Old-era coding was scientist-style — logical, linear, zero errors. AI-era coding is artist-style — sketch, run, discard, sketch again. When the cost of a single iteration nears zero, trial and error becomes the engine of creation.
So, you may never have written code. You may have written a lot of it. Either way, you're welcome here — today's point applies to both. The way people have learned coding for the last 50 years is about to change.
This essay walks through why old-era coding was scientist-style, and why AI-era coding has to become artist-style. Every technical term gets unpacked as it appears. Let's go slowly.
Let's start with the principle. When humans build, we tend to move in one of two grains. One is scientist-style. The other is artist-style. Scientist-style means: plan, break it down logically, finish it correctly on the first pass. Like a building's blueprint. Mistakes are costly.
Artist-style is the opposite. Sketch first. Run it. Dislike it. Throw it out. Sketch again. A painter can throw away fifty studies to reach one finished piece. Discarding is part of the process — the esquisse sits in the floor of every atelier.
People pick between the two grains by context. But coding, strangely, has been stuck in only one grain for a long time. Scientist-style. Why?
The reason is simple. Old computers were slow. Running a line of code took time, the machines were expensive, and the programmer's hour was expensive. So the person who got it right in one pass was the good one. One execution cost tens of dollars.
Education followed. In university programming classes, the first rule was "write it without errors." Draw diagrams, write pseudocode, line by line, then let the computer have it. If it didn't run the first time, you were treated as someone who had designed poorly. For 50 years, scientist-style coding was the unquestioned default.
That assumption started to wobble around 2023. Why? The cost of one run dropped to nearly zero.
One example. Building a simple login button used to take roughly 30 minutes frontend, 1 hour backend, 30 minutes integration test. Two hours total. With AI in the loop, the same task drops to about 10 minutes. The code may not be polished, but in ten minutes you have something that runs.
The exact number depends on the tool. It might become 3 minutes, or 2. The ratio is what matters. One run now costs a tenth of what it used to, or less. That one shift flips coding itself.
One example from 2023. ChatGPT recommended three "generative art" coding tools: Processing, p5.js, and openFrameworks. The first two were built for non-specialists. Processing in particular came out of John Maeda's "Design By Numbers" research at the MIT Media Lab — a language built so that designers and artists could make visual work in a few lines of code. Artist-style coding had a path even before AI. The language was waiting.
Think of a kitchen. There are two ways to cook.
One is measured: salt 5 grams, water 200 milliliters, heat 12 minutes. Safe for beginners. Predictable results. Scientist-style cooking.
The other is tasting as you go. Simmer, taste, add, taste again. The more times you taste, the better the dish. Artist-style cooking. The way a grandmother makes soup.
AI coding sits closer to the second. Not one perfect pass — many tastings and fixes.
Try the experiment. To build the same small app, scientist-style runs 60 min plan + 90 min code + 30 min fix = 180 min. Artist-style runs 5 min sketch + 2 min run + 5 min sketch + 2 min run… ten times, about 70 minutes. More than 2.5× faster. And here's the surprise — the quality.
In the same time, the one who attempts ten times more always wins.
That's the aha. Quality in creative work does not come from genius. It comes from the number of trials. When the cost per trial approaches zero, trial and error becomes free. At that moment, artist-style is suddenly the most efficient style.
Ask yourself one question before coding.
"Is this task a blueprint or a sketch?"
Two answers.
Financial transaction systems, medical records, large-scale databases. One wrong step is catastrophic. The old method is correct: plan, verify, test, ship.
Your personal site, a prototype, a small automation script, generative art. Most of the things you personally want to make live here. Running and throwing away is the natural rhythm.
If you're starting today, open Processing or p5.js. p5.js runs in a browser, no install.
// p5.js first sketch — 20 lines and a circle follows the mouse
function setup() {
createCanvas(800, 800);
}
function draw() {
background(20);
fill(201, 100, 66);
circle(mouseX, mouseY, 60);
}
Show this to an AI and say "make it more interesting." Ten seconds later a variation arrives. Don't like it? Throw it away, ask again. Throwing away costs zero.
To wrap up. Old-era coding was scientist-style because one run was expensive. In the AI era that cost is near zero. So coding shifts to artist-style. Whether the tool is Processing or p5.js today, or something else in three years, doesn't matter. Expensive runs → scientist-style. Cheap runs → artist-style. That mapping doesn't bend.
One more thing matters. When artist-style is allowed, anyone can become a maker. The advantage moves from people with the highest logic score to people with the most ideas and the highest willingness to keep trying. Fifty years of being pushed out of coding suddenly reverses for a whole kind of person.
Carry one question — "Is this task a blueprint or a sketch?" That single question ports you from the old habit to the new one. The people who last are not those who write good code — they're those who throw away a lot of code. Technology changes. Principles do not.
Sketch. Run. Again.