There is a particular kind of grief floating around right now. You see it in blog posts, in conference talks, in late-night threads: a mourning for the joy of coding before AI. People describe it as if a forest has been paved over. Something they loved is gone, and something colder has taken its place.

I think most of these conversations talk past each other because they skip the only question that matters: what was the joy actually made of?

"Coding" is not one activity. It is at least two, braided together so tightly that for decades nobody had to separate them. AI pulls on one of those strands and not the other, and whether that feels like loss or liberation depends entirely on which strand you were holding.

Two kinds of joy

Strip a programming session down to its emotional core and you find two distinct rewards.

The first is the joy of materializing a vision. You see something in your head — a tool, an interface, a system, a small clever thing that does not exist yet — and you bring it into the world. The pleasure here is in the gap closing. The thing in your imagination and the thing on the screen converge until they are the same thing. The keyboard, the language, the build system: these are friction. Necessary friction, often beautiful friction, but friction. The joy lives at the moment of arrival.

The second is the joy of figuring something out. A problem resists you. You sit with it. You pull on threads, build mental models, get them wrong, refine them, and somewhere — sometimes in the shower, sometimes at 2 a.m., sometimes mid-sentence in a meeting — the shape of the answer clicks into place. The pleasure here is not in arrival but in the act of comprehension itself. You understand something now that you did not understand an hour ago, and your brain rewards you for it the way it rewards eating when you are hungry.

These are not the same feeling. They use different muscles. They satisfy different hungers. And — this is the important part — they leave behind different kinds of memory. A vision-materializer remembers what they built. A problem-solver remembers how the world bent into a new shape inside their head.

Most working programmers feel both, in different proportions, sometimes in the same hour. But if you ask honestly which one is the core — the thing that made you a programmer rather than something else — almost everyone has an answer.

What AI actually changes

AI coding assistants are extraordinary materializers. That is what they are built to be. You describe the thing, they produce the thing. The friction between vision and artifact collapses dramatically. What used to be an afternoon of plumbing is now a paragraph and a review pass.

If your joy was in the materialization — in seeing the thing exist — AI is not stealing anything from you. It is giving you more of what you loved. The gap closes faster, which means you can close more gaps, which means more visions per unit of life. The friction you tolerated was never the source of the joy; the arrival was. You can build the second thing now, and the third, and the weird side project you never had time for. The hands-on craft loss is real but it is a craft loss, not a joy loss. You can still write the loop by hand on a Saturday if you want to. Nothing stops you.

One thing has to be said clearly here, because it is the most common bad-faith reading of the materializer position: materialization is not "whatever shipped, shipped." A vision is not just a silhouette of a feature; it has internal coherence, a way it behaves under pressure, a quality it carries. A materializer who accepts slop because it superficially resembles the artifact in their head has not closed the gap — they have moved the goalposts to meet the output. That is not the joy of materializing a vision. That is the relief of being done. They are different feelings, and conflating them is how teams end up shipping confident-sounding garbage at unprecedented speed. The AI gives you a draft. The work — the actual materializer's work — is to keep pushing the draft until it matches the thing in your head, including the parts of the thing in your head that have to do with correctness, taste, performance, and how it will read to the next person. Acceleration is acceleration toward the right artifact, not toward any artifact. A materializer who forgets this is no longer practicing their craft; they are just hitting accept.

If your joy was in the figuring out, the picture is genuinely different — and the grief is genuinely earned. The AI is not just removing friction; it is removing the problem itself. The puzzle you would have sat with for three days, turning it over in your head on the train and in the shower, building the mental model that becomes part of how you think forever — the AI hands you an answer in twelve seconds. The answer is often correct. And the comprehension that would have grown in you while you struggled does not grow, because you did not struggle.

This is not a complaint about laziness or skill atrophy, though those are real concerns. It is something more specific: a category of human pleasure, the pleasure of understanding something hard, requires the hardness. Remove the hardness and you remove the pleasure, even if you keep the answer. You cannot have the satisfaction of a crossword without the crossword.

Why the debate is so confused

Once you see this split, the public conversation about AI and coding starts to make more sense. The two camps are not actually disagreeing about AI. They are reporting honestly on two different inner experiences.

The "AI is wonderful, I ship five times faster" camp is overwhelmingly populated by materializers. They are telling the truth. Their joy is intact and amplified.

The "AI is hollowing out my craft" camp is overwhelmingly populated by problem-solvers. They are also telling the truth. Their joy is, in fact, being eroded — not by malice or hype, but by the specific mechanism of having the puzzles solved before they get to play with them.

When these two groups argue, they sound like they are arguing about a tool. They are actually arguing about which of two pleasures is the real one. There is no answer to that question, because there are two answers, and they are both correct for the person giving them.

Notice how each camp uses the same phrase to dismiss the other's grief. To the materializer, the problem-solving was always an implementation detail — a means, a tax you paid on the way to the artifact, something a sufficiently advanced tool was supposed to absorb eventually. To the problem-solver, the shipped artifact was the implementation detail — the residue, the visible echo of an internal event that had already happened in their head. Each side, in good faith, treats the other side's joy as the boring scaffolding around their own. That is why the conversation goes nowhere: both sides are correctly identifying what is, for them, incidental.

What to do about it

If you are mourning, the first useful move is to ask yourself the honest version of the question. Not "do I miss coding," but: which kind of coding? When you replay the moments you would call joyful, are you watching yourself ship the thing, or are you watching yourself understand the thing? The answer tells you whether you are facing an opportunity or a loss.

For materializers, the path is mostly forward. Use the tools. Build more. The thing you loved is more available now, not less.

For problem-solvers, the answer is harder and more deliberate. The puzzles still exist; they have just stopped arriving on their own. Production code paths now route around them. To keep the joy, you have to choose the friction back in — pick problems the AI cannot solve cleanly, work in domains where the model is weak, build from scratch on weekends, read papers, do the leetcode-equivalent that is actually interesting to you, contribute to runtimes and compilers and other places where the problem space is still deep enough that no autocomplete can shortcut it. The protected hour where you do not ask the assistant is not a Luddite stance; it is a deliberate preservation of the conditions your joy requires.

Both responses are healthy. Both are grown-up. What is not healthy is conflating them — using a materializer's optimism to dismiss a problem-solver's grief, or using a problem-solver's grief to deny a materializer's genuine, earned acceleration.

The thing under the thing

The deeper claim hiding inside all of this is that coding was never one thing. It was a workbench where two very different human pleasures happened to use the same tools. The industry treated them as one because the workflow forced them to be one — you could not materialize a vision without solving a hundred small problems along the way, and you could not solve interesting problems without something to materialize them into.

AI is the first force strong enough to pull those two pleasures apart. It is doing so cleanly and without asking permission. What we are watching is not the death of the joy of coding. It is the unbundling of two joys that were always separate, finally being forced to admit it.

Which one was yours?