Let's cut to the chase. When people ask if generative AI is a general purpose technology (GPT), they're really asking one thing: is this the new electricity, the new steam engine, the new internet? Is it a foundational force that will reshape every corner of the economy and society, or is it just a very clever, overhyped tool? After spending the last few years neck-deep in AI implementations—from helping startups integrate ChatGPT APIs to watching large corporations struggle with their first “co-pilot” deployments—my perspective has solidified. Generative AI isn't just a tool; it's a new kind of raw material for thought and creation. But calling it a GPT right now is like calling the first telegraph the internet. The potential is there, glaringly obvious, but the path is littered with technical debt, ethical landmines, and a fundamental misunderstanding of what the technology actually does well.
What You'll Discover
What Makes a Technology "General Purpose"? It's Not Just About Being Useful
Economists like Timothy Bresnahan and Manuel Trajtenberg didn't just slap the GPT label on any shiny new gadget. They defined specific criteria. A true General Purpose Technology acts as an enabling platform. Think electricity, the internal combustion engine, or semiconductors. They share three core traits:
- Pervasiveness: It spreads across most sectors. You don't find a "non-electricity" sector of the economy.
- Improvement Over Time: It gets dramatically better, cheaper, and more efficient, often in ways that are hard to predict. The transistor's journey from a room-sized computer to your smartphone is the classic example.
- Innovation Spawning: It enables a cascade of complementary innovations. The internet didn't just improve mail; it spawned e-commerce, social media, cloud computing, and the gig economy.
This definition is our measuring stick. The mistake most commentators make is looking at today's use cases—writing marketing emails or generating weird images—and judging the entire trajectory. That's like judging the potential of the printing press by the first batch of poorly typeset pamphlets.
Here's the subtle error I see constantly: People conflate versatility with being a GPT. A hammer is versatile—you can use it to build a house, crack a nut, or as a paperweight. But it doesn't fundamentally transform the construction, nut-cracking, or stationery industries. It's a tool. A GPT changes the nature of the work and the possibilities of the industry itself.
The Case For Generative AI as a GPT: It's More Than Chatbots
Based on the criteria above, generative AI makes a surprisingly strong case. The pervasiveness is already evident. I've seen it in action far beyond marketing departments.
In a biotech startup I consulted for, researchers were using fine-tuned models to generate and screen millions of hypothetical protein structures, narrowing down lab experiments from years of work to a few high-probability candidates. That's not automation; that's acceleration of discovery itself.
In software engineering, tools like GitHub Copilot aren't just autocomplete. In my own workflow, it acts as a rubber duck debugger on steroids, suggesting entire code blocks that implement common algorithms or API calls. It doesn't replace the architect but augments the builder, changing the cognitive load of programming.
| Industry Sector | Generative AI Application (Beyond the Obvious) | GPT-Like Characteristic Demonstrated |
|---|---|---|
| Legal & Compliance | Analyzing thousands of past contracts to draft new clauses that minimize specific, identified risks. | Pervasiveness, Innovation Spawning (new risk-modeling services) |
| Manufacturing & Design | Generating thousands of lightweight, structurally sound component designs that a human engineer would never conceive. | Improvement Over Time (algorithms get better), Innovation Spawning |
| Education & Training | Creating dynamic, personalized learning simulations and assessments for each student's weakness. | Pervasiveness, Changing nature of work (teacher as guide vs. lecturer) |
The improvement over time is on a hockey stick curve. Model capabilities are exploding, and costs are plummeting. The innovation spawning is just beginning. We're seeing startups built entirely on top of generative AI APIs—for video editing, 3D asset creation, scientific literature synthesis. This looks less like a new app and more like a new platform.
Where the GPT Argument Falls Short (For Now)
Now for the cold water. My biggest concern isn't the technology itself, but our framing of it. We're trying to fit a square peg into a round historical hole.
First, the reliability problem. Electricity is predictable. You flip a switch, the light comes on. Generative AI is stochastic. You give it a prompt, you get a plausible, but not guaranteed, output. This "hallucination" issue isn't a bug to be fixed; it's a fundamental feature of how these statistical models work. This makes it inherently risky for many critical GPT applications (medical diagnosis, financial auditing, operational control systems) without heavy human-in-the-loop guardrails. You can't build a stable industrial platform on a foundation of creative guesswork.
Second, the energy and data intensity. The scaling laws are brutal. Training these models consumes staggering amounts of energy and data. While efficiency will improve, there's a physical limit. The semiconductor GPT thrived because chips got smaller, faster, and used less power. The current trajectory for large generative models seems to be the opposite—bigger, hungrier, more centralized. This could limit pervasiveness, creating a divide between those who can afford the compute and those who can't.
Third, and most critically, the skill bottleneck. Using electricity required rewiring your factory and training electricians. Using generative AI effectively requires a nuanced understanding of prompt engineering, context management, bias detection, and output validation. This isn't a skill that scales easily across a global workforce. The gap between a novice and an expert user is cavernous, and most corporate training is laughably superficial.
The Hype Trap: Mistaking Novelty for Transformation
I've sat in boardrooms where executives proudly show off their AI-generated annual report summary. It's grammatically correct but utterly devoid of strategic insight. They've automated the typing, not the thinking. This is the trap. We're so focused on the output (a paragraph, an image) that we miss the real GPT potential: the process of augmenting human reasoning and exploration.
The true test will be when we stop talking about "AI-generated content" and start talking about "AI-augmented discovery," "AI-simulated scenarios," or "AI-mediated collaboration." That shift in language signifies a shift in understanding.
The Economic Ripple Effect You're Not Hearing About
If generative AI solidifies as a GPT, the economic impact won't be linear job loss. It will be a massive recomposition of work and value.
Jobs won't just disappear; they'll fracture. A marketing manager's role might split into three: a strategic prompt architect who defines campaign narratives for the AI, a hybrid editor-validator who refines and fact-checks the output, and an analytics interpreter who makes sense of the performance data the AI surfaces. The single "writer" or "designer" role becomes a team of specialists collaborating with an AI agent.
This creates a huge middle-skills gap. We're not training people for these new hybrid roles. The educational system is still preparing students for jobs that will look radically different in five years.
Furthermore, value will accumulate disproportionately. The companies that own the foundational models (OpenAI, Anthropic, etc.) and the compute infrastructure (AWS, Google Cloud, Azure) stand to capture immense value, much like Intel and Microsoft did during the PC revolution. The risk is a new form of technological oligopoly, which could actually stifle the very innovation a GPT is supposed to fuel.
A Practical Roadmap for Businesses, Not Hype
So, what should you do if you're trying to navigate this? Don't start with the question "How can we use generative AI?" Start with: "Where do we have high-volume, repetitive, cognitive tasks that follow patterns but require some nuance?"
My advice, drawn from painful experience:
- Pilot in low-risk, high-feedback domains first. Internal knowledge base Q&A, drafting first versions of routine documents, summarizing long meeting transcripts. The feedback loop is tight, and errors are contained.
- Invest in "AI literacy" training, not just tool training. Teach your teams about probabilistic outputs, confirmation bias, and how to craft effective prompts. This is the new digital literacy.
- Focus on process augmentation, not job replacement. Look for tasks that are bottlenecks. Is it competitive research? Code review? Customer email triage? Augment those specific points with AI, measure the change in throughput and quality, and iterate.
Treat generative AI not as a magic wand, but as a profoundly powerful, yet quirky, new team member that needs constant supervision and clear instructions.
Your Burning Questions, Answered Without Fluff
The journey of generative AI from a fascinating tool to a true General Purpose Technology is not guaranteed. It hinges on solving the reliability-energy-skills trilemma. But the evidence of pervasiveness and innovation spawning is already piling up. The question is no longer if it will be transformative, but how we will guide that transformation to avoid the pitfalls of past technological revolutions. One thing feels certain: we are not just adopting a new technology; we are learning to collaborate with a new form of intelligence, and that will change everything.
Leave a comment