The Rise and Hangover of Vibe Coding And Why the Design Engineer Isn’t the Cure

In February 2025, Andrej Karpathy, co-founder of OpenAI and former AI lead at Tesla, posted something that felt like a joke. He called it “vibe coding”: fully giving in to the vibes, embracing exponentials, and forgetting that the code even exists. “I just see stuff, say stuff, run stuff, and copy-paste stuff,” he wrote, “and it mostly works.”

It did not feel like the birth of a movement. It felt like a sleep-deprived founder describing a bad habit. And then almost instantly, half the internet recognized themselves in it.

The Tweet that Launched a Thousand Startups

The timing was not accidental. In early 2025, AI coding tools had crossed an invisible quality threshold. Models could reason about architecture, debug across multiple files, and generate production-quality code without hallucinating every other function. The tooling had caught up to Cursor, Bolt.new, Replit Agent giving AI agents actual access to a terminal, a browser, and a deployment key. Not just autocomplete. Autonomy.

The result was a Cambrian explosion. Non-programmers were shipping full-stack apps over a weekend. Solo founders were replacing entire engineering sprint cycles with a single well-crafted prompt. By March 2025, Merriam-Webster had listed “vibe coding” as a trending expression. By year’s end, Collins English Dictionary named it Word of the Year.

6,700%  Search spike, spring 2025     ·     41%  Of all code AI generated     ·     95%  AI written, YC W25 codebases

Among Y Combinator’s Winter 2025 batch, 25% of startups had codebases that were 95% AI generated. GitHub Copilot crossed $2 billion in annual recurring revenue. Cursor’s parent company went from a $2.6 billion valuation to $29.3 billion in under a year. The investment world was not hedging. It was all-in.

“The era of the syntax-obsessed developer is fading, replaced by a world where the best code is the code you never had to see.”

The Hangover

By Q4 2025, the industry had a name for what came next: the Vibe Coding Hangover. Applications built purely through vibes started showing their seams. The code looked right. It ran. And then, quietly, it started to fail in ways that were hard to see and harder to fix.

01  Security holes at scale.  ~45% of AI-generated code failed security tests, including OWASP Top 10 vulnerabilities. AI-co-authored pull requests had 2.74× higher vulnerability rates than human-written code.

02  The shadow bug problem.  AI produces code that looks perfect but contains deep structural flaws. If you don’t understand the vibe you asked for, you can’t fix it when the vibe shifts.

03  Open source erosion.  Vibe coding quietly undermined OSS sustainability. Tailwind CSS saw documentation traffic drop ~40% even as downloads rose. Revenue: down close to 80%.

04  Technical debt at warp speed.  Teams could build faster than they could validate. Compliance-heavy industries found AI speed became a liability without engineering discipline to contain it.

Enter the Design Engineer: The Next Big Idea

Just as the vibe coding hangover settled in, the founders' community started floating a new concept. Meet the Design Engineer: part designer, part technical operator, directly connected to the final output. The pitch is compelling.

The old workflow design in Figma, hand off to developers, wait, iterate, repeat, creates enormous friction. Every handoff loses fidelity. Every back-and-forth costs days. The design engineer collapses this into a single loop.

OLD WORKFLOW   →   DESIGN ENGINEER WORKFLOW

✕  Design in Figma            →  Create rough concept

✕  Hand off to developers            →  Use AI to generate UI

✕  Wait for implementation            →  Work directly in codebase

✕  Review, request changes            →  Iterate in production

✕  Repeat            

The design engineer owns the entire product front end, from information architecture to final implementation. They enforce design standards. They monitor usage signals. They reduce friction without waiting on anyone else. On paper, it’s the most efficient creative loop ever devised.

“Designers must move closer to code and use AI directly to generate production-ready work.”

Sounds familiar, doesn’t it? Replace “developer” with “designer,” and you’ve essentially described the same promise vibe coding made twelve months ago. And like vibe coding, it will run headlong into the same wall.

The 90% Problem and Why the Last Mile is Everything

Here is the brutal math of software design. AI can get you to 90% remarkably fast. With Claude and a solid design system, Tailwind, Shadcn, Material UI, you can have something that looks like a product in an afternoon. It will be functional. It will be coherent. It will feel almost right.

Almost.

AI + design engineer: 90% of the way there     ·     With human designer polish: 99.99%

The cost of going from 90% → 99.99% is roughly equal to 0% → 90%. But that last mile is what determines whether users stay or churn.

That gap, the 10% between “functional” and “delightful,”  is where products are won and lost. It lives in the micro-interactions. The weight of a hover state. The rhythm of a transition. The way an error message makes you feel. These are not details you can prompt your way into. They require a human who cares.

The design engineer role, like vibe coding before it, will stall at 90% and mistake arrival for completion.

The Deeper Problem: The Founder Already is the Design Engineer

There is a second reason this role will struggle to find its footing, and it is more fundamental than polish.

The best person to make design decisions for a product is the one who understands the market and users most deeply. In an early-stage company, that is almost always the founder. And today, a founder with Claude, a Tailwind component library, and a clear product vision can reach 90% on their own. No designer. No front-end engineer. Just intent, translated directly into the interface.

So what exactly does the design engineer add? They sit between the founder, who has the judgment, and the AI, which has the execution. They are a layer of interpretation inserted into a process that is increasingly direct. The value proposition hollows out precisely as the tools improve.

→  Founders can already get to 90%.  Claude + Tailwind + shadcn is a design system most founders can wield themselves. The design engineer’s core value proposition disappears when the tools democratize it.

→  Judgment lives with the founder.  No designer or front-end engineer has the market context a founder carries. The person who decides what the product should feel like needs to understand what the product is for.

→  The last 10% still requires a specialist.  The final polish, the kind that makes users love a product instead of merely using it, requires a dedicated human designer with deep craft, not a hybrid role split between code and concept.

Where it Actually Lands

Vibe coding didn’t fail. It matured. The same will likely be true of the design engineer. The role will not disappear; it will narrow. It will be useful in specific contexts: mature product teams that need someone to own front-end quality without full design-developer separation, or companies scaling past the founder-as-designer stage but not yet large enough for dedicated specialists.

But as a structural answer to the design and development gap? It faces the same ceiling vibe coding hit. AI gets you most of the way there. The founder has the judgment. The specialist has the craft. The design engineer, caught in the middle, will need to clearly articulate what they bring that neither of those alternatives can provide.

So far, the answer has been: speed and ownership. That is real. But it is not enough to justify a distinct role in a world where the tools keep closing the gap.

“The last mile is not a technical problem. It’s a taste problem. And the taste, so far, does not scale.”

The era of forgetting that the code exists is over. The era of forgetting that taste matters never began.