Kohinoor

On the recut of a famous stone, and what sharp words cost the people still learning.

In 1851 the Koh-i-Noor was the most famous object in London, and it was a disappointment.

It sat in the Crystal Palace at the Great Exhibition, six million people filed past it, and the general verdict was that it looked like a lump of glass. The organizers tried to fix the viewing before they tried anything else — a locked gilded cage, then a darkened enclosure lined with gas jets to force some fire out of it. It stayed dull.

The stone was not dull. It had been cut in the Mughal manner, which optimizes for retained weight and for the light the stone was actually going to be seen in. Set against London's grey daylight, in front of a public that had just been taught what a brilliant cut looks like, an honest Mughal stone reads as a rock.

So Prince Albert had it recut. In 1852, at Garrard's, with the Duke of Wellington ceremonially striking the first facet and a cutter brought over from Coster of Amsterdam doing the remaining thirty-eight days of work, the Koh-i-Noor went from roughly 186 carats to about 105. Two-fifths of the stone became dust on a workshop floor — permanently, irreversibly — to satisfy a room of people who had concluded that the problem was the stone.

Albert is said to have been unhappy with the result anyway.

I know this is not a neutral story to reach for, and that the men who judged the stone dull were not the people it came from. That is part of why the metaphor fits.

The same move, in our decade

The genre is familiar by now. A leader with a stage says a version of: this is the worst cohort of engineers we have ever produced. They cannot debug. They ship what they do not understand. They are not really engineers. Don't bother learning to code.

I am deliberately not attributing any of it, because the specific speaker is the least interesting part. The genre matters more than any instance of it, and the genre has an incentive built into it: a hedged claim about how people acquire proficiency does not travel, and a verdict about a generation does. Sharpness is not a side effect of these statements. It is the distribution mechanism.

And here is the part that makes it hard to argue with: the critics are not wrong about what they are seeing.

The critique, at its strongest

Stated the way I would state it if I wanted to win with it:

Generation has become cheap, so people now produce far more code than they understand. The output compiles and often passes tests, which makes the gap invisible. The person who shipped it cannot tell you what it owns, what it assumes, or what breaks it. Worse, the next generation is learning this way from the start — so they never acquire the judgment that used to come from writing every line themselves. Velocity is being converted directly into debt, and into a cohort of engineers who cannot debug what they deployed.

Every clause of that is true about a real and common practice. It also misidentifies the disease, and the misidentification is exactly why the sharp version does damage instead of work.

What the critique gets wrong

It assumes that generating code and understanding code are the same act — that you understood your system because you typed it.

That was true, and it was an accident. Typing was never a comprehension mechanism. It was a slow enough process that comprehension had time to happen alongside it, for free. Remove the slowness and you do not remove the need for comprehension; you remove the thing that used to smuggle it in.

So the prescription that follows from the critique — type more, use less, go back — is aimed at the wrong layer. It asks people to reinstall the delivery vehicle rather than the cargo. And when it is delivered as a verdict on character rather than a claim about process, it stops being a prescription at all.

Uncut gems

Diamond is the hardest natural material we know, and it splits like a deck of cards if you know where.

It has perfect octahedral cleavage: four planes, fixed by the lattice, not negotiable. A cleaver spends days studying a rough stone before marking one line on it, and then strikes once. Aligned to the plane, the same blow that would have shattered the stone opens a clean face instead.

Joseph Asscher studied the Cullinan for months before he touched it. When he finally struck, on the tenth of February 1908, the blade broke. Not the stone — the blade. (The version where he faints is probably invented; the broken blade is not.) He reground, struck again, and the stone opened exactly along the line he had marked.

Force was never the variable. Alignment was.

Pressure is also what makes diamond in the first place: carbon held at enormous pressure, a hundred and fifty kilometres down, for a geological length of time. The same quantity that forms the stone destroys it. The only thing separating the two outcomes is whether the pressure is aligned to the structure of the material it is applied to.

That is my whole objection to the current discourse. Friction produces pressure, and pressure that is not aligned to the structure of the thing it lands on does not shape it — it fractures it, in a direction nobody predicted.

Somewhere for the pressure to go

Sharp words from a stage are force. They land on a population whose planes nobody has mapped, because almost nobody senior has sat down and watched how a person entering this field in 2026 actually learns an unfamiliar framework. We are striking a stone we have not studied, in public, at volume, for reach.

What that produces is never what the speaker intended. It produces concealment rather than correction — people do not stop generating when they are told generating is shameful, they stop showing that they generated, and the behaviour you needed to watch in order to teach into it becomes the thing they have learned to hide from you. Raise the social cost of "I don't understand this yet" and you get less admitting, not less not-understanding: a cohort that looks more confident and is less legible, to you and to itself. And "this generation" is an identity claim rather than a process claim, so it arrives as shame with no procedure attached. Nobody can fix being their cohort.

The fix is not softer words. It is a process with somewhere for the pressure to go.

If comprehension no longer arrives free with typing, it has to be built deliberately, as its own instrumented pass. In my own work that pass has a shape and a name: burst, then refine.

A machine can locate the generation and change; only a person can say whether every cut matters, and answering that is what turns code I generated into a system I own. It is also a better teacher than writing all the branches myself would have been.

The stone

And this journey will be the choice in front of anyone with a stage and an opinion about the people entering this field of computer science right now. You can be certain about the stone, in public, at volume. Or you can accept that cutting is irreversible and comes last — after you have studied the stone, marked one line, and understood which way the planes run.

Engineering is changing and the bar is raising in ways we didn't expect. It's best we learn how to tame the next generation of working engineers rather than argue with no progress in sight.

Let's build the magic lanterns to study the stones.