AI Isn’t a Bubble. I Watched It Do the Work of Four Engineers in My Living Room

Every few weeks, a new headline lands in the same inbox. “AI will cause mass layoffs.” “AI is the next dot-com bubble.” “AI is overhyped.” “AI is the end of white-collar work.” Pick one, and there’s a hot take to match it.

I want to offer a different perspective — one that I think gets closer to the truth than either the doomers or the dismissers. AI is real. AI is disruptive. AI is not a bubble. But what AI actually does, and who it actually threatens, is more specific and more interesting than the headlines suggest.

We’ve been here before — sort of

Technological revolutions are not new. The steam engine, electricity, the personal computer, the internet — each one rearranged the economy and put a generation of workers on the wrong side of the curve. AI belongs in that lineage. But it’s also different, and the difference matters.

Think about the calculator. When the pocket calculator became cheap and ubiquitous in the 1970s, an entire layer of clerical work — human “computers,” bookkeepers doing long division by hand, junior accountants checking arithmetic — got compressed. One person with a calculator could do the work of five. But the calculator was a single-purpose tool. It sped up arithmetic. That’s it. It didn’t write your memo, it didn’t draft your contract, it didn’t summarise your meeting.

AI is not a single-purpose tool. That is the thing people keep underestimating. It writes, codes, summarises, translates, drafts, debugs, designs, plans, and explains — across domains, in multiple languages, twenty-four hours a day, with no coffee breaks and no Monday-morning grumpiness. The calculator was a hammer. AI is a workshop.

A weekend project that should have needed a team

Let me make this concrete with my own experience.

I’m a firmware engineer. I work with VHDL on FPGAs — programmable logic devices used in everything from telecom to defence to high-frequency trading. At work, a project like building a complex signal processing chain on the programmable logic side of an FPGA is a team sport. You’d typically have a PL engineer writing the HDL, a separate engineer modelling the filter in MATLAB or Python to define the reference behavior and tune coefficients, an embedded software engineer building the Linux image and writing the application that runs on the processor, a systems engineer wiring it all together, and a performance engineer validating it all. Five people, minimum, plus a lead.

I do not have five people sitting in my flat on a Saturday morning. But I recently built — as a hobby project — a system on an FPGA with multiple processor cores, a GPU, and Ubuntu running on it, with a non-trivial filter implemented on the PL side. Alone.

How? AI acted as my junior engineer(s).

I asked it to write a Python reference model for the filter — what functions I needed, what the input and output behaviour should look like. It produced a script. The first version had bugs. By the end of a weekend, it was clean. Two days, for someone who is not a software engineer by training, to produce a working reference model. That used to be a person’s job for a week.

For the embedded side, I told it what I was trying to do on Ubuntu and it walked me through the boot image and the bare minimum functions like gstreamer pipeline that I would need. There were issues. There always are. But I could push through them. The PL work I already knew how to do, but I used AI to generate small utility functions that I could read, sanity-check, and drop in.

It even helped with diagnostics — when something didn’t behave, I could describe the symptom and get plausible hypotheses faster than I’d get them from a forum.

Could AI have replaced me on this project? Absolutely not. It cannot architect a non-trivial FPGA system, it cannot reason about timing closure across clock domains, and it certainly cannot debug a flaky board on the bench. But did it let one person do the work of a small team? Yes. And that scaling factor is the entire story.

The hyperactive colleague analogy

Most of us have worked with that one colleague. The one who somehow does the work of two people, ships everything early, answers emails at midnight, and is simultaneously the boss’s favourite and a nightmare for everyone else on the team. You can’t out-hustle them. You can’t out-organise them. They reset the baseline for what “productive” looks like, and suddenly the rest of you are sweating to keep up.

Now imagine AI is that colleague. Or, more accurately, imagine AI is a tool given to that colleague, so that one person now does the work of four or five.

That is roughly the shape of the disruption we are walking into. AI is not (yet) a full-time replacement for a full-time employee. But it is a force multiplier so powerful that the economics of headcount start to look very different. You don’t fire the engineer. You give them AI, and then you don’t hire the next three engineers you would have hired.

This is the part of the story that gets lost in the “AI will replace all jobs” / “AI will replace no jobs” shouting match. The honest answer is: AI doesn’t need to replace you to reduce hiring. It just needs to make the people already there meaningfully more productive. And it is doing exactly that.

Why India should be especially concerned

Now here is where the picture gets uncomfortable, particularly for Indian readers.

A very large chunk of the Indian IT and ITES economy is built on volume execution of pattern-heavy work. BPOs handling structured customer queries. Recruitment firms doing first-pass CV screening. Paralegal support, transcription, basic accounting, data entry, tier-one tech support, admin coordination, content moderation. Even significant slices of application development — the unglamorous truth is that a lot of it is, as the joke goes, copy from Stack Overflow and paste into the IDE. The “plumbing” work.

Pattern-heavy, repetitive, well-documented work is exactly what AI eats first. It does not get tired. It does not take chai breaks. It does not ask for a hike in March. The error rate, on tasks within its competence, is low and getting lower.

This is not a five-year-from-now problem. The squeeze on entry-level hiring at the big Indian IT services firms is already visible in the numbers, and most credible analyses point to AI-driven productivity as a major reason. The pyramid model — hire thousands of freshers every year, train them, bill them out, promote a fraction — depends on there being a steady supply of pattern-heavy work for those freshers to do. That supply is shrinking.

Meanwhile, AI’s coding ability has crossed a threshold that even optimists did not predict two years ago. It is not just generating boilerplate any more. It is writing genuinely complex functionality, doing meaningful refactoring, and debugging at a level that would have sounded like science fiction in 2022. Each model release moves the goalposts further.

Is it a bubble? No. Is there a problem? Yes.

So no, AI is not a bubble. Bubbles are built on speculation untethered from utility. AI’s utility is real, measurable, and compounding every quarter. The valuations of individual AI companies may or may not be frothy — that is a separate question — but the underlying technology is doing economically useful work right now, in millions of workflows, including mine.

The real problem is not whether AI is real. The real problem is a question almost no one is asking clearly: do we have enough work?

If demand for software, services, products, and ideas grows faster than AI grows productivity, we are fine. People shift up the value chain, the pie expands, and the doomsday headlines age badly — exactly as they did with every previous technological revolution. If demand does not grow that fast, we have an unemployment problem on our hands, and a structural one rather than a cyclical one.

This is fundamentally a test of imagination and policy. Of whether companies can identify new categories of work that AI-augmented humans can pursue. Of whether governments can create the conditions — education, capital, infrastructure, social safety — for that pivot. Of whether individual professionals can stop treating AI as a threat to argue with and start treating it as a tool to master.

For India, the opportunity is genuine. We are a young country with enormous unmet demand in healthcare, education, agriculture, manufacturing, climate adaptation, and public services. There is, in principle, far more work than AI can possibly displace. The question is whether our companies, our policymakers, and our talent are agile enough to redirect the workforce toward that work, fast enough, before the bottom of the existing pyramid gives way.

I don’t know what AI looks like in five years. Nobody does. But I know what I saw on my own desk, on my own weekend, with my own hobby project. One person doing the work of five. That is not a bubble. That is a revolution, quietly, in your living room.

The only real question is whether we are going to ride it or get rolled by it.

“The industrial revolution moved machines into factories. This one has moved them onto desks. Quietly, cheaply, and at planetary scale.”