Getting Started with AI Agents.

Well you all can just simply google the define for AI Agents , and I’m not writing this to bore you , we’ll get straight to the point .

What was the problem with traditional LLM’s ?

Large Language Models (LLMs) like GPT, Gemini, Claude, or Mistral are incredibly smart — but they're static.

Once trained, they don’t learn new things unless retrained — and let’s be real, retraining isn’t cheap or frequent. Most foundation models are updated every few months, at best.

So what if you ask your AI:

“Tell me the latest news about Apple's WWDC 2025.”

It’ll reply based on what it knew at the time of training — not what happened this morning.

Even if a model is trained daily (which almost no one does), the knowledge is still 24 hours old. For fast-moving domains like news, finance, sports, or even social media trends — this is unacceptable.

We want fresh , current data / knowledge sometimes to perform our tasks , that’s where AI Agents come into play .

What are AI Agents ?

AI agents are a game-changer. They don’t just generate — they act.

Think of them as intelligent bots that can:

• Read your command in natural language

• Decide what tools or actions are needed for the task

• Take action using APIs, search engines, or files

• Return results that are up-to-date and useful

They combine the brains of an LLM with the agency to interact with the outside world.

It’s like attaching hands and legs to a powerful LLM brain 🧠

Benefits of AI Agents — Simplified

1. They Automate Complex Tasks

AI agents can do tasks that normally need human effort — like collecting data, sending emails, or organizing stuff.
They work faster, cost less, and don’t need someone to guide them every time.
Once set up, they figure out the steps on their own.

---

2. They Work Better Together

When multiple AI agents work as a team, they give better results than just one agent.
Why? Because they learn from each other, share ideas, and fill in missing information.
Think of it like a group project where everyone brings a skill — the final work is stronger.

---

3. They Give Smarter, Personalized Answers

AI agents don’t just reply — they:

• Look things up in real-time

• Use tools (APIs, search, etc.)

• Remember what you said before

So the answers you get are more accurate, more relevant, and feel human-like — not robotic or outdated.

Risks & Limitations of AI Agents — Simplified

1. Too Much Dependency on Each Other

Some tasks need multiple AI agents working together.
But if one agent makes a mistake, others might follow — leading to a chain of failures.
If all agents are built on the same model, one weakness can break the entire system.
That's why careful testing and good data rules are important.

---

2. Stuck in a Loop

Sometimes, an AI agent might get confused or stuck, using the same tools again and again without solving the problem. This is called an infinite loop.
To prevent this, some human monitoring or control might still be needed.

---

3. High Cost and Time

Training smart AI agents takes a lot of computer power, time, and effort.
If the task is complex, the agent might take hours or even days to finish it.
Building agents from scratch isn’t easy — especially for small teams.

---

🔐 4. Privacy Risks

If not handled properly, AI agents working with personal or company data can cause privacy and security issues.
Imagine an AI setting prices or writing code without human checks — the results could be risky or wrong.
That’s why companies like OpenAI, Microsoft, and IBM need to ensure strong safety rules are in place.

Building Stuff Part :

🛠️ I built a CLI tool where you can ask for the latest news with your own words —

Just type something like:

latest news on apple
…and it’ll instantly fetch and return real-time updates, not some cached summary.

No fancy UI. Just raw speed and intelligence.
Built for devs, researchers, and curious minds who hate outdated info.

1) Setting up system_prompt Correctly is the Key :

2) Handling Agent Calling via OpenAI API (feel free to use any other)

messages = [{"role":"system","content":SYSTEM_PROMPT}]
query = input("User: ")
messages.append({"role": "user", "content": query})

while True:
    response = client.chat.completions.create(
        model = "gpt-4o-mini",
        response_format={"type":"json_object"},
        messages = messages
    )
    messages.append({"role":"assistant","content":response.choices[0].message.content})
    parsed = json.loads(response.choices[0].message.content)

    step = parsed.get("step")   
    content = parsed.get("content")

    if step == "action" :
        news_list = tools[parsed.get("function")](parsed.get("topic"), parsed.get("category"),int(parsed.get("ps")))
        messages.append({"role":"assistant","content":json.dumps({"step":"observe","output":news_list})})
    if step == "output" :
        print(content)
        break

3) The Brain of AI-Agent (A simple function)

def get_news(topic,category,ps=2):
    top_headlines = newsapi.get_top_headlines(q=topic,
                                          category=category,
                                          language='en',
                                          country='us',
                                          page=1,page_size=ps)
    news_list = []
    for a in top_headlines["articles"]:
        news_url = a["url"]
        news_title = a["title"]
        news_list.append({"title":news_title,"url":news_url})

    return news_list

Working Demo:

Provides the news article Title as well as the URL to read the complete news .

That’s about it about the AI Agents .

⛓️‍💥 Source Code : https://github.com/gaurangsaini01/GenAI/blob/main/06-news-ai-agent/main.py
🖇️ LinkedIn : https://www.linkedin.com/in/gaurang-saini-00a930253/

🔚 Wrapping Up

AI agents are a big step forward — they don’t just think, they act. From automating tasks to giving real-time, personalized responses, they’re changing how we interact with technology.

But they’re not perfect — challenges like feedback loops, privacy issues, and system dependencies still need attention. With careful design and strong safety measures, AI agents can become powerful tools in both tech and business.

So, what’s next?

In the next article, we’ll explore something just as exciting — RAG (Retrieval-Augmented Generation).
It’s a smart way to make AI even more accurate and up-to-date by combining the power of LLMs with external knowledge sources. or before that I might make a Mini CLI based Cursor also , thats lets you vibe code from the terminal :)

Stay tuned. The future of intelligent AI isn’t just about generating — it’s about retrieving, reasoning, and acting smartly. 🔍🤖

Bye !