The AI Paradox: Is AI Overhyped or Underhyped? The Answer is... Both, Simultaneously.
The Disconnect Between AI Buzz, Business Value & Ease of Execution
🔑 TL;DR (Key Takeaways)
AI is Underhyped in Its Potential
AI isn’t just another tech wave—it’s a foundational shift in what computers can do, not just how easily they do it. It’s enabling the once-impossible.AI is Overhyped in Execution Ease
While building demos is easier than ever, delivering production-grade AI systems that drive real business outcomes is still complex, iterative, and organization-wide.The Real Opportunity is Massive
From autonomous systems to AI-native education and databases, AI is enabling a whole new category of companies. Every existing business now faces an existential challenge—or opportunity.Most Leaders Underestimate AI
The biggest mistake? Thinking AI is like mobile, cloud, or the internet. It’s not a wave of convenience. It’s the start of a new epoch.
As an AI coach to CXOs and Boards, I get this question all the time:
“Some say AI is overhyped. Others insist it’s underhyped. What’s the reality?”
My answer? Both. Simultaneously. Let’s break this down.
AI Is Deeply Underhyped
If you listen to any conversation with the world’s most influential corporate leaders, startup founders, or iconic VCs—Satya Nadella, Sundar Pichai, Sam Altman, Elon Musk, Vinod Khosla, Paul Graham, Marc Andreessen, Ben Horowitz, Eric Schmidt, Andrew Ng etc—you’ll hear one thing loud and clear:
AI isn’t just another wave. It’s the next epoch.
They talk about the transformative power of AI—how it will change everything.
And they’re absolutely right.
But here’s the catch: Even their words fall short of capturing how profoundly AI is going to reshape our world.
Most leaders view AI as “just another technology”—like the internet, cloud, or mobile. That is the biggest mistake.
Those technologies made hard things easier:
The internet made it easy to connect remote machines and exchange information.
The cloud made data accessible from anywhere.
Mobile made connectivity seamless and constant.
Each of these was a wave of convenience. But none of them changed the fundamental capability of computing itself.
AI does. And thats where its true power lies.
Lets take an example - image segmentation.
🖼️ What Is Image Segmentation? (In Simple Words)
Imagine you’re looking at a photo of a street — with cars, people, buildings, trees, and the road.
Now imagine telling a computer:
“Pick out just the cars.” (blue color)
“Highlight only the people.” (red color)
“Separate the road from everything else.” (megenta color)
That’s image segmentation : It’s the problem of dividing an image into different parts, where each part corresponds to a specific object or region — like a person, a cat, the sky, or a tumor in a medical scan.
In technical terms:
Image segmentation is about labeling every single pixel in an image with a category (e.g. “this pixel is part of a dog,” “this pixel is background”).
Segmentation is one of the hardest problems in computer vision. 20 years ago, during my grad school days, I attended a talk by top computer vision researchers from Microsoft Research who proudly presented a breakthrough: After years of work, they had pushed image segmentation accuracy from the low 80s to the high 80s. That was then considered state-of-the-art.
Fast forward to today — and you have tools like Segment Anything from Meta, which can segment objects in the most complex scenes with near-perfect accuracy. And it's not just for researchers. Any developer can write a 4 line code and incorporate its power in their application.
Have you ever wanted to remove a specific object from an image — say a person in the background or an unwanted item? Today, you can literally draw a rough outline with your finger, and poof — it’s gone.
That magic? It’s powered by image segmentation.
And just five years ago, this would’ve been flat-out impossible.
Object removal in your iPhone or android is powered by Image Segmentation
This is a tiny example of something bigger:
AI isn’t just making hard things easy — it’s making impossible things possible.
It’s now widely believed that from climate to crime, space to sports, crops to culture, there’s no area of human life that won’t be touched — or transformed — by AI.
Even the best minds at the frontier of AI can’t fully express the scale of what’s AI can do for mankind.
That’s the part that’s deeply underhyped.
What does this mean for Businesses, CXOs, Startups & VCs
Owing to the crazy advances in AI, suddenly, one can solve a whole new class of problems - from self driving to AI powered financial decisions. This was not possible earlier.
This means - 3 smart kids can create the next AI powered urban transportation company taking on the likes of Uber. AI powered classrooms & material challenegeing the teaching metheds of best in class ivy league colleges. AI powered intelligent Database that can give the might oracle a run for its money. In short - the next unicorns, decacorns, business tycons will all come from the AI era. This is a threat to every existing business. This is why AI is no more an option for any individual, business, multinational, even countries.
We wrote a seperate article on this:
So far we have understood the underhyped part. What is the over hyped part?AI is highly overhyped
So what is overhyped? The ease with which one can get to a production-grade AI system that delivers business outcomes.
There are no two questions that in the last 2.5 years, since the ChatGPT moment, AI has made drastic progress:
Time, cost, and difficulty of building AI systems has come down drastically.
Quality of output has gone up drastically.
Ease of incorporating AI into products is better than ever before.
To build PoCs (Proof of Concepts), you don’t even need full-fledged AI teams — a small business leader, a smart product manager, and 1-2 engineers can be enough.
To make AI development and deployment easier, there are a host of tools.
Despite all this — building, deploying & operationalizing a production-grade AI system that delivers business outcomes remains a significant challenge.
It is nowhere close to the ease with which one can build traditional software. It is far from the "plug-n-play" state we have reached in the realm of traditional software development.
Even today, building, deploying & operationalizing a production-grade AI systems that deliver business outcomes is:
Fairly complex
Time-consuming
Full of experimentation and iteration
Requires organization-wide alignment
If you found this useful & you are using it anywhere, please cite this write-up as:
Gupta, Anuj. (June 2025). The AI Paradox: Is AI Overhyped or Underhyped?The Answer is... Both, Simultaneously. anujgupta.co https://pragmaticai1.substack.com/p/what-sam-altmans-role-in-openais
or
@article{gupta2025OpenAI-without-sam,
title = {The AI Paradox: Is AI Overhyped or Underhyped? The Answer
is... Both, Simultaneously},
author = {Gupta, Anuj},
journal = {anujgupta.co},
year = {2025},
month = {June},
url = {https://pragmaticai1.substack.com/p/the-ai-paradox-is-ai-
overhyped-or}