# Is This Our Last Programmers’ Day?

LLMS index: [llms.txt](/en/llms.txt)

---

Today is October 24. In China, it is commonly celebrated as Programmers’ Day: after all, 1024 = 2¹⁰, a pleasingly round and lucky number in the binary world. It seems a good day to share a few thoughts on programming as a profession.

I have a strong hunch: **this may be the last Programmers’ Day we celebrate.**

Of course, October 24 will come around again next year. By “last,” I mean the end of programming as a distinct occupation. Before long, “programmer” may be a historical footnote, just as the old job of “computer” is today.

## 1. From Human Computers to a Historical Footnote

Why do I say that? Today, “computer” means the electronic machine on your desk. But decades ago, it was an actual human occupation: **someone who computed for a living**.

These computers used paper, pencils, and slide rules to work out powers, logarithms, square roots, and calculus problems by hand. The film **Hidden Figures** tells the story of such people: during the space race of the early 1960s, Black women working as computers at NASA’s Langley Research Center verified the trajectory for America’s first crewed orbital flight by hand. Being a human computer was a respected white-collar job that demanded real skill.

We know what happened next. As electronic computers spread, calculation went from a human craft to a routine operation performed by machines. Beyond elementary school exercises, nobody earns a living taking square roots by hand anymore.

The most telling story in the film, though, is Dorothy Vaughan’s. When she saw that an IBM computer was coming, she did not wait to become obsolete. She taught herself FORTRAN, then taught the human computers she supervised, bringing her whole team into the new profession. **The occupation of human computer disappeared, but those who embraced the new machines early became the first generation of programmers.**

![A woman calculating on paper alongside a modern programmer at a computer](computers.webp)

**Today’s programmers stand where human computers once stood.**

I made this complaint recently in [DeepSeek Interviews Bojie](/en/ai/deepseek-interview/): if you are hiring a new generation of AI-native engineers, judging them by how well they hand-code solutions to online-judge algorithm problems is absurd. It is like hiring programmers back then by testing their pencil-and-paper arithmetic and square-root calculations.

The era of programming as a distinct occupation is coming to an end. It is leaving the stage in much the same way that human computers did when digital computers took over.

## 2. The Priesthood’s Twilight and the Rise of the Superpowered Individual

I understand the psychological shock this upheaval has brought to many of my peers. Just two or three years ago, programmers were treated as something of a priesthood. We knew arcane incantations ordinary people could not read; with a few keystrokes, we could summon power from cyberspace and command enviable salaries.

Within a year or two, that sacred monopoly was broken. A technological Reformation arrived: more and more people who cannot code are picking up Claude Code and Codex as their “Bible in the vernacular,” bypassing the priests to speak directly to “God” and build software themselves.

That is a heavy blow to traditional programmers’ pride and livelihoods. If your sense of superiority still rests on being good at test problems and writing code by hand, consider these harsh but clear-eyed words from AI pioneer Ilya Sutskever:

> “if you value intelligence above all other human qualities, you’re gonna have a bad time”
>
> — Ilya Sutskever

![Screenshot of Ilya Sutskever’s post about valuing intelligence above other human qualities](ilya.webp)

**But there is another side to this: if you can set aside your pride and embrace this power, this is no apocalypse. It is a golden age in which compute lets one person accomplish more than ever.**

Cheap compute lets you take on engineering projects on a scale once unimaginable for one person. A project that previously required a team of a dozen or so people and hundreds of thousands or over a million RMB in startup capital might now be within your reach alone, with a few subscriptions to leading models at US$200 a month each.

Meanwhile, when software becomes cheaper to build, demand often rises rather than falls: Jevons paradox. The coming wave of automation across industries will create software demand far beyond anything we have seen. The old moats are being filled in, but the new territory is enormous.

## 3. Programmers Will Disappear, but Engineers Will Endure

If “programmer” becomes a historical footnote, what takes its place? The answer lies in an older, more fundamental title: **programmers will disappear, but engineers will not.**

In real software engineering, typing code has never accounted for all that much of the work. The software engineering classic *The Mythical Man-Month* offered a familiar breakdown long ago: 1/3 of the time for design, 1/6 for coding, and 1/2 for testing.

I think that ratio still holds up. Outside a few specialized low-level fields, the **1/6 spent on coding** is no longer the bottleneck. Once calculators became widespread, nobody competed on how fast they could calculate by hand. What mattered was knowing what to calculate and whether the answer was right.

The real challenges therefore shift to either end: **design** and **testing**. To be honest, AI agents already do impressive work in both. But humans still have two indispensable roles to play at the beginning and end:

1. **In design: provide intent and taste.** At least today, AI does not come up with a purpose of its own. It has no desires, feels no pain, and does not know why it exists. It needs humans to give it goals and meaning, supply ideas and tokens, and pay the bills. Human **intent** is what sets it in motion. Coming up with an idea that is actually worthwhile, deciding what to build and what to leave alone, takes **taste**.

2. **In testing and delivery: provide distribution and accountability.** Software generated entirely by agents has no one vouching for it. To put it into production and deliver it commercially, an actual person or organization must ultimately **take responsibility**, putting real money and business reputation on the line.

The grunt work between those two ends can already be handed to agents, which do it quickly and cheaply.

Looking further ahead, as the cost of software engineering falls sharply, **the work outside engineering** becomes more important: distribution, marketing, and, underneath it all, **building trust**.

In the AI era, anyone can knock out a pile of flashy projects over a weekend. Most will quietly rot in a neglected corner of GitHub. The real test for engineers now is getting other humans or agents to use what you build, establishing a positive feedback loop, and turning the project into a software operation that keeps running and pays its own way.

There is a half-joking line going around: every startup now needs a **CIO**, not a Chief Information Officer but a **Chief Influencer Officer**. Only by building real influence and relationships based on trust can you get that software machine running.

## 4. Facing the Wave: It Is Not Too Late to Start

Is resisting or rejecting AI a viable option? Honestly, no. Both China and the United States have made their bets clear: full speed ahead on AI and robotics as the next engines of productivity. Structural unemployment during the transition tends to get a passing mention in these sweeping visions of progress, dismissed as inevitable growing pains. Think of textile workers during the Industrial Revolution, carriage drivers as automobiles spread, or human computers after electronic ones arrived.

If your core skills happen to be in AI’s path, that is brutal, and it is not your fault. Anxiety and anger are understandable. But neither solves the problem, and clinging to the past will not improve your position.

When a big wave comes, the most dangerous place is the shallows, where it rises highest and breaks hardest. Either head for deeper water, specializing in low-level fields that AI cannot yet crack, or learn to surf and ride the wave. **The crest may not be safe, but it offers the best view. Only from there can you clearly see the course ahead and its risks.**

More importantly, I believe **there is still plenty of time to get started.**

Do not let the tech industry’s echo chamber distort your view. AI’s adoption in real production work is nowhere near as widespread as you might think. My rough estimate is that the world has on the order of a million programmers experienced in working with AI and a hundred thousand “commanders of agent armies.” Against 30–50 million developers worldwide, let alone a population of 8 billion, those numbers are still tiny.

[AI Adoption May Not Be as Widespread as You Think](/en/ai/funnel/)

![Chinese infographic showing five levels of AI adoption in a funnel](funnel.webp)

Everyone on this new battlefield started from zero. Even the world’s most seasoned Claude Code users have less than two years of experience. The practical feel they have developed in that time has already put their productivity an order of magnitude ahead of traditional programmers. But if you are willing to get involved now, you can still catch up.

Besides, seeing a technology of historic importance go from zero to one, helping it take shape, growing alongside it, and learning how each new generation of models changes the possibilities is a rare experience in its own right.

## 5. From Foot Soldier to Commander of a Thousand: Leading an Army of Agents

So how do you get started? My answer is simple: **put your compute to full use, and develop your instincts as a commander through real projects.**

If you spend US$20 a month on a chatbot to pass the time, at best you have a decent digital adviser. But if you get everything you can out of a leading AI service’s top-tier US$200 monthly plan, you may accomplish things you once would not have thought possible.

I know that several premium subscriptions add up to a substantial expense for many people. You do not need to buy them all at once. Start with one top-tier plan that suits you, use it until you hit the limit, and then decide whether to add another. Compare the cost with hiring one person, and the arithmetic is straightforward.

The most important thing to learn along the way is each AI’s temperament.

Do not treat Codex, Claude, Gemini, Grok, GLM, and DeepSeek as mere tools. Think of them as digital colleagues with distinct personalities and strengths. Each has its specialties, quirks of reasoning, habits of cutting corners, and blind spots. I assign work across a dozen or so top-tier subscriptions every day, so I know this well. Learning to work alongside these new colleagues, understand their temperaments, and **match the right colleague to the right job** will be essential for every engineer.

![Different AI models illustrated as humanlike characters](agents.webp)

You cannot develop that feel for directing AI by reading 10 tutorials or listening to 100 talks. The only way is to roll up your sleeves and work through demanding, real engineering projects, making mistakes and learning from them. That is how you learn to direct agents fluently: knowing which job belongs with which agent, how to break down goals, delegate authority, set boundaries, and review and accept the results.

Every programmer needs to make this shift in identity: **from a foot soldier with their head down writing code to a commander directing the operation.**

In traditional engineering management, leading 8 people made you a team lead; a few dozen, a middle manager; one or two hundred, a senior manager. In the coming era, serious engineers will start by commanding 10 agents, then move up to 100 or 1,000. Your standing will no longer depend on how many lines of code you type. The size of the agent army you can effectively direct will define your new rank and scope.

Looking back, electronic computers did eliminate the human computers with their pencils and paper. But they also ushered in the extraordinary age of software engineering. Among the first programmers were people like Dorothy Vaughan, human computers who embraced the new machines early.

Today, the rise of AI agents is bringing the era of programmers who type code by hand to an end, while opening one in which a single person can be an entire team. Yesterday’s human computers became the first programmers. Today’s programmers can become the first fleet commanders.

Happy 1024 Programmers’ Day, everyone—perhaps our last. And when you wake up tomorrow, welcome to the new world of engineers and commanders.