LLMs and integrated circuits

Large Language Models (LLMs) come in for a lot of hate in some quarters. It's not hard to see why, if you're in line to lose your livelihood to one. In the IT industry, there are sound arguments that they de-skill engineers, and that we could end up in a situation where nobody actually knows how to program any more.

I don't know whether we're right to be concerned. I know that I don't like LLMs, because I find them inelegant. Mine is an aesthetic objection, rather than a pragmatic one.

It's common to hear LLMs' detractors referred to as "Luddites", as if they were hide-bound reactionaries, uselessly standing in the way of progress. I think a more apt comparison, however, is not with the mechanical textile looms of the 19th century, but with integrated circuits (ICs) in the 1970s.

I'm old enough to remember -- as if it were yesterday --the controversy surrounding the rise in popularity of ICs. Until the 1970s, almost all electronic equipment was constructed from discrete components -- transistors, diodes, and so on -- even though the technology underlying ICs had been described about twenty years ago. Understanding the behaviour of individual transistors was still considered essential when I was an engineering student in the 1980s. Although this subject is still taught, I think it's now only of practical interest to IC designers.

An IC combines many -- perhaps millions these days -- of electronic components into a single, functional block. ICs range in complexity from logic gates and amplifiers, to complete microprocessors. Designing with ICs is radically different from designing with discrete components: in most applications it's quicker, easier, and cheaper. It wasn't really until the late 70s that ICs became sufficiently affordable to be used routinely and, when that happened, they were greeted with a mixture of great enthusiasm and crippling anxiety.

Opponents of ICs -- and in the 70s there were many -- claimed that they would de-skill the electronics industry. Rather than spending years and years learning how to combine dozens or hundreds of transistors, designers would just use one or two ICs, whose detailed operation they didn't understand, and wouldn't even be capable of understanding. Skilled designers would be put out of work as, using ICs, one person could do the work of ten (or so it was claimed).

This wasn't just an academic debate: in the 1970s there were dozens of popular electronics periodicals -- hard to imagine, today -- and their "letters to the editor" pages were full of anti-IC rhetoric. It seemed that anybody who had undergone any kind of traditional electronics training simply hated ICs, and that ICs were going to destroy the world.

I started building electronic equipment as a teenager, and I thought ICs were marvellous. Everybody my age did -- an interest in electronics as a hobby was far from unusual among teenagers back then. It's true that I didn't understand how ICs worked, and it's true that I wouldn't have been able to understand them if I'd tried -- I needed four years of university for that. But I could read a data sheet, and experiment. I could, in a sense, "vibe code" an audio amplifier or a short-wave receiver, with only the most rudimentary grasp of circuit theory. I could build more elaborate things with ICs, than I ever could have before they entered my price range.

We criticize LLMs today for similar reasons to those levied against ICs fifty years ago. They're de-skilling; they'll put trained people out of work; nobody understands how they work; they're wasteful and inelegant. And so one.

These days, it's hard to find anybody working in electronics who thinks ICs are a bad idea. You'll find a few designers who favour discrete components; some even promote thermionic valves (tubes), which predate the transistor. But, for the most part, today's electronics engineers grew up with ICs already well-established or, if you're retired like me, on the way to being established. Avoiding ICs is now seen as a peculiar, niche affectation. I don't think the advantages and disadvantages of ICs are even discussed any more; I suspect that many electronics engineers would be surprised to learn there ever was a controversy. On the whole, ICs didn't put electronics engineers out of work, although they did change that work.

In engineering, something that increases productivity, even if disruptive and broadly disliked, tends to be adopted. We tend not to count the human cost, just as we didn't when Nedd Ludd was being blamed for smashing up mechanical looms. If LLMs turn out to be successful, I suspect a time will come when nobody even remembers the controversy surrounding them.

I retired before the use of LLMs was forced on me. I doubt they would have improved my working life; frankly, I might not even have had a working life for much longer. But people entering the IT industry now will, I suspect, adopt them enthusiastically. Experience has shown that, in twenty or thirty years, they'll become an everyday part of everybody's practice.

In a sense I'm glad that I won't be around to see it.

Published 2026-03-05, updated 2026-03-05

LLMs and integrated circuits

Categories