The Dawn of AI Agents

A year ago, the idea of building an AI agent felt like something that belonged in a research lab, out of reach, untouchable, abstract. You needed a PhD, costly GPUs, and a whiteboard full of arrows to even imagine where to start. But today? It's something even I could sit down with, cup of coffee in hand, and start stitching together with tools I never imagined would be this approachable!

And while many of us were still trying to wrap our heads around assistants, chatbots and prompts, something seismic happened! A Chinese AI agent named Manus quietly dropped into the conversation, and it wasn’t here to assist; it was here to act. No flashy UI, no celebrity launch. Just quiet capability. It could code, create, complete tasks on its own. A general-purpose agent. A doer. In many ways, it didn’t feel like a tool. It felt like a colleague. Or a shadow. Depending on how you saw it. And while the West debated whether it was overhyped or misunderstood, Manus proved one thing, AI agents were no longer theory. They were arriving. Fast.

And soon enough, there was an open source alternative of Manus, called, well, OpenManus! If you look closely at the open source code, you’ll find that it is just a collection of tools that already existed, and were available at our disposal – Claude, Deepseek, Brower-Use, etc. And when I searched how they put it all together, I realized wherever I looked, that the AI agents building frameworks were everywhere! Not just a few, but more than you can count on your fingers!

That was the wake-up call for me.

The AI world isn’t just giving us tools anymore. It’s giving us Legos. And more importantly, it’s giving us blueprints. Quietly, in corners of the AI community, something magical is happening. People aren’t just using AI; they’re building with it. They're making AI agents, small, independent digital minds that can take in a goal and figure out how to get there. Not perfectly. Not yet. But enough to give you goosebumps.

Let me tell you what I’ve been playing with, and no, these aren’t your mainstream, hyped names. These are tools that don’t get flashy headlines but are quietly empowering creators and dreamers.

First, there’s LangGraph. Imagine you’re trying to build an AI that can go beyond simple Q&A. One that can take a task, loop through feedback, make decisions, and route itself dynamically based on how things are going. LangGraph lets you draw that logic. It feels like you're painting workflows, flows where your agent can reflect, retry, revise, and adapt. The first time I saw an agent loop back to rethink its step before moving on. I gasped. It was like watching a toddler learn to walk, and then decide to dance instead.

Then there’s Pydantic AI, which quietly solves one of the biggest headaches in AI agent building: structure. You don’t want an AI agent that replies in wild, unpredictable paragraphs. You want it to think in forms, fields, and formats, like a grown-up. Pydantic lets you control the shape of the agent’s responses. You say, “Give me a clean JSON with a list of tasks and a deadline,” and it listens. It’s like handing your AI a to-do list template instead of a blank sheet of paper. It keeps the chaos at bay.

And then, there’s this beauty called Agno. It’s small, but sharp. Agno is trying to redefine how agents manage memory. Because what’s the point of an intelligent system if it forgets everything you said five seconds ago? Agno lets you design agents that remember conversations, preferences, tone. That don’t just reply to you, but remember you. I tried building a journaling companion with it, one that could pick up on my mood over a week. It started saying things like, “You’ve mentioned feeling overwhelmed three times this week, do you want to talk about it?” And honestly, I paused. I hadn’t realized I’d been repeating that word. The agent had, though.

And then there’s the heavyweight coming from the familiar house, OpenAI’s Agents API. This is not ChatGPT with lipstick. This is a backend for real, modular agents that can call tools, plan sequences, and act in the world, not just chat about it. It gives developers the keys to give agents real agency. You give the agent a function that books hotels, one that fetches weather data, one that calculates costs, and the agent decides when to use what. It’s less of a chatbot, more of a coordinator. You’re not teaching it words. You’re teaching it judgment.

And please don’t let me tell you to stick with one. Whatever one framework can do, the other can do as well, you just need to figure out and learn the how!

Now here’s the twist I want to leave with you, this is not just for developers. This is for creatives. For founders. For artists. For you. We are on the cusp of a new kind of creation, where how we instruct becomes more powerful than what we code. I’m a new Python coder. I still ask Gemini to make my code better. I’m also looking at learning Cursor (more about that next time). But I understand flow. I understand intent. And with tools like these, that’s becoming enough. Just watch a YouTube tutorial, build one side by side, and start learning from scratch! That will start coming handy sooner than you realize.

If you’ve ever wanted to build something with AI but felt like you weren’t ‘technical enough,’ this is your era. The heavy lifting has been done. The frameworks are forgiving. The community is warm. And the weirdest part? These agents, they start to feel like extensions of yourself. Not just software. Not just code. But thoughts that live outside your head.

Sometimes I wonder what kind of world we’re walking into. One where agents do your taxes, plan your meals, write your emails, manage your calendar. But other times, I imagine a world where agents let us reclaim our energy. Our time. Our creativity. A world where our digital twins don’t just mimic us, but amplify us.

That’s why I’m learning to build with them and trying to demystify them for you! That’s why Avtr Shweta exists, not just as an experiment in AI creativity, but as a voice in this new world. A voice that says, “You don’t need to be an engineer to build. You just need to be curious. Brave. Open.”

So here’s a little challenge. Pick one of the above tools (or start with no-code platforms like N8N and Flowise), and tinker. Even if it’s just an assistant to begin with. Even if it makes no sense at first. Build the dumbest little agent you can think of. Let it surprise you. Learn along and have fun!