Prototyping is not shipping

Almost half of designers generate half or more of their output with AI. Only 1.4% trust that output without reviewing it. This article is about why both things make sense together.

The first time I read these numbers, I thought contradiction. Today I think they describe, precisely, what a prototype is.

The survey question is about production, and the most common answer is “first draft, I edit heavily” (34.2%). Add “exploration only” (29.2%) and 63.4% treat what the AI generates as raw material, not as a deliverable.

I grouped the six answers into three blocks:

• Draft or exploration, 63.4%. The AI writes, the person rewrites or throws it away.
• Ship with review, 34.2%. “I review before sending” (24.7%), “ship with small tweaks” (8.1%) and “full trust” (1.4%).
• Doesn't use it, 2.4%. A smaller group than the anti-AI discourse suggests.

The middle block is the interesting one. A third of the profession already operates in “the AI writes, I review and release” mode. That's where the conversation about quality and accountability needs to happen.

Cross this article with the previous one. A large group generates a lot of code and trusts it very little. That only makes sense if the generated code lives mostly in prototypes, proofs of concept and internal tools, not in production.

The survey is called State of Prototyping and the name is literal. Vibe coding didn't replace engineering. It replaced the clickable prototype. What used to be a Figma screen pretending to work is now an app that actually works, and is still disposable.

That's an enormous change for people who do design. The prototype now has real behavior, real data, real bugs. And the distance between prototype and product got, at the same time, shorter and more dangerous.

59.1% built a tool of their own with AI-generated code in the last six months. Add the people who want to build one and it's 89.5%. Only 10.4% dismiss the idea. The survey's own line: “more than half of designers are now builders. Not for the company. For themselves.”

If the AI writes and the person reviews, the skill that sets apart a designer who codes isn't writing code. It's knowing what to accept, what to rewrite and what to throw away.

That demands things most design courses don't teach: reading code you didn't write, recognizing a fragile pattern, actually testing a flow, deciding when the prototype turned into debt. The good news is that designers critique other people's work all the time. Reviewing generated code is a design critique with a different material.

What from your last prototype would you let go to production without a review?

What to do on Monday: write down, on one page, your team's review criteria for AI-generated code: what always gets read line by line, what only needs to be tested, what never goes to production without an engineer. Then explicitly label every prototype as “disposable” or “product candidate.” Confusing the two is where the debt is born.

The people who generate the most code with AI feel more valuable. The people who generate the least feel less secure. The third article is about the group that is 58% of the profession and the only one, among designers, with more fear than confidence.