Everywhere you look, headlines scream that AI is going to change the world overnight. Yet in practice, most organizations are still in the experimentation phase, and that’s not going to change soon. If AI really is the biggest shift since the internet, why don’t we see the impact everywhere already?
The short answer: because this is exactly how it’s supposed to look. Transformative technologies never arrive in one big bang. They crawl in through trial, error, and iteration. Electricity took decades to rewire factories (source). The internet took decades to reshape commerce. AI is walking the same path.
This slow pace of AI adoption is a feeling born from high expectations and enthusiasm, coupled with a low understanding of the history of technological revolutions. Let’s dive into that.
Why it feels slow
We have heard that AI is so slow to spread, but compared to what exactly? I think we have to look at history for other technological revolutions and compare how quickly they transformed the world. Spoiler alert, it was much, much slower. I believe that after looking at previous technological advancements, you will realize that the feeling that AI adoption is so slow is just a feeling. AI is probably the fastest technological revolution in history.
The historical precedents & The Productivity Paradox
When factories first got electric power in the 1890s, productivity barely budged for decades. Economists were baffled, calling it the “productivity paradox.” The reason was simple: factory owners were using a new power source to run an old system. They just replaced one big steam engine with one big electric motor, keeping the same inefficient layout of belts and pulleys.
The real revolution arrived a generation later, when factories were completely redesigned around the unique capabilities of electricity: small, distributed motors driving the modern assembly line. This transformation required rethinking the entire process of manufacturing itself.
We are in the “swapping motors” phase of AI today. We are using it to do old tasks a little better. This is a necessary learning period. The real transformation will begin when we start redesigning our businesses around the capabilities of AI. I don’t think we will have to wait a generation this time.
The Revolution is Being Built, Not Delayed
The messy, thoughtful period we are in now is not a sign of failure. It is the essential groundwork. Technology adoption follows an exponential curve that seems flat for a long time before it suddenly skyrockets. We are on that flat part of the curve, building the infrastructure, skills, and governance needed for the explosion to come.
The AI revolution does not need to be rushed. It needs to be understood. The transformation is not stalled; it is being built, one methodical step at a time. This is a moment not for impatience, but for strategic foresight.
The Invisible Revolution
This feeling of slowness also comes because AI is kind of an invisible technology. Past revolutions were physically obvious. The steam engine announced itself with great machines of iron and smoke. The internet filled our offices with beige computer towers. AI is different. It runs on the same servers and cables as the old internet, looks like any other app, and appears as just another piece of code.
The consequence is that we feel that the AI revolution is slow because we don’t realize AI is already everywhere. It powers your Amazon product recommendations and is designing the next cancer treatments. Generative AI isn’t just a chatbot like ChatGPT; it can also perform topological optimization to build lighter planes that consume significantly less fuel. It’s already optimizing global supply chains and reducing CO2 emissions.
When we talk about AI, we often think of ChatGPT because that’s what most people can see. But AI’s biggest impact comes from improving industrial processes, creating better-designed products, and enhancing our lives in ways we don’t even notice.
Human frictions & what really slows down AI adoption
Any new technology must overcome the friction of human nature. The challenges we see with AI adoption aren’t new; they are the same patterns of inertia, trust, and misuse that have accompanied every major technological shift.
The Inertia of “Good Enough”
Humans are creatures of habit. We resist change not because we are lazy, but because established processes are predictable. The mindset of “if it ain’t broke, don’t fix it” often means “if it’s not that bad, why improve?” This inertia is a powerful force. We can see this in city administrations that still run on paper forms, decades after computers became ubiquitous.
This isn’t because paper is superior; it’s because the existing system works, and the activation energy required to implement a new one is enormous. AI must not just be better than current solutions but so much better to overcome this fundamental resistance to change.
Many Are Using the Tool Wrong
Another reason for the perceived slowness in AI advancement is that I often hear people complain that AI makes mistakes or isn’t good enough at this point. But then I ask them how they use it and what they expect from it. Then it all makes sense. I think people expect a perfect, magical tool that can do anything. The truth is, AI is good at some things and bad at others. Most often, it can do 75% of the job, and you have to do the rest yourself. Some people don’t think that’s good enough. I look at that and think my productivity has increased 3x.
AI isn’t a mind reader; it’s a partner in a process. It excels when you treat it as such, providing context and clarification step-by-step. Its real strength isn’t generating novelty from a vacuum, but helping where the context is already rich. Complaining that AI is “bad” after giving it a poor prompt is like telling an architect to “build a nice house” without providing any blueprints or preferences and then being disappointed with the result. We must learn how to use the tool for what it’s actually designed to do: generate solutions based on context.
For example, AI works poorly if you ask it to generate a blog article on a subject from scratch. However, provide an outline, some links to sources you researched, your own notes, and examples of your style. Suddenly, it can produce something much closer to your expectations.
Note that you still had to do part of the job; the creative part. This is where AI fails at the moment. It’s really good at re-arranging data into something coherent. Producing a blog article from notes and sources, generating a summary, or producing documentation based on code. But AI sucks at being creative, which leaves us with the interesting part of the work.
The Problem of Trusting a Black Box
Trust presents another significant hurdle. People are rightfully skeptical of “black box” systems. When a tool’s decision-making process remains opaque, we can’t fully rely on its output, especially in high-stakes fields like finance, law, or medicine, where explainability is essential. Thus, slowing down adoption in some fields.
This creates a critical issue for modern generative AI. While it can generate fluent and plausible explanations for its conclusions, we can’t audit its internal reasoning. As I explained in my article “The thermometer is not the temperature,” a measure of performance isn’t the same as the performance itself. An AI can achieve high accuracy on a test dataset for all the wrong reasons.
Yes, the current generation of generative AI is capable of “thinking”. They generate an inner dialogue before producing a response. However, while this technique works really well, we can’t be sure that the final answer directly follows from this thinking process.
Without a transparent reasoning process, we cannot verify that the AI produced the right response. This makes current AI excellent for non-critical administrative tasks like assisting in the writing of reports or coding. However, we still cannot trust them to act on their own reliably as a human would. Would we want that anyway?
Why it’s actually fast
In medicine AlphaFold from DeepMind. It also uses the same technology as ChatGPT to fold proteins. Before AlphaFold, it took years to understand a protein; now we analyze thousands per day. The consequence is that we are able to build better medicine much faster than ever before. For example, USAN is a new experimental drug that has been designed with AI, and it shows promise for the treatment of some lung diseases.
Other examples include AI designing new rocket engines in a few weeks instead of months. AI is also redesigning plane wings, which are capable of reducing fuel consumption 100s of tonnes per year per plane. Finally, Amazon deployed DeepFleet, a generative AI model able to control thousands of warehouse robots at a faster speed.
In my personal experience, I have seen great productivity improvements with AI. For example with Cursor, a tool for software development using an AI agent. It helped me build tens of pages of documentation in a few days instead of weeks. Recently, I have built & published a plugin for WordPress in 2 days. It would have taken me a few weeks without Cursor. The AI revolution is here, and it’s already impacting many billion-dollar industries.
Is the AI revolution slow? It’s not
In some ways, this article is in response to the infamous report from MIT that states that 95% of AI Projects Fail. This study made quite some noise. However, many have pointed out that the report is flawed and the authors are just trying to sell their own AI solution. I suggest you read this article, which has done a great analysis of this report.
In pratice, AI is already providing great return on investment, and it’s only the beginning. One comprehensive report by Google from 2024 shows “74% of enterprises using gen AI report ROI within the first year, with 86% of those reporting increased revenue, noting a 6% or more increase.” This study is quite extensive. I suggest you read it if you are interested.
This article is also a response to the article from The Economist: “Why is AI so slow to spread? Economics can explain”. Which, while interesting, only scratches the surface of the issue, and I wanted to dive deeper into the issue.
The revolution is taking place at its own pace. Just like any other. The productivity paradox of the industrial revolution and the dot-com crash of yesteryear show us that technological revolutions don’t happen in an instant or even smoothly. Humans are creatures of habit, and we still have only scratched the surface of what we can do with AI.
Should we be disappointed by the speed of change? No, we should learn to expect it. Will the stock market crash like it did with the dot-com bubble? Possibly yes. But the AI revolution is coming, and in 10-20 years, all of our medicine and all of our software will be built by AI. The AI revolution may come slower than you expect, but it’s coming faster than previous technological revolutions.
As an independent AI researcher/developer specialized in Natural Language Processing (NLP), I have a comprehensive expertise in the development and integration of AI systems, as well as data analysis.
Is your company looking to integrate AI solutions, analyze data, or strengthen its back-end development? Contact me!
Mail : pro.judicael.poumay@gmail.com
Linkedin : Judicaël Poumay
Portfolio : www.judicael-poumay.be
- Why is the AI revolution so slow? (It’s not) - 18 September 2025
- How Does LLMs Store Knowledge? A Deep Dive Into Feature Superposition - 15 August 2025
- How RLHF works for LLMs : A Deep Dive - 1 July 2025