Hey! Ankur here, and this is the 9th edition of Lazy AI — India’s only AI newsletter for non-tech folks — 5 mins of reading every day, to help you stay ahead of the AI curve.
Happy day for me — We’re already 220 subscribers in just 8 editions! Glad that people are resonating with the content 🙂
If you are liking it too, it would mean a lot if you could share this article with a friend.
Okay, today I’m going to talk about Gen AI, Agentic AI and AI agents. Most importantly, I’ll talk about why you must be careful with products that use these terms. Because most “agentic AI” softwares are just automation workflows that don’t actually need or use “AI”.
Don’t worry, I’ll get into the details.
But first, I want to make sure I write relevant content for you, so I’d like to understand what function you guys work in. I don’t want to end up writing about AI in marketing, when most of you are from finance, or vice versa 🤷🏻♂️
So…please tell me!
Now, let’s begin..
What is Gen AI?
Generative AI (Gen AI for short), is what we use everyday — ChatGPT, Claude, Grok and the likes. Basically these tools take prompts from you, and generate something. Could be text, pictures, videos or any other form of output.
Please note here, that Gen AI is not just about content creation. Any output that ChatGPT (or any other AI tool) gives you, in the form of text, is Generative AI.
So you ask it a question about how to travel in winters, and it gives you a response — that’s Generative AI.
Now, Agentic AI
This is where the AI stops being the talker and starts being the doer.
Generative AI takes your prompt and gives you a response. It’s a 2-way communication where you and the AI communicate with each other, one-by-one.
Agentic AI changes this.
Agentic AI is basically a capability of the AI software that
takes a goal from you
figures out how to achieve it, and
then executes the steps needed to achieve that goal.
The “AI Agent” is the system that executes the goal. Agentic AI is the capability of that system. They’re related terms and most people use them interchangeably.
Think of it as you telling an AI agent “I want to travel from Mumbai to Bangkok on 5th March 2026. Check Yatra, Ixigo and MMT to find flights, and book me the cheapest one-way ticket. Make sure it’s a direct flight. Also book me a hotel for 3 nights in Bangkok. Make sure that the hotel has good food, based on reviews from Tripadvisor. My credit card details are XXXX XXXX XXXX XXXX.”
The AI agent will crawl the 3 websites, find the cheapest price and book your flight + accommodation.
Of course, you can ask it to either run autonomously, or take your approval after each step — that’s your call. But it has the capability to run autonomously and reach the goal you have set for it.
However, an important thing to keep in mind — we are still in the very early days of AI, so a single prompt like the above will most likely not give you the desired result. Instead, you should break it down into steps and tell the agent to take your approval after crucial steps.
For example, “Before entering my credit card details, first show me the flight you are booking” can be a simple way to ensure that the agent is not hallucinating (or going bonkers) during the workflow.
And that’s how your AI agent will execute the steps and achieve the goal for you.
Essentially, while Generative AI will prepare your travel itinerary, Agentic AI will execute the itinerary for you. That’s the core difference. AI Agents mostly use Generative AI as well, as part of their workflow.
If this article added any value to you, do consider subscribing so that you keep receiving this newsletter regularly. It’s free, lasts 5 minutes, and keeps you ahead of the curve!
Now, a word of caution
We all know that AI is the “next big thing”. So you’ll see a lot of startups talking about “AI-based workflow systems”.
The reality though, is that most of these are just workflow automations, with no real “intelligence” capability.
You see, there are two kinds of workflows for achieving goals:
Where you simply automate the task without applying any intelligence. In the above example for travel bookings, the first part of the goal — about booking flight tickets — is where you don’t need intelligence. The software simply needs to scan websites, check prices and book the cheapest one. All of it is possible without using AI. With simple code. This isn’t an “AI workflow”. It’s just workflow automation that you can do if you know how to code with python. The second kind below, is the real AI agent.
Where you automate a task that requires complex reasoning/intelligence. This is where you actually need AI. In the above example, the second part of the goal — about booking hotels that have good food based on tripadvisor reviews — is where the AI agent will collect data from tripadvisor reviews, run an analysis on it, use its own intelligence and then determine whether the food is good or not. THIS is where you need AI agents. Because it needs intelligence.
The problem today is, a lot of startups actually just run workflow automations with python and call it an AI agent, just because they want to be seen as building something that’s trending right now.
Nothing wrong with workflow automations. They’re brilliant and save a lot of time. But don’t call them “AI based workflows” if they’re not actually using AI.
If this article added any value, do consider subscribing so that you keep receiving this newsletter regularly. It’s free, lasts 5 minutes, and keeps you ahead of the curve!
Alright, now that we spoke about Gen AI, AI agents and how to determine “true” agentic workflows, I’m sure you’ll be able to ask the right questions to determine whether a software is an AI agent or not.
Whether you’re in marketing, finance, ops, HR, product or any other function, while evaluating tools, make sure you know whether you’re paying for AI, or just for workflow automation. Because the former is expensive, while the latter can be bought at a fraction of the cost.
That’s it for today folks!
If this added any value, please like it and share it with a friend. It would mean a lot to me!
See you next time..
Cheers,
Ankur