Design Once, It Runs Forever

Here's a pattern most people live in: you do the same task every day. Redo the same report template, redo the same calculation, redo the same reply. Each one is five minutes. But those five minutes stack up. Two and a half hours a month. Thirty hours a year. Why does this happen? The answer is in an old principle — we keep making outputs, not designs.

This essay explains that distinction all the way through. Claude Live Artifact is today's example, but the principle works with or without AI. Three years from now, when "Live Artifact" is replaced by something else, the spine of this essay still holds. We'll go slowly.

Outputs and designs are different

Let me separate two words first. An output is something you use once and it's gone. Today's report, yesterday's reply, last week's spreadsheet. Each one is consumed once and disappears.

A design is different. A design is the method that produces outputs. The report template, the reply framework, the spreadsheet formula. Make it once, and it will generate a hundred, a thousand outputs for you.

Every task in the world is one of these two. Strangely, most people only make outputs. Starting from scratch every day. Why? Because making a design looks harder upfront. One output takes five minutes. A design takes thirty. Cheap looks cheap in the moment. But a year later, the output-maker spent thirty hours. The design-maker spent thirty minutes.

Example — Claude Live Artifact

Let's make the principle concrete. Claude has a feature called Live Artifact, launched in April 2026.

Before, if you asked Claude "turn this data into a table," Claude would return a table and that was it. Next time you needed a similar table, you asked again. You were collecting outputs.

Live Artifact flipped this. It publishes the table-making tool itself as a public URL. Swap in new data, and the same design produces a new table. Send a friend the link, and they can use it with their own data.

The difference laid flat:

Let me put numbers on it one more time. Say a 5-minute task, once a week. Over a year: 52 × 5 = 260 minutes, roughly 4 hours 20 minutes. Design the same thing once with Live Artifact in 30 minutes, and the yearly cost drops to "click the link × 52" — about 5 minutes. A 4-hour-15-minute difference, from a single 5-minute task. Ten such tasks — 42 hours. A hundred — a full week returned.

What Live Artifact really did is one thing. It nudged users to make designs instead of outputs. That's the whole trick. Whatever feature comes next from any AI company, this direction keeps going. The principle is already correct.

Analogy — kitchen recipes

Imagine two cooks.

Cook A makes kimchi stew from scratch every day. Today a little salty, tomorrow a little bland. The flavor drifts. When a guest says "yesterday's was better," she can't reproduce it. It lives only in her head.

Cook B spent three hours on day one writing the recipe. Two tablespoons of doenjang. 300 grams of kimchi. High heat five minutes, medium fifteen. Once that single sheet exists, thirty minutes gives the same flavor every day. Bring in a new helper — same flavor. Hand the recipe to a guest — they can make it too.

Those three hours pay off forever.

• Cook A = the output-maker. Starting fresh each day.
• Cook B = the design-maker. Made once, works on its own.

Here's the first aha moment.

If you're going to do it twice, make the design, not the output.

One question to sort the work

How do you know which tasks should turn into designs? One question.

"Am I doing this three or more times this month?"

Three or more: turn it into a design. One or two: just do it once and move on. Why three? Because two won't recover the cost of designing. Starting with the third, profit begins.

Concrete example. Every week I resize YouTube thumbnails to 1280×720. That's four times a month. Design target. One Live Artifact — drag in any image, auto-crop returned — and now it takes three seconds.

Last year I made a wedding invitation. Once. Not a design target. Just ship the output.

I run this question every Friday afternoon. I list every task I did three or more times that week, then mark which haven't been turned into designs yet. Usually one per week becomes a Live Artifact. Fifty-two a year. Every week's 30 minutes keeps paying itself back from the next week onward.

What you can do today — a 30-minute experiment

Block 30 minutes.

First, name one task you've done twice this week. Email reply template, calculation, schedule formatting, anything.

Second, open Claude and ask:

I do this every week. Don't give me the output.
Give me a Live Artifact tool where I swap in data
and the result comes out automatically.

[task + sample data]

Third, bookmark the URL. Next week, same task — just open the link. Thirty minutes replaces a year of repetition.

The first time you try this, you'll feel something strange. "Wait — is that really all?" Yes. It only felt normal to make outputs because tools couldn't easily produce designs. Now they can.

Summary

Outputs are used once and gone. Designs, made once, work forever. Today we explained this through Claude Live Artifact, but it applies to whatever AI tool comes next, and to your work life with or without AI.

Keep one question in your pocket — "Am I doing this three or more times this month?" Three or more, make the design. Build it once, and that design does a hundred jobs on your behalf.

Three words to remember: Once. Fully. No repeat. Those three change how your days feel. The technology changes. The principle does not.