DEV Community: Marika Tchavtchavadze

Developers have been communicating with AI for a while — writing prompts, guiding tools like ChatGPT or Copilot. It’s slowly becoming part of how we build and solve problems. Will it soon become a must-have skill for developers?

Marika Tchavtchavadze — Mon, 20 Oct 2025 14:48:26 +0000

The AI Era: New Prompt Engineering is Just the Start: The Real Skill Developers Need is AI Product Leadership

Marika Tchavtchavadze ・ Oct 19

#ai #promptengineering #newera #learngoogleaistudio

The AI Era: New Prompt Engineering is Just the Start: The Real Skill Developers Need is AI Product Leadership

Marika Tchavtchavadze — Sun, 19 Oct 2025 20:19:10 +0000

🧠 Communicating With AI: The New Skill Developers Need in 2025

Talking to AI is becoming a core skill for developers — and it’s not just about writing prompts.

This week I was exploring Google AI Studio, and it made me think a lot about this new kind of skill — instructing AI clearly and guiding it to build exactly what you need.

💬 Talking to AI is Not Just Prompting

We already have a new profession called Prompt Engineering, with courses and tutorials everywhere.
It’s a fairly new concept — it started emerging around 2020 with GPT‑3 and became more formalized with ChatGPT in late 2022, so it’s been around for roughly 3–5 years.
But it’s evolving fast — and it’s not just about “writing a good prompt.”

Communicating with AI is more like managing a team: you need to understand your product, what it should do, and how to explain it clearly.
That’s when AI stops being confusing and starts helping you. Some call this the art of asking the right question

⚙️ My Experiment in Google AI Studio

Even though I can code, I wanted to see what’s trending now, so I decided to build a task scheduler and habit tracker app — with dashboards — using just prompts.

The first version appeared in minutes. But when I tried to adjust it, I noticed something important:

Large updates often gave results I didn’t like.
The trick was step-by-step commands, not everything at once.

After iterating this way, I got a version I liked — and I could share it to GitHub or deploy it with one click.

🧩 What Skills Actually Matter?

For developers using AI tools, the key skills are:

Coding: helps you understand the output and tweak it later if needed.
UI / product sense: AI can build interfaces, but you know what works best.
Product thinking: understanding the goal keeps your prompts effective.

The better you understand these areas, the better you can command AI.

📘 Learn From the AI-Generated Code

Even when using only prompts, AI tools often show the code behind the output.
You can study it, understand how it works, and learn from it — which helps you later build your own products more confidently.

🧭 The Real Skill: Giving Clear Commands

Communicating with AI is like leadership.
You give instructions, review the results, adjust, and guide again.

Good prompts are not too general and not overloaded with details.
They are clear, precise, and goal-focused.

A little knowledge about LLMs (Large Language Models) and NLP helps you understand how AI interprets your instructions.

🔗 Prompt Engineering Guide

🌟 Why This Skill Matters in 2025

AI tools are everywhere — from app builders to content creators.
Knowing how to communicate with AI is becoming a core skill for developers.

This is just how the landscape is changing: the better you get at guiding AI, the more you can prototype faster, learn faster, and create smarter products.

🌍 New Jobs, New Era

People used to say AI will take jobs.
But I see something different — AI is creating new opportunities.

From side income projects to full products, developers can leverage AI to build, learn, and experiment faster.
This era is here — and it’s up to us to adapt.

💭 Final Thought

If you’re a developer, start experimenting with AI tools.
Play with Google AI Studio, ChatGPT Canvas, or similar tools.
See what happens when you talk to your code instead of writing it line by line.

You’ll learn a lot — about AI, about clear thinking, and about guiding complex systems effectively.

What do you think?
Will “AI communication” become a must-have developer skill?
Or is it just a temporary trend?

Would love to hear your thoughts 👇

The Future Isn't Machines. It's Algorithms

Marika Tchavtchavadze — Fri, 17 Oct 2025 22:59:48 +0000

I remember movies from my childhood — robots slowly gaining feelings, struggling with identity, and even dying like humans. Flying cars filled the skies in futuristic cityscapes. Robots weren’t just helpers; they took over jobs like cooking, cleaning, and even nursing. Holograms taught students in classrooms, guiding lessons as if they were real teachers. Back then, school discussions and headlines speculated what 2020 would look like — a world run by intelligent machines, where technology felt magical and omnipresent

And now? We got something very different. Mostly recommendation systems, automation, and conversational chatbots. Not the sci-fi spectacle we imagined.
The future isn’t full of machines around us. It’s algorithms inside our apps. Invisible. Predictive. Quiet.

We don’t have conscious robots — and honestly, I’m happy about that. Robots shouldn’t mimic humans in areas where emotions and shared life matter — where we co-live, co-exist, and interact socially. Tasks that require trust, empathy, or subtle human judgment shouldn’t be handed over to machines pretending to feel. Technology should support us, not replace the nuances of human connection. What we do have are systems that predict our preferences, understand our personalities, and take care of boring, repetitive tasks.

Yes, someone could build machines in human shape. Intelligent systems already exist, capable of learning, predicting, and adapting. Combine that with human-like form, and you could have machines that seem human — acting, speaking, and reacting like us. But appearances can be deceiving. They might imitate behavior, but they don’t share experience, emotion, or understanding. From my perspective, the value isn’t in machines pretending to be people - the point is humans designing systems to make life and work easier.

Maybe the future isn’t about sci-fi fantasies. Maybe it’s about making the invisible intelligent — and letting us focus on the things that actually need us.

- What 'sci-fi' tech from your childhood do you wish we actually had?

How Much Should We Rely on AI?

Marika Tchavtchavadze — Wed, 01 Oct 2025 12:13:15 +0000

AI Makes Starting Easier

AI has changed how we build, think, and start. For me, it has made it so much easier to shift between fields, start new projects quickly, and let large language models (LLMs) guide me depending on my commands.

It’s not just about switching fields — AI is also amazing for one-time projects. For example, I once helped a friend with web scraping. I was hesitant to start because I didn’t have much experience with scraping, and it wasn’t really my interest. But this time, AI guided me through the process. Thanks to my coding knowledge, I knew how to ask questions, what information to give AI, and how to solve errors. Without AI, I might never have started.

But here’s the truth:
I don’t let AI do everything for me.

It’s great for repetitive tasks, or busywork, but every task has its own “soul”—its unique niche that reflects me as a developer. By keeping the core ideas mine, I’ve developed a better understanding of my work and made it more modular and reusable. If I relied on AI fully, I’d just be producing copies of work, not creating something new.

Using AI Wisely: Better Code, Less Waste

And there’s a bigger picture: relying on AI fully just produces copies, which require server computation and electricity. One person’s copies may seem small, but billions add up — more energy, more emissions, more pollution. By keeping the core work mine and reusing it, I reduce AI usage, learn more and minimize environmental impact.

Thinking vs. Decision-Making

Sometimes I have a vivid idea. I’ll share it with an LLM and ask it to ask me questions back. This makes me think harder, and thinking is vital — it’s what differentiates us from AI.

AI is a decision-maker from data.
Humans are thinkers.

What’s crucial is that my idea is still mine. I’m not “stealing” from AI models, which are built from other people’s perspectives. I’m using AI to challenge my thoughts, not to replace them.

AI as an Assistant, Not a Leader

Doing the same stuff again and again can be overwhelming, and here AI shines. It automates repetitive tasks, freeing my time. But I still know what I’m doing; it’s under my ownership.

AI is my personal assistant.
It is not my leader.

Never Lose Your Identity

This is the most important part. Authenticity still matters. AI may be faster than someone else, but in the end, authenticity wins.

AI helps us start easier. Even this post is polished with AI, but all the ideas and thinking are mine. This is my thought. This is authentic.

Takeaway

Use AI to start faster, automate the boring, and spark better questions. But keep your identity, your thinking, and your authenticity at the center. That’s how you rely on AI without losing yourself.

(Follow me if you want more thoughts and lessons from my AI + dev journey.)

The Big AI Hype: What’s Really Behind the Name

Marika Tchavtchavadze — Fri, 26 Sep 2025 12:58:20 +0000

What Does “Artificial Intelligence” Really Mean?

Artificial intelligence is often presented as the innovation of our time. But what truly stands behind this name?

If you look closer, you’ll see that AI is mostly automation and probability models. Behind the label are mathematical algorithms that have existed for decades. The difference today is that we finally have the data to apply them at scale — to train, test, and use them in real-world systems.

Probability isn’t new. Automation isn’t new. What’s new is the combination of these with massive datasets.

When Recommendations Feel Like “Intelligence”

When YouTube suggests what to listen to next, it doesn’t actually know you. It just calculates probabilities based on people like you.
And what does “people like you” mean?
It simply groups you with others who clicked or watched similar things. That’s not mind-reading — it’s math.

But because it feels personal, we can easily think: “Oh no, maybe there’s really AI behind this 🤯.”
(Let that thought sit for a second.)

Automation ≠ Consciousness

AI is not consciousness.
AI doesn’t feel for you.
AI only follows models trained on past data

# The 'Conscious' AI 

user_input = "I'm stressed about work"

# Predefined keywords and responses
keywords = {
    "stressed": "Stress detected. Take a break!",
    "anxious": "Anxiety detected. Remember to breathe.",
    "sad": "Sadness detected. I’m here to listen."
}

# Pattern recognition + probability
response = "I don’t understand."  # default
for word, reply in keywords.items():
    if word in user_input.lower():
        response = reply
        break

print(f"AI Response: {response}")

If there’s enough data and a model is trained on huge amounts of examples, it can feel like it has consciousness, empathy, or understands your feelings. But don’t let that fool you. It’s still pattern recognition, probability, and automation at work — not real understanding.

AI as a Tool, Not a Threat

The name “AI” makes it sound huge, futuristic, and even scary. Some people fear it could take their jobs. But at its core, AI is just a tool — built on automation, probability, and human-designed models.

Think about cars: when they replaced walking, they didn’t destroy work, they created it. Driving became a profession. Similarly, “AI” isn’t a job-killer by itself — it’s a tool that opens up new opportunities.

Final Thought

Don’t let the wrapping fool you. Artificial Intelligence is not magic. It is a tool—a complex form of automation built on decades of math and powered by unprecedented amounts of data. The power is real, but the intelligence is an illusion of probability, not consciousness. Treat it as the powerful, human-designed instrument it is.

What do you think — should we stop calling it “Artificial Intelligence” and start calling it what it really is or leave it as it is?

Messenger Bot Setup: The Mistake That Taught Me the Most

Marika Tchavtchavadze — Wed, 17 Sep 2025 17:47:30 +0000

I’m diving into chatbots recently, so the next station was a Messenger bot. I thought I could just follow some steps with ChatGPT as my guide… but I missed one very important point.

👉 If you want to create a Messenger bot, you must have a business page under a BUSINESS type Meta app.

I didn’t know that. I was creating Meta apps and choosing anything but the right type—and that’s where the loop started.

Here’s the trap: once you create an app, you can never change its type. I went to the App Dashboard, tried to add Messenger under “Add Product”… but the magic button was gone.

I tried everything—different setups, endless searches, even trolling the web for answers. Nothing. Days passed.

And then I found the hidden treasure 🗝️:
When creating an app, you need to choose Other → Business. That’s it. If you miss this, Messenger will never show up as an option.

Here is the link where you can create a new app:
developers.facebook.com/apps

Once you start creating the app, in Use Cases choose Others

Then click Next and choose Business type

This is the magic I missed. After finding this, I could finally move forward! 🎉

Page Access Token

Next, you need to create a Page Access Token from Graph API Explorer
and make sure to select your business page—this is a MUST.

After that, you can add Messenger as a product in your app.

Also, set up a Verify Token (you can write your own). Be careful to save it somewhere safe and never share it.

Ngrok — The Local Tunnel

The journey is not finished here. Now you need ngrok. When you run your bot locally, Facebook cannot reach localhost. Ngrok creates a secure HTTPS tunnel to your machine so Messenger can communicate with your bot.

How to install ngrok

- Download from ngrok.com
- Unzip/Install it
- Authenticate once with your ngrok account:

ngrok config add-authtoken YOUR_AUTH_TOKEN

After running your bot script

from flask import Flask, request
import requests
import json

app = Flask(__name__)

PAGE_ACCESS_TOKEN = "PAGE_ACCESS_TOKEN"
VERIFY_TOKEN = "VERIFY_TOKEN"

# The bot will look for an exact, case-insensitive match.
qa_database = {
    "hello": "Hello! I am messenger bot and I am under development. Can't answer any questions yet.",
}

# --- Root route for testing ---
@app.route("/", methods=["GET"])
def home():
    return "Messenger bot is running!"

# --- Webhook route for Facebook ---
@app.route("/webhook", methods=["GET", "POST"])
def webhook():
    if request.method == "GET":
        # Verification handshake
        token_sent = request.args.get("hub.verify_token")
        return verify_fb_token(token_sent)
    else:
        # Handle messages
        output = request.get_json()
        for event in output.get("entry", []):
            for messaging_event in event.get("messaging", []):
                sender_id = messaging_event["sender"]["id"]
                if messaging_event.get("message"):
                    message_text = messaging_event["message"].get("text")
                    if message_text:
                        # Convert message to lowercase for case-insensitive matching
                        message_text_lower = message_text.lower().strip()

                        # Look for a response in the QA database
                        response_text = qa_database.get(message_text_lower)

                        if response_text:
                            send_message(sender_id, response_text)
                        else:
                            # Default response if no match is found
                            default_response = "I'm sorry, I don't understand that question. You can try asking one of the following:\n\n"
                            # List all available questions
                            for q in qa_database.keys():
                                default_response += f"- {q}\n"
                            send_message(sender_id, default_response)
        return "Message Processed", 200

def verify_fb_token(token_sent):
    if token_sent == VERIFY_TOKEN:
        return request.args.get("hub.challenge")
    return "Invalid verification token"

def send_message(recipient_id, text):
    """Send message back to Facebook Messenger user"""
    url = "https://graph.facebook.com/v21.0/me/messages"
    params = {"access_token": PAGE_ACCESS_TOKEN}
    headers = {"Content-Type": "application/json"}
    data = {
        "recipient": {"id": recipient_id},
        "message": {"text": text}
    }
    response = requests.post(url, params=params, headers=headers, json=data)
    if response.status_code != 200:
        print(f"Error sending message: {response.text}")

if __name__ == "__main__":
    app.run(port=5000)

press the link bellow and you will see Messenger bot is running!

Now open a new terminal and type:

ngrok http 5000

You will get a link like:

https://qwerty123.ngrok-free.app/

Paste the URL into Meta App Dashboard → Messenger → Webhooks like this:

https://qwerty123.ngrok-free.app/webhook

then press Verify and press button.

⚠️ If you’re using the free version of ngrok, the link changes every time you restart it. That means you’ll need to update the Callback URL in your Meta App every time you run your bot.

💡 Tip: For testing, I saved all my tokens in a separate file config.py and imported them in my bot script:

# config.py
VERIFY_TOKEN = "my_secret_token"
PAGE_ACCESS_TOKEN = "your_page_access_token_here"

Then in your main bot code:

from config import VERIFY_TOKEN, PAGE_ACCESS_TOKEN

This keeps your tokens out of the main script, making it safer and easier to manage.

Key Takeaways:

Small setup mistakes can cost days.
Planning is crucial.
Keep your tokens safe.
Use ngrok for local testing.
Double-check your app types.

How I Built My First Telegram Bot (and Why Small Steps Matter)

Marika Tchavtchavadze — Mon, 08 Sep 2025 16:14:12 +0000

I used to think I needed to do everything perfectly before I even started. I would plan endlessly, imagine the ideal architecture, and hesitate because “it’s not ready.” Meanwhile, others had already started building, learning, and shipping.

Now I realize: it’s not necessary to be perfect or pro to begin. I just started—small steps, imperfect, but real. This Telegram bot is my first deployed project, a step on my journey.

What matters is progress, not perfection. I plan, I start, I learn, and then I improve. Each small step builds confidence and experience, bringing me closer to bigger goals. I finally understand the power of doing over idealizing. The journey itself is the reward.

And here is my very first idea delivered as a product. My first Telegram bot may be small, but it was a perfect first step. I set it up with Python, hosted it on Heroku, and tested simple commands. Seeing it work live for the first time was incredibly rewarding—it turned theory into real experience. This bot is far from perfect, but it taught me more than planning ever could, and now I have a foundation to build more advanced bots for businesses.

Feel that vibe? Share your story or drop a thought—I’d love to hear from you!
And don’t forget to try my bot out: KiwiCafeBot

Greetings DEV Community

Marika Tchavtchavadze — Tue, 02 Sep 2025 13:44:18 +0000

Hi, I am Marika. I am a mathematician with hands on experience in python and machine learning algorithms. Right now, I am experimenting with building AI bots. I’ve already created some basic ones and want to dive into deeper into the field.

I am here, because I want to

Share my journey and thoughts about tech & AI
Make connections with others in the community
Interact with peers and get inspired by new ideas

I’m excited to start sharing and growing together with this community