I love programmers. The best programmers are among the smartest people in the world. They built AI. But I will not bat an eyelid when AI eliminates the work of writing the same thing someone else is writing down the street.
Just three years ago, a hundred programmers in Los Angeles were writing the same code at the same time. Email validation. The same form. The same functionality, commissioned separately, paid for separately, implemented again. It made no sense, but companies kept paying and salaries kept rising. There was money in doing it, so people did it. Today, that happens only occasionally, and I am glad.
Of course there were frameworks. Of course people tried to reuse code. The optimization was difficult. AI has done in three years what would otherwise have taken another fifteen or twenty. I welcome that correction. Respecting programmers does not oblige me to defend work that no longer needs doing.
The same goes for people re-entering invoices and forwarding emails. Across customer service centers and accounting departments, thousands of people spend their days on this dull, repetitive work: copying information between systems, entering the same data again, passing a message to the next person. I do not want us to preserve pointless work just because someone has been paid to do it until now.
Karim Lakhani says that the competition is another person using AI. I agree. There are two things going on: we are improving the quality and efficiency of work, and we are automating work that should not need doing in the first place. One person can replace ten. That person's pay will rise. What happens to the other nine?
I do not know. It frightens me. I accept that this is happening and that we cannot reverse it. I agree with Lakhani, and I grieve for the people left behind. My enthusiasm for eliminating the work does not give me an answer for the people who lose it.
I co-own a company that produces 3D imagery. In four years, the progress has been extraordinary. Architecture still needs an architect. Design still needs a designer. But simpler modeling and finishing tasks have already moved to AI. A designer spending ninety percent of the day resizing images, adjusting resolution, and fiddling with production details was spending most of the day on work that wasted their abilities.
The same change is happening with websites. GPT Codex or Claude Code can build a site in minutes that beats sites agencies charged tens of thousands of dollars for. You have to be creative enough to get there. Tell it to make a ten-thousand-dollar website and it will give you something unremarkable. The price is not a creative brief. Put in the creative energy you would put into a proper website, and the technical work can move at a completely different speed.
The amount of thought involved and the amount of production labor involved are beginning to come apart. We can execute our ideas faster, make changes, and see their effects without turning every decision into another round of expensive production.
When Jensen Huang says everyone is now a programmer, he means we can give computers instructions in ordinary language. In that sense, he is right. The limit of the metaphor is professional competence: being able to ask for software does not mean knowing whether it works properly or solves the right problem. The same applies to design and writing. Access to production does not give everyone the same judgment.
The best programmers will be needed and paid even more. The money could be astronomical. The same goes for the best designers. We will need one where we needed ten, perhaps one where we needed a hundred. But I cannot turn that into a claim that the people who keep their jobs must be the most talented. I am a co-owner. I can decide how work is organized and invest weeks in tools. An employee assigned repetitive tasks does not automatically have that freedom. The task tells me what they were paid to do, not everything they are capable of.
Those weeks of preparation also help explain the productivity gain. I build instructions, files, and skills that carry decisions into later projects. A one-sentence request can draw on weeks of work already stored in that environment.
Ted Chiang argues that small decisions made during execution matter to art, and he allows for detailed, iterative work with a generator. Preparing my environment does not settle his question about who makes those decisions in the actual result.
This essay is a concrete example. I supplied the experiences and arguments, established the language and tone, and chose the title. The model proposed sentences, transitions, and an order for the material. I rejected ideas that were not mine and passages that softened what I wanted to say. I also left it room to propose wording I had not specified. My instructions did not determine every sentence; reading and changing the draft was a separate part of the work.
That is the division I need to understand when I reuse a system: what I have already decided, what I will check in this particular result, and what I am leaving to the model. Reusable context does not make all its decisions mine. It does let me carry earlier work forward instead of briefing everything from scratch. I still have work to do, but I do not have to repeat all of it. That is how I can take on more without bringing in the same amount of production labor.
I have been lucky. I am in a position to organize work around these tools, and they have unlocked more of what I can do. I am happy about that. But I have no idea, no plan for the people whose work is disappearing. I do not know what they will do or what will happen to those other nine out of ten. I can see what this change has given me. I do not know what it leaves for them.