A lot is changing with AI. It’s been confusing, and slightly overwhelming, for me to get a grip on what’s changing, and what it all means for me personally, for my job, and for humanity. What’s going to happen in the next 5 years? Will my skills be relevant? How do I truly add value with AI getting smarter in every way? How does it change life for me and my family?
It doesn’t fit into any of my existing mental models. I use it in every aspect of life already, and the inability to understand it in a similar way to anything in the past makes me feel unsettled. So, I’m writing to break it down, and create a mental model that does work. For myself. And maybe for others out there dealing with all of this too.
It’s a new General Purpose Technology (GPT)
Andrew Ng popularised “AI is the new electricity”. It’s not just any random analogy about a future possibility, it’s a well researched reality. It’s useful to understand what this really means.
When science results in a breakthrough like the steam engine, or electricity, or the internet, or the transformer, it’s called a paradigm shift, or a scientific revolution. Thomas Kuhn defines and explains this well, in The Structure of Scientific Revolutions, 1962. And the paradigm shifts in AI as a science, have been studied via Kuhn’s framework in 2012, and 2023.
However, a paradigm shift in science also causes paradigm shifts in engineering. Engineering rethinks its assumptions, methods, and goals. It can both use the new technology, and build the next order of technology on top of it, causing the ripple effect that leads to the birth of entire industries and economies.
While Kuhn doesn’t go into it, the technological diffusion, and the economic impact of scientific revolutions are better studied through the GPT (general purpose technology) paper from Bresnahan & Trajtenberg in 1995.
“General Purpose Technologies (GPTs) are technologies that can affect an entire economy (usually at a national or global level). They have the potential for pervasive use in a wide range of sectors and, as they improve, they contribute to overall productivity growth.”
And Calvino et al in June 2025, finds that AI meets the key criteria of a General Purpose Technology. It’s pervasive, rapidly improving, enables new products, services and research methodologies, and enhances other sectors’ R&D and productivity.
Here’s a sample of the ripple effect with GPTs, in simple terms. The diffusion lag from one column to the next in number of years, is listed below it. The table is an approximation to give you an idea, and in reality tracking all this is fairly fuzzy.
| Technology | 1st Order Effects | 2nd Order Effects |
|---|---|---|
| Steam Engine | Trains, steamships 40-60 years | Travel across countries, factory jobs 10-30 years |
| Electricity | Light bulbs, electric motors 5-15 years | Home lighting, appliances, 24-hour cities 15-30 years |
| Semiconductors | Microchips, personal computers 10 years | Smartphones, digital life, automation 20-40 years |
| Internet | Web browsers, Wi-Fi 10 years | Online shopping, social media, remote work 5-15 years |
| AI (Transformer) | LLMs, vision models, GPU boom, Fine tuning, Compression 2-4 years | Customer service automation, Agentic coding, Content creation 3-6 years |
You can see the diffusion (another overloaded term, sorry) lags reduce over time, and with AI the diffusion lag is very low because a part of the distribution infrastructure, the internet, already exists, and is mature. There are still major gaps, for example with inference and compute scaling, that will take years to stabilise.
Still, it’s the fastest GPT diffusion ever, we’re in the middle of it, and it’s changing the world around us. And the sheer scale of it is staggering. 10s of thousands of researchers and engineers across the world, burning billions of dollars with governments and mega-corps, to make progress happen. It is remarkable. It’s like the world coming up with covid vaccines all over again, every few weeks.
It’s important to understand that rapid progress in science and technology alone isn’t enough to speed up economic diffusion. It’s a two-way street where R&D in AI labs has to drive economic growth, for the economic growth to invest more into R&D in AI labs. The state of the economy, the attitude of world leaders towards AI, can slow down, or speed up progress.
It’s not just another GPT though. It’s more.
It’s a paradigm shift in accessing knowledge
Languages, Writing, Printing, Broadcasting, Internet.
These are all GPTs too, but of a different, more radical kind. Each of these have been a revolution in sharing knowledge. And now, AI is yet another revolution, one that makes all knowledge intuitively accessible. Knowledge access diffuses more pervasively and quickly compared to other technology.
The internet had these aspects too, especially evident in the Solow Paradox.
“You can see the computer age everywhere but in the productivity statistics.”
WSJ wrote about it recently. Andrej Karpathy wrote about this too, in his power to the people article. The revolution with AI is apparent in the speed that AI took over google search as the way to access the internet. But the implications of accessible knowledge are far beyond replacing search engines.
“Why do I need to know the capital city of Cambodia, I can just Google it” use to be a popular phrase in my school. General knowledge, we called it, and that skill became obsolete with the internet and search engines. With AI, we’ve transcended that with… “Why do I need to know what quarks are, or how steam engines work? Why do I need to know anything about anything? I can just look it up with AI”. I’m kidding… but only partly. I’m confused about the implications of that question because I do have an occasional existential question of that sort with AI. Does this mean we learn more or less? Does this mean we can be curious about more things now, or does it mean our curiosity will reduce because we’re satisfied with the knowledge we get from AI?
Education for my son will look wildly different from mine. He can have a personal tutor with longitudinal awareness, one that can embody a Richard Feynman, or Walter Lewin, or Grant Sanderson over voice or video, teach hard concepts, and answer silly, or complex questions untiringly.
AI enabled me to transition to a new country quickly and comfortably. I can stand before an aisle of cough medicines, or shampoos, or groceries in Toronto, and get suggestions similar to what I’m used to in Bangalore. I’ve understood the healthcare, and school systems here through AI. I even use it to troubleshoot my dryer, my dishwasher, a 3-dial-lock, and learned to operate an oven for the first time. I didn’t use any of these things before in my life.
There’s an important nuance in the stochastic nature of AI. We need to understand AI’s emergent psychology of gullibility, hallucination, jagged intelligence, anterograde amnesia, etc. And then we need to get good at context engineering the same way we got good at searching the internet with keywords, to use it well.
Further, AI makes knowledge accessible not just to tech savvy users, but to children, elders, and other software, and AI too. The ripples are coming.
It’s not just the new internet though.
It’s a paradigm shift in thinking
Thinking, or reasoning, is a defining aspect of humans. It’s the skill that differentiates us from animals. Immanuel Kant describes it as:
Revolution der Denkart
revolution of the way of thinking
Through our history, we’ve had various landmarks in the centuries of evolution of human thinking abilities. And each of these has been a revolution that impacts our defining skill as a species.
- Thinking by questioning (Socrates) Thinking by logic (Aristotle) Thinking by observing: Empiricism / Scientific Method (Bacon) Thinking by heuristics and biases: System 1 and 2 (Kahneman) Thinking by… augmentation? (AI)
Even without AGI, AI is changing our way of thinking already.
Everyone has their personal, tireless, emotionless, brain power augmenter. AI can read, write, see, hear, and speak, and combined with some common sense, it can truly understand and interpret these signals.
I think this is the weirdest part of AI that’s difficult to understand and comprehend for us all.
Thinking to me was… sitting with my thoughts, alone. Reading up, writing down my thoughts and reflecting on them. Sharing my thoughts with others, getting their thoughts on it, and then assimilating my own point of view. Thinking is… following a trail of thought to its logical conclusion.
I don’t think quite like this anymore.
I chat with a voice AI, ramble on for a while, and have it summarise my thoughts back to me. I write to the AI, and have it reflect back to me, instantly. I can invoke the internet spirits of my favourite famous people (Rich Hickey and Andrej Karpathy these days), and chat with them. Standing on the shoulders of giants has never been easier. It is overwhelming, and I can’t think and comprehend fast enough. It’s a good problem to have though. I’m the bottleneck, as I often am, with AI.
It’s not just me, I’m sure. Most people, like me, already use it to break problems down, to brainstorm, and to create decision matrices for software (or civil?) architecture decisions. Even governments, presidents and prime ministers use it to shape policies and laws. Some people use it as therapists, and some even have emotional AI companions. We have an extension of our brains that’s an expert in nuclear physics, or quantum mechanics, or computational music. We can consult anytime.
So for software developers, it’s not just a small paradigm shift like changing from C to Lisp, or from object-oriented to functional programming. It’s Software 3.0 where we create software by using AI, and we’re also building the next order of technology in the GPT’s diffusion. We need to build the railroads for the new steam engines, the light bulbs for the new electricity.
Further, software can use AI’s ability to think and access knowledge, just like humans can. Oh, and AI can use software too. And oh, AI IS software too.
Trippy.
So, it’s not just a paradigm shift in thinking.
It’s a recursive paradigm shifting paradigm shift
Code is data is code. Lispers get this, its turtles all the way down. Let’s walk through it.
- Software can use AI. We write pieces of text in between code that uses AI’s thinking or knowledge access capability. Like with autocompletion, or chatbots. AI can use software. It can access the filesystem and run programs on your OS, if you let it. It can access the web through search engines, or web browsers, if you let it. AI can build software. It writes and executes small python scripts to analyse data when thinking. The agency and autonomy needed to build full fledged meaningful software isn’t there with AI yet, but it’s advancing quickly. AI assisted coding is pretty big, you know this. AI IS software. AI is a trained neural network, and with sufficiently advanced capabilities, it can build itself.
AGI isn’t here yet, and AI isn’t building itself without humans. But, AI is already fairly bootstrapped. Kimi K2 is LLMs all the way down. This is one of the reasons the technology is evolving really quickly. AI, Software, and Humans, are together enhanced by AI.
Humans, software, and AI… they each build and use each other, and themselves. Well okay, no one is building humans (yet). But humans build and use Software and AI. Software builds and uses AI. AI builds and uses Software. Arguably, Software and AI use humans too. There’s an autonomy slider on each edge here, and different kinds of tasks have different levels of autonomy.
And history doesn’t quite help us here, maybe fiction does. But yeah, we’re in fairly uncharted territory. If and when AGI does come, all bets are off.
Thrilling.
Embracing change
AI represents a multi-dimensional revolution. One that impacts our defining human ability of thinking. One with the potential to transcend us.
Thinking of it as an evolution of existing paradigms helps in appreciating the similarities with the familiar, and differentiating what’s new. Most people comparing AI with the internet or the industrial revolution are doing the same thing.
We need to accept that we’re in the middle of a revolution and that things will be confusing for a while. If Kuhn is right, at some point, we’ll hit a plateau in science, and that will bring some stability to the world. Until then though, this magnitude of change will be the norm.
And change is hard. Paradigm changes are harder. Depending on whether your product is going through an existential crisis, or you’re knee deep in researching or applying AI, or you’re just entering the software industry, your mileage may vary on how you experience this phenomenon. But, you’re likely swimming through a change wave right now.
For those of us in tech, the most imminent, and significant change wave, is the industrial revolution of Software itself. Software is becoming more probabilistic than deterministic. How we hire, what skills we value, who we fire, how we organise product teams, who we write software for, and what constraints we optimise for… are all changing already. I don’t think I’ll recognise the software industry in another 5 years.
What are the light-bulbs or the railroads this time around? If the GPT is about thinking, what are its 2nd and 3rd order effects, and what is our place in them?
What is your role in this revolution?
Footnotes
Because of the enigmatically stochastic nature of the AI, it’s not an exact science to apply or evolve it. So, it gives everyone (not just scientists) the opportunity to unlock its benefits. The effectiveness of LLMs is evident and relevant to science mostly because engineering (and everyone else too) tries to exploit it every way possible.
- This is the reason interpretability is big, it allows us to look behind the curtain.
Here are some examples that should get you in the mind space of a paradigm shift:
Philosophic shifts
- Ptolemaic (earth at centre) to Copernican cosmology (sun at centre) Newtonian gravity (inverse square law) to Einstein’s general relativity (mc2)
Operational shifts in programming
- Object oriented programming to Functional programming C family of languages to Lisp family of languages
I’m grateful to Udit, Atharva and Deepa, who helped me think this through, and reviewed the writing too.
