What Are We Not Asking Yet About the AI Buzz?

I use AI a lot.

As a software engineer, I have experienced first-hand what it can do to productivity. The first version of Monesize took us about eight months to build with a team. I built Monesize Core largely by myself in about two months, with AI being a significant part of the development workflow.

So this is not an article about how AI is useless, as it clearly is not.

What bothers me is something else. There is an enormous amount of noise around AI, and much of the conversation seems to move from a small number of things we can actually observe to a very large number of conclusions that we have not established.

AI can make a skilled person dramatically more productive. That is real. AI can also allow someone with little or no professional competence in a field to produce something that looks surprisingly competent. That is also real. Beyond those two things, however, I think we have accumulated an extraordinary number of assumptions.

And I keep coming back to a question that seems strangely absent from the conversation: what are we not asking yet about the AI buzz?

PS: This post was revised/articulated by Claude. The initial draft and thoughts are 100% mine.

What Do We Actually Know?

For me, the starting point is simple. AI is a productivity tool. Everything else is fluff.

Give a competent software engineer a capable coding model and that engineer can get considerably more done. Give a researcher an AI system that can search, summarize, and organize information, and some parts of the research process become faster. Give a competent writer an AI assistant and drafting, editing, and iteration can become faster.

That is valuable. It is also not particularly mysterious. The human already knows what they are trying to accomplish. AI reduces the amount of time and effort required to get there.

My experience with Monesize is one example. I did not become a software engineer because of AI. I already knew how to design systems, write software, work with databases, deploy applications, reason about APIs, and make architectural decisions. AI simply made the implementation process much faster.

That distinction matters. The strongest case for AI does not require us to pretend that AI has suddenly created competence. It requires us to acknowledge that competent people can become dramatically more productive with it. And I think that is already a profound technological development.

But then I have another question. Where is the corresponding economic value?

Are Productivity Gains Becoming Proportional Economic Gains?

It is easy to demonstrate that someone saved two hours using AI. It is much harder to demonstrate that the company is now materially more profitable because of those two hours. Those are not the same thing.

There is already evidence of what researchers have started calling an AI productivity paradox. A 2026 NBER working paper based on corporate executive data found that perceived productivity gains from AI were larger than measured productivity gains, suggesting that the benefits people experience at the task level do not necessarily appear immediately in company-level revenue or productivity statistics.

That does not mean the productivity gains are fake. It means the economic transmission mechanism is more complicated than AI makes an employee twice as fast, so the company makes twice as much money.

A company can make an engineer twice as fast and use that additional capacity to build more features. It can run more experiments. It can reduce development time. It can improve quality. It can attempt projects it previously could not afford. All of those things can be useful without immediately showing up as a proportional increase in profit.

There are also companies where AI adoption is producing measurable commercial results. That is important. We should not pretend otherwise. But that is precisely why I think the question deserves more attention: how much of the enormous productivity being reported is actually becoming durable economic value?

If AI is producing an extraordinary productivity revolution, I would expect the evidence to eventually become impossible to miss in corporate margins, output, revenue per employee, and broader economic productivity. We are not there yet.

What Does All of This AI Infrastructure Actually Cost?

This is where the conversation becomes even more interesting. When we talk about the cost of AI, we often talk about the price of a token or the electricity consumed by an individual prompt. That is not the real cost.

The real system includes data centers, GPUs, networking equipment, semiconductor manufacturing, construction, electricity generation, cooling, water, land, maintenance, engineering, research, replacement hardware, and the enormous capital required to build the infrastructure in the first place.

The scale is already enormous. The International Energy Agency projects that global data-center electricity consumption will more than double from 2024 to about 945 TWh by 2030. It identifies AI as the most important driver of that growth alongside other digital services.

That is an extraordinary amount of energy, and the capital commitments are equally extraordinary. Reuters reported that OpenAI expected to spend around $600 billion on computing resources through 2030. The same report said OpenAI generated $13 billion in revenue in 2025 against approximately $8 billion in expenses, while inference costs quadrupled during 2025 and adjusted gross margins fell from 40% to 33%.

That does not mean OpenAI is a bad business. It means the scale of the bet is enormous.

Alibaba offers another useful example. In August 2026, the company reported that quarterly net profit fell 75% despite a 9% increase in revenue, while capital expenditure rose sharply as it invested heavily in AI infrastructure. Alibaba has committed 380 billion yuan, roughly $56 billion, to AI investment through 2029.

Again, none of this proves that AI will not eventually produce enormous returns. It simply raises a question I think deserves to be asked much more aggressively: are the returns going to justify the resources being committed?

And What About the Environmental Cost?

This is where I think the AI conversation becomes particularly incomplete. The planet was not an empty data center waiting for GPUs. We already had climate change. We already had greenhouse-gas emissions. We already had water scarcity. We already had resource constraints. We already had semiconductor manufacturing with its own environmental footprint. AI is arriving on top of all of that.

The environmental cost of AI therefore cannot be reduced to the electricity used to answer a prompt. There is the electricity itself. There is the infrastructure required to generate it. There is cooling. There is water. There is semiconductor manufacturing. There are raw materials. There is construction. There is transportation. There is hardware replacement and electronic waste.

And the numbers are growing. The IEA expects data-center electricity demand to grow around 15% annually between 2024 and 2030, more than four times the growth rate of electricity consumption from the rest of the global economy.

I am not arguing that AI should not consume resources. Every useful technology consumes resources. The question is much more fundamental: is the value we get from AI large enough to justify the physical resources we are allocating to it? That is not an anti-AI question. It is an accounting question.

But What If Much of the Computation Produces Almost No Value?

This is where I think the conversation gets even stranger. AI has made the production of digital things extraordinarily cheap. That sounds unambiguously good until you remember that humans have never had unlimited demand for digital things. We have unlimited capacity to produce information now. We still have approximately the same number of human brains capable of paying attention to it. So we are increasingly generating things simply because we can.

Look at the internet today: AI-generated articles, AI-generated images, AI-generated songs, AI-generated videos, AI-generated social posts, AI-generated websites, AI-generated applications. Some of these things are useful. Some are entertaining. Some are genuinely creative. And some are just garbage.

I have seen the same bizarre AI videos everyone else has seen. Cats doing impossible things. Strange cinematic scenes that exist for no reason other than someone thought it was amusing to generate them. Images that are obviously synthetic at first glance. Songs with lyrics that technically make sense but feel completely mechanical.

The interesting question is not whether these things can be generated. Obviously they can. The question is: what value did we create by generating them?

That matters because generation is not actually free. Even a useless piece of content consumes computation, electricity, storage, and bandwidth. It occupies recommendation systems and search indexes. It competes for human attention. And if millions of people are generating millions of pieces of content that nobody meaningfully wants, the fact that each individual generation is cheap does not mean the aggregate resource consumption is meaningless.

We Are Seeing the Same Thing in Software

This is one of the things I find particularly funny about the current SaaS environment. We are producing software at an incredible rate here and there. Expense trackers. Invoice makers. Inventory trackers. Habit trackers. CRMs for a very specific profession. Operating systems for very specific industries. AI dashboards. AI assistants. AI workflow tools. AI platforms and so on.

Lol. Just lol.

Thousands upon thousands of applications can now be produced with a fraction of the engineering effort that used to be required. But I keep asking myself: how many of these applications actually need to exist?

The cost of producing software has fallen. That does not mean the demand for software has increased proportionally. A person can now build an application over a weekend that previously might have required a small development team. Wonderful right? But if nobody needs the application, we have not created much value. We have simply made the production of software cheaper.

That distinction is incredibly important. Output is not value. And AI may be creating an enormous amount of output.

The Illusion of Competence

There is another effect of AI that I think deserves much more attention. I call it the "illusion of competence."

Someone who could not write a "hello world!" program can now prompt an AI system into producing a functional application. Someone who struggles to write a paragraph can ask an AI system to produce an essay. Someone with little design experience can generate a poster. Someone who knows very little about a subject can ask an AI system to produce an apparently authoritative explanation.

This is genuinely useful in some situations. But it creates a psychological problem. The person can now produce the output without necessarily acquiring the competence behind the output. That distinction is enormous.

A person can generate software without understanding software engineering. They can generate an essay without becoming a writer. They can generate a graphic without becoming a designer. They can generate an analysis without becoming an analyst. They can generate a medical explanation without becoming a doctor.

The danger is not that the AI output is always wrong. The danger is that it can be plausibly right often enough for the user to believe they have acquired a capability they have not actually acquired. This is particularly important because the better AI becomes, the harder that illusion can become to detect. If the output were obviously terrible, there would be no problem. But when a system produces something polished, articulate, and convincing, a person without the underlying domain knowledge may not know which parts are wrong.

There is already research examining this relationship between AI use, cognitive offloading, and critical thinking. One 2025 study found an association between greater reliance on AI and reduced critical-thinking engagement, while Microsoft Research and Carnegie Mellon researchers have also studied how knowledge workers' confidence and critical-thinking effort change when using generative AI. These studies do not establish that AI makes people universally less intelligent, but they do establish that cognitive offloading is an important question worth investigating.

That brings me to another question.

Are Humans Actually Becoming Smarter?

We now have access to more information than any generation before us. AI makes access to that information even more seamless. But are humans becoming cognitively smarter? I don't know, and I don't think anyone should confidently claim that we are.

Access to information is not the same thing as understanding. Getting an answer is not the same thing as knowing how to derive the answer. Having AI write an explanation is not the same thing as learning the subject. Having AI solve a programming problem is not the same thing as developing programming ability.

There is a difference between externalized intelligence and internalized intelligence. AI may be giving us extraordinary access to the former. Whether it is substantially increasing the latter remains an open question.

And Then There Is the Quality Problem

This is where my personal experience makes me less enthusiastic than a lot of the current discourse. AI is faster. I will happily concede that. But faster is not the same as better.

I have read enough AI-generated writing to recognize the texture of it. A good human writer has a voice. There is something behind the words. The writer has a rhythm, a preference for certain constructions, a particular way of seeing the world. Sometimes you can almost hear the writer's inaudible words inside the text.

AI-generated writing often feels different. It can be grammatically excellent. It can be coherent. It can be polished. But it often feels statistically smooth. The sentences converge toward familiar structures. The vocabulary converges toward familiar patterns. The prose can be technically competent while being strangely anonymous.

The same thing has happened visually. Before generative AI became ubiquitous, websites could look remarkably different from one another. Companies had brand guidelines. Designers developed visual systems. A company's website reflected decisions made over years about typography, spacing, imagery, interaction, and identity.

Now I regularly encounter websites that feel like variations of the same generated template. Different colors. Different gradients. Different logos. The same underlying visual language. The same cards. The same rounded buttons. The same dashboard structures. The same animations. The same aesthetic vocabulary.

The same thing is happening with graphic design. The same thing is happening with code. Professional engineering teams have internal conventions, architectural preferences, linting rules, naming standards, and institutional knowledge. Code written by those teams reflects the history of the organization and the people who maintain it. AI-generated code can be perfectly functional while still converging toward recognizable patterns.

That is not necessarily bad. Consistency is useful. But there is something we lose when the entire internet begins to look like variations of the same statistical center.

Humans Were Already Doing These Things Well

This is perhaps the biggest point I want to make. Before AI, human professionals were already producing excellent work. Human writers produced excellent writing. Human designers produced excellent designs. Human engineers produced excellent software. Human photographers produced excellent photographs. Human musicians produced excellent music. Human researchers produced excellent research.

AI did not arrive to rescue humanity from an inability to produce these things. It arrived and made the production process faster. That is valuable. But I don't see evidence that the average AI-generated output is inherently superior to the output of a genuinely skilled professional. In many cases, I think the opposite is true.

A professional has judgment. A professional has taste. A professional has context. A professional has accumulated experience. A professional knows what not to do. AI can assist with all of these things. But assistance is not the same as replacement.

What About the Problems That Actually Matter?

There is another thing that makes the AI discourse strange to me. We talk about AI as though civilization is approaching some extraordinary transformation. Maybe it is. But I look around and see the same major human problems.

People are still hungry. People are still poor. People are still homeless. People still lack access to adequate healthcare. Millions of people still lack reliable electricity and clean water. Wars still happen. Governance still fails. Inequality still exists.

AI may eventually help with some of these problems. It may improve agricultural planning, logistics, scientific research, medicine, education, and public administration. But "AI could help solve a problem" is not the same as "AI solved the problem." The distinction matters because we are making enormous investments today based partly on expectations about what AI will eventually do. I want to see what it actually does.

So Is AI More Useful to Humanity Than It Costs?

This is the question I keep coming back to. Not whether AI is useful. It obviously is. Not whether AI increases productivity. It clearly can. Not whether people enjoy using it. Millions of people obviously do.

I mean something much bigger: is AI more useful to humanity than it costs humanity? And by costs, I mean everything. The capital. The labor. The electricity. The semiconductor manufacturing. The data centers. The water. The land. The environmental impact. The hardware. The research. The opportunity cost of all that capital and talent. The information pollution. The synthetic content. The low-value computation. The social consequences of cognitive offloading. The potential erosion of trust in what we see and read.

Against all of that, what is the total value we are creating? I don't think we know. And I don't think the answer can be inferred simply from the fact that AI companies have enormous valuations or that millions of people use AI.

Usage is not value. Revenue is not necessarily profit. Productivity is not necessarily economic output. Output is not necessarily value. And technological capability is not necessarily human progress.

I Don't Think AI Is Useless

In fact, I think that conclusion would be absurd. I have personally benefited from it too much to make that argument. AI has made me faster. It has allowed me to build more. It has reduced the friction between an idea and an implementation. For a competent professional, that is enormously powerful. I expect that capability to keep improving.

What I question is everything that has been attached to that capability. The assumption that because software is easier to produce, every new piece of software is valuable. The assumption that because people can generate professional-looking work, they have acquired professional competence. The assumption that because AI increases task-level productivity, the economy will automatically experience proportional productivity and profit gains. The assumption that because AI can generate almost unlimited content, that content represents unlimited value. The assumption that more information automatically makes humans smarter. The assumption that increasing AI capability automatically means increasing human welfare. And finally, the assumption that because something is technologically possible, it is necessarily worth the enormous resources required to make it ubiquitous.

Faster != Better

I think this is where I land personally. AI is very good at making things faster. That is probably its clearest and most defensible contribution today. A skilled engineer can build faster. A skilled researcher can research faster. A skilled writer can draft faster. A skilled designer can iterate faster. That is real.

But speed is only one dimension of quality. A faster article is not necessarily a better article. A faster website is not necessarily a better website. Faster code is not necessarily better code. A faster image is not necessarily a better image. More software is not necessarily more value. More content is not necessarily more knowledge. More information is not necessarily more intelligence. And more AI is not necessarily more progress.

That is why I think we should slow down, ironically, and ask some very old-fashioned questions. What did we create? Who actually benefited? How much value did it create? What did it cost? And was it worth it?

I don't think we have answered those questions yet. And perhaps before we spend another trillion dollars trying to make AI more capable, we should become a little more interested in answering them.