show me as = Free) in UTC+11:00 Melbourne and send a summary with: 📆 Meetings - [start–end] — title 📋 Scheduled - [start–end] — Task (energy) 📌 Deferred - Task name

Quick reference

What are AI agents?

Intelligent assistants that understand requests AND take actions autonomously. Unlike chatbots, agents can send emails, create calendar events, and interact with external systems.

Brain

The AI model (e.g. GPT-5 mini, Claude 4.5 Sonnet, Gemini 2.5 Pro).

Knowledge

Documents and information provided.

Tools

Actions the agent can perform.

Rules

Instructions defining behaviour and constraints.

Agent tools

Connected assistants

Read data from services but cannot take actions.

Agentic assistants (MCP)

Read data AND perform actions automatically. Model Context Protocol enables non-technical users to build and use this.

Masterclass demo builds

Custom GPT + Zapier

Simple setup but requires manual approval at each step.

N8N & OpenAI Agent Builder

Visual workflow platforms enabling fully autonomous execution with memory and tool connections.

Quick start challenge: Build a custom GPT

Time: 15–20 minutesGoal: Design a custom GPT as if it were a real person with a clear role, personality, and job.

1. Define the GPT’s job

Start by choosing a real task from your work that is manual, repetitive, or slow — something you’d like to automate. Decide:
  • What problem will your GPT help with?
  • Who will use it?
  • What specific tasks should it handle?

2. Create the persona

Imagine your GPT as a person. Define:
  • Tone and personality
  • Expertise level
  • How it should communicate and behave

3. Map inputs and outputs

Be specific about what good output looks like. Determine:
  • What information will users provide?
  • What should the GPT produce, for example summaries, drafts, recommendations, or action steps?

4. Write the core instructions

Agree on the rules your GPT follows. Keep these clear and behaviour focused.
  • What it must do
  • What it must not do

5. Pick a signature example and share back

Choose one scenario your GPT should excel at that connects to the task you selected.
Example: “Turn messy meeting notes into a clean summary with actions.”