# What is an AI agent? A practical guide from someone who ships them

By Kush Ahuja · Sep 30, 2026 · Building agents-hub, part 1
Source: https://www.thekush.codes/blog/what-is-an-ai-agent

> A plain-English definition of AI agents, how they differ from chatbots, the four parts every agent needs, and what I learned building the agents-hub AI agents for Gmail, Calendar, Tasks and Docs.

## Key takeaways
- An AI agent is software that uses a language model to decide which actions to take, takes them through tools such as Gmail or a calendar, and checks the result, instead of only replying with text.
- Every useful agent has four parts: a model, tools, memory and guardrails. Most failures come from the last two.
- agents-hub runs separate AI agents for Gmail, Calendar, Tasks, Docs and Maps, with one hub that routes each message to the right agent.

## What is an AI agent?

An AI agent is a program that uses a large language model to decide what to do next, does it through real tools, and looks at the result before answering. A chatbot tells you how to add a meeting. An agent adds the meeting to your calendar, checks for a clash, and tells you it is done.

I build AI agents for a living. I am CTO of BuildYour.Company and I build agents-hub, a team of AI agents that works inside your Gmail, Google Calendar, Google Tasks and Drive, and that you talk to in Telegram or a mobile app. This guide is what I wish someone had told me before I started.

## AI agent vs chatbot

- A chatbot answers. An AI agent acts: it calls tools, reads what came back and decides the next step.
- A chatbot forgets. An agent needs memory, so "send that to Priya" knows what "that" is.
- A chatbot can be wrong in words. An agent can be wrong in your inbox, which is why guardrails matter more than prompts.

## The four parts of every AI agent

1. A model that plans. In agents-hub that is Azure OpenAI with function calling: the model chooses which tool to call and with what arguments.
2. Tools that do real work. Gmail search, create draft, calendar events, Google Tasks, a Python sandbox for documents, Google Maps for places and routes.
3. Memory. Recent messages, facts that never scroll away, and an archive searched only when needed. See part 5 of this series for the three-layer design.
4. Guardrails. Hard limits the model cannot talk its way around, such as "the AI can draft an email, only a human tap can send it".

## Single agent or many?

One giant agent with every tool gets confused and expensive. agents-hub uses one small agent per job (Gmail, Calendar, Tasks, Docs, Maps) and a hub that reads each message, splits it if needed, and routes each part to the right agent. "Mail Priya the deck and block Friday 4pm" becomes a Gmail task and a Calendar task that run in the same round.

## What makes an AI agent good

- It waits for you to finish typing before it acts (part 4).
- It shows progress on long jobs instead of going silent (part 6).
- It never sends, deletes or pays without a clear human yes (part 7).
- It remembers what you told it last week without re-reading everything every time (part 5).

## Where to start

If you want to build one, start with one tool and one guardrail, not ten tools. If you want to use one, pick an agent that works in the accounts you already have. That is the idea behind agents-hub: sign in with Google once, unlock the agents you need, and they do the work where your work already lives.