What Are AI Agents?

Mental Model (TL;DR)

Think of an AI agent as a traditional server app + LLM integration + function tools. The model plans and explains; the tools do real work (math, API calls, etc.); your server glues it together.

Understand AI Agents through Simple Problems

Instead of abstract definitions, think in terms of tasks the agent can do via tools.

function sum(a: number, b: number) {
  return a + b;
}

function isPrime(n: number) {
  if (n < 2) return false;
  if (n % 2 === 0) return n === 2;
  for (let i = 3; i * i <= n; i += 2) {
    if (n % i === 0) return false;
  }
  return true;
}

async function getCryptoPrice(
symbol: string, vs = 'usd'
) {

  /* 

  fetch from API and return price
  e.g., https://api.coingecko.com/api/v3
  /simple/price?ids=bitcoin&vs_currencies=usd

  */
}

async function getWeather(
  city: string,
  unit: 'celsius' | 'fahrenheit' = 'celsius'
) 
{
  /*

   geocode the city, then fetch weather
   e.g., Open-Meteo geocoding 
   + forecast endpoints
   
   */
}

From User Input to Function Calls

The model reads the user message and decides if a tool is needed. If yes, your server calls the right function with structured args, returns the result to the model, and the model explains the final answer.

• "is 97 prime?" → call isPrime(97)
• "btc price in usd" → call getCryptoPrice('btc','usd')
• "weather in London" → call getWeather('London')

How the Pieces Fit

1. Your server/API routes host small, reliable tool functions.
2. The LLM handles language understanding and planning.
3. A tiny orchestrator loop decides which tool to call, feeds results back to the model, and stops when a final answer is ready.
4. Optional: add context/memory by keeping recent chat history.