The Last Craftsmen

I was a craftsman of code. I’ve been trying for a while to put this particular feeling of mine into words. I’ve been feeling it for some time now, and I think I can finally name the approximate shape of it. It’s not fear. I’m not afraid of what is to come. It’s not anger. I’m not angry about how the world is moving forward. It’s meaningless to be angry. The world moves forward regardless. I think it’s closest to a form of grief.

I’m in mourning.

For much of my professional life I was a craftsman. I liked writing code. Sure, I wrote code because it was necessary for me to make the things that I wanted to make as a roboticist. Writing code is how you breathe life into the mass of copper, steel, and silicon that is a robot.

But the act of writing the code was in itself an expression of craftsmanship. I liked finding ways to express logic elegantly. I liked the satisfaction of replacing spaghetti with a tight, well-written routine. I liked watching the code transform into the best, most robust version of itself as I patched failures. An edge case at midnight, an unexpected input from an undocumented API call, some assumption that I didn’t realize I was making.

Iteration after iteration, the code was wrought and folded and reworked like the steel of the master blacksmiths of old until it was strong and resilient. There was as much satisfaction in writing a good piece of software as making a good knife.

Over the last eight months or so, I’ve barely written a single line of code by hand.

Was it a deliberate decision? Somewhat. The tools were getting good enough that not using them meant that I couldn’t execute as fast as someone who was. And in research, as in business, speed of execution is paramount. It became increasingly difficult to justify not using generative AI—even to myself.

I’m paid to make things work, after all, not to indulge in the experience of making them work.

Long ago we built high level languages like Python so that we wouldn’t have to torture ourselves by writing assembly. And even longer ago we invented assembly so that we wouldn’t have to torture ourselves by writing machine code by hand. We built compilers, debuggers, IDEs, and auto-complete. We camped out in Stack Overflow, hoping the disgruntled mod wouldn’t close our question un-answered.

Programming has always involved building abstractions that allow us to tell the computer what we want while worrying less about how it happens.

Finally, we’ve created the abstraction to rule all abstractions. Just tell the computer what you want.

The genie has granted our wish.

And yet, I mourn.

I sometimes think about legendary names like John Carmack, spoken of in whispers in programming forums. His “evil floating point bit level hacking” in the Quake III source code was a story to be told around the campfire. As these myths and legends fade from memory, what are we left with?

There will always be brilliant people who can do amazing things with the tools at their disposal. But no one today is an expert in cavalry charges in the way Alexander’s commanders were. The skill still exists in history, but the world no longer creates people who spend their lives mastering it.

Someone born in 2040 might have had the potential to be the greatest C++ wizard who ever lived. But if machines can already produce better code than they can, faster than they can, why spend 20 years developing the intuition?

I think the greatest programmer who will ever live has already been born.

Translation

Consider translation.

You have to spend decades immersing yourself in the second language. But this immersion was only the beginning. You might live in the country where the language is spoken, live as they do to better understand the cultural context of the words. You might permanently live between two worlds to understand both the worlds well enough that you translate faithfully between them.

A great translator was also a bridge between two worlds. Someone who could be depended on to accurately convey intent and meaning when two countries decided to negotiate trade agreements or ceasefires. An accurate translation could be the difference between peace and war, life and death.

Machine translation has been around for a while and it had been improving steadily. But modern large language models have made it easier and easier to remove the human from the process. Enormous amounts of everyday professional translation can now be done with little or no human involvement. Perhaps literary translation will remain protected for longer, because there we explicitly value interpretation and authorship. But much of the economic machinery that once gave people a reason to become exceptionally good at bridging two languages may disappear forever.

This fading of human expertise goes beyond the replacement of human translators.

Experts are not created in a vacuum. There is a ladder, a pipeline. A large number of people begin as mediocre translators. Some become competent. Some spend decades mastering their craft. A tiny fraction become extraordinary. The existence of ordinary professional work gives people a reason to climb that ladder.

Remove the lower rungs, and it’s not just today’s experts that are erased. You break the process that will produce the experts of tomorrow.

The greatest Japanese-to-English translator who will ever live may already have been born. Someone in the future might possess even greater natural ability but never have a reason to hone their skills.

Art

Something similar seems to be happening in the visual arts.

For most of human history, imagining an image and producing it required vastly different levels of effort. Even the most extraordinarily imaginative person might never see their idea realized without developing the skill to bring it onto the canvas. Anatomy, perspective, lighting, composition, color and control of the medium took decades to master.

Generative AI flattens the hill between vision and craft. It is increasingly possible to possess the idea without possessing the technical ability to realize it.

As with most technological advances, there’s something genuinely wonderful about this too. Someone who could never draw can now create images that only existed within the confines of their mind. New forms of expression have opened up, with far less technical skill required to turn an idea into an image.

The other side, of course, is that things like commercial illustration, concept art, advertising and all the other mundane applications of art —i.e. the economic machinery that created demand for artists and let them pursue their art by getting paid for it — may disappear.

Even the great artists of old like Michelangelo had rich patrons who wanted to see more of their art. When the economics moves on to pay art factories with minds of silicon, will as many people still spend decades honing their craft? Just as the greatest sculptor may have lived and died during the Renaissance, the greatest practitioner of many forms of visual craft may already have been born.

People will continue to draw and create art of course, just as people continued to paint portraits after the invention of photography. But far fewer people hire a painter to paint their portrait than hire a photographer.

Portrait painting is a preserved craft. The same is true of sculpture and many older forms of art. They remain alive, and people continue to push them forward, but they no longer occupy the same place in the machinery of mass culture.

Photography

Photography itself offers a useful complication to this argument.

It has already been through several rounds of automation. Photographers once had to understand exposure simply to produce a usable photograph. Automatic exposure weakened that requirement. Autofocus removed another technical burden. Digital photography removed the need to understand film stocks, chemical development, and the darkroom simply to make an image. Computational photography now makes dozens of decisions before the person taking the photograph ever sees the image.

But photography is not dead.

People still take photos. Some people still shoot film, like I do. Some still use manual-focus lenses. And some still use their own darkrooms to develop, edit and make prints of their photos. They revel in the added difficulty. They find value in waiting for the photo to be developed and in not knowing how the photo turned out keeps them more present in the moment. The scarcity increases the value of each photo, they say.

I’m not sure all difficulty is the same though. Some difficulty is just pointless friction. Some of it changes you. And I don’t know that we are very good at telling the difference while we are busy removing all the friction from our lives.

But photography will never be the same as it was in the past.

The photos that modern smartphones, or simple point and shoot cameras produce, operated by the average person is good enough for most situations and more than good enough for the purpose of preserving a memory. We will never return to the age of daguerreotypes.

A craft practiced by enthusiasts can survive. But the ecosystem that reliably produces masters can shrink.

Writing

Is writing dead?

Anyone with a half formed idea and an outline can now feed them into a large language model to produce a camera ready essay with perfect prose and coherent arguments. While humans and AI do tend to have very different ways of writing that can — with varying amounts of effort — be identified, I think most people would agree with the general trend. Writing is getting increasingly easy to automate and it’s getting harder and harder to distinguish human writing from machine generated writing.

Even this essay which I mostly wrote by hand — with some AI assistance to check for grammar and awkward phrasing — may one day be completely generateable with just the seed of the idea. I spent maybe five or six hours thinking carefully and probing my thoughts before this essay crystallized into identifiable thought in my mind. When an LLM can do something that looks like the same thing in 10 minutes, will I still spend those hours probing my own thoughts?

Writing is a little special, in my opinion and may be especially dangerous to outsource. It’s not just the transcription of thoughts into words. Writing sharpens the mind, clarifies ideas and lets humans understand themselves better. To write is often to discover what you think rather than to simply narrate your thoughts. To write is to change yourself.

Thoughts emerge in the act of trying to express them. You begin a sentence, believing one thing and by the end, discover that you cannot defend this belief. Still dissatisfied, you revise the sentence, you delete it and rewrite it completely. And in the process you find that the thought itself has changed. It may even be accurate to say that writing comes more from the unconscious parts of your mind and is perhaps the clearest way for your unconscious mind to tell your conscious mind what it thinks.

The struggle to express what you think is part of the process that creates thoughts and opinions. If this struggle is outsourced, what happens to thinking?

Craft and Capability

I think many conversations and discussions about AI gloss over the distinction between the thing that we want made and the human ability to make it.

Many things that humans make today cannot be made by a single human. I think I, Pencil, the essay by Leonard E. Read, gives a good example through the manufacturing of a pencil. Nobody knows the entire chain of processes involved. One person knows how to cut down the tree. One person knows how to shape the wood. One person knows the best formula for the graphite. Another knows how to distribute the pencils to the right places and market them. No one person can make a pencil from scratch.

The same thing applies to many things that humans make today. So if we already rely on systems in which no individual can make the whole thing from scratch, why worry about AI taking over part of this process?

From the perspective of output, humans have progressed. We can make things more efficiently. We can write text faster, draw things faster, and make more things in general. We can make as much as people want, more cheaply, so that more of the people who want something can afford it. What’s the problem here?

I think the thing that makes me mourn is that automating the process of creation does something a little different. On the output side, humans do become more productive. But on average, individual capability has reduced.

We may have better software with fewer programmers, or even without programmers. We may have better translations without translators. We may have better portraits without portrait painters. And maybe even better essays without better writers.

But does it matter?

Can a humanity that has lost touch with writing appreciate better writing? Can people who have lost the link between the art and the artist’s story appreciate better art? Can people who have outsourced their thinking appreciate better thought?

What does it mean for a humanity that has lost its ability to appreciate craftsmanship? Can such a humanity really receive—or even recognize—the better craftsmanship that machines might give us?

The Genie

So I’m not writing any of this to suggest that we stop using AI. I’m still using it for writing code and to get things done at work. And I don’t expect to stop using it as a tool any time in the future. In fact, I’m actively reading about AI and finding better and better ways to integrate it into my work.

The benefits are too great and the incentives are too powerful for me to seriously entertain a personal ban on AI. I’m easily two to three times more productive as an AI assisted programmer by my own estimate and my accumulated experience still lets me review the code, catch mistakes, and produce good software faster than I could before. No, asking people to stop using AI for programming is like asking people to stop using spreadsheets since manual bookkeeping feels more meaningful.

Save for some kind of Butlerian Jihad, we’re never going back to how things were. I’m a realist in the end.

But progress and loss are not mutually exclusive. In fact I would argue that they are two sides of the same coin. To move forward, you must leave something behind. And I’m now in the position to mourn what humanity will leave behind as we integrate AI ever more deeply into our lives and our civilization.

The Last Craftsmen

I sometimes think about the person learning programming today.

Will they ever spend six hours tracking down a bug caused by undocumented behavior? Will they ever experience that strange mixture of frustration and satisfaction that comes from finally making the connection? Will they ever curse the author of the library they are using for not documenting it well enough? Will they ever accumulate the countless variations of these encounters that slowly crystallize into expertise?

Maybe none of those frustrations were valuable in themselves. Maybe six hours lost to a stupid bug is just six hours lost to a stupid bug. But somewhere in all that pointless difficulty, something was being built besides the software. The programmer was being built too.

Maybe.

But I think very few people will need to. And I think the same will be true for translators, visual artists, musicians, writers, singers, and the whole brilliant, chaotic mess that is human creative endeavor.

I will use AI. And I will probably be able to accomplish things that would otherwise have remained dreams. And so will many others.

But the number of craftsmen will probably dwindle and fade.

I was a craftsman of code, and I may never practice that craft professionally again.

And as we march onward, I think it is worth pausing for just a moment to mourn.

The greatest craftsmen may already have been born.