Anyone can vibe code

In February 2025, Andrej Karpathy tweeted a phrase that would define an era: "vibe coding." The idea was simple. Tell an AI what you want, watch the code appear, and nudge it until it works. Forget syntax. Forget architecture. Just vibe. A year later, millions of people are doing exactly that. Non-technical founders are shipping SaaS products in a weekend. Designers are building their own apps. Kids are creating games before they learn algebra. The barrier to building software has never been lower. But here's the thing nobody talks about at the hackathon afterparty: building was never the hard part.

The build trap

There's a seductive lie embedded in every vibe coding demo. It goes like this: if you can build it, you can succeed with it. The demo shows someone going from idea to working prototype in minutes. The crowd cheers. Everyone goes home to build their own thing. What the demo doesn't show is what happens next. The prototype sits on a subdomain nobody visits. The Product Hunt launch gets 12 upvotes. The landing page converts at 0.3%. The founder spends three months adding features that nobody asked for, because building is comfortable and everything else feels terrifying. This isn't a new problem. But vibe coding has amplified it by orders of magnitude. When building takes months, you're forced to think carefully about whether anyone actually wants what you're making. When building takes an afternoon, that forcing function disappears. You can ship ten products before ever talking to a single potential customer.

Distribution is the real game

As investor Rick Koleta put it in 2025, getting from zero to $500K in annual recurring revenue is now a distribution challenge, not a technical one. When 83% of early-stage SaaS teams use the same no-code and AI-assisted stack, and most apps can be cloned in 72 hours, the product itself is no longer a moat. Forbes reported in April 2026 that VCs are now evaluating startups primarily through a distribution lens. The question isn't "can your team build this?" It's "can your team get this in front of the right people, repeatedly, at a cost that makes sense?" Distribution means understanding where your audience already spends their attention. It means building relationships before you need them. It means creating content that earns trust over time, not just running ads. It means developing partnerships, integrations, and community loops that compound. None of this can be vibe coded.

Marketing requires taste

Michal Malewicz wrote a piece in March 2026 titled "Vibe Coding is Over," and his core observation was sharp: everything built with AI tools is starting to look the same. The same landing pages. The same gradient backgrounds. The same hero sections with the same stock illustrations. When everyone has the same building tools, differentiation moves upstream. It moves to brand. To voice. To the specific way you frame a problem that makes someone stop scrolling and think, "this person gets it." Marketing in the age of commoditized software is less about features and more about positioning. It's about understanding the conversation already happening in your customer's head and inserting yourself into it with clarity and relevance. This takes deep customer empathy, not just a prompt.

Selling is a human skill

Vibe coding can generate a checkout page. It cannot sit across from a skeptical enterprise buyer and navigate their objections. It cannot build the kind of trust that closes a six-figure deal. It cannot read the room on a sales call and know when to push and when to listen. Even in product-led growth models where the product "sells itself," someone has to design the onboarding flow that converts a free user into a paying one. Someone has to write the email sequence that re-engages churned users. Someone has to look at the data, form a hypothesis, and run the experiment. The human skills around selling, persuasion, relationship-building, and negotiation are becoming more valuable precisely because the technical skills are becoming less scarce.

Polish separates toys from tools

There's a canyon between a working prototype and a product people trust with their money or their data. That canyon is called polish. Polish is the loading state that doesn't feel janky. It's the error message that actually explains what went wrong. It's the settings page that doesn't make users feel stupid. It's the empty state that guides instead of confuses. It's the thousand tiny decisions that add up to something that feels intentional rather than generated. Scott Young, writing about the future of skilled work, made an observation that captures this perfectly: a world where everyone can code is probably one where there's a lot more slop, but also a much higher bar for what counts as genuinely useful software. Vibe coding increases the variance more than the mean. The people who win in this environment aren't the ones who build fastest. They're the ones who care most about the details that users notice but can't articulate.

Iteration requires conviction

Building version one is the easy part. Knowing what to build for version two, that's where it gets hard. Iteration requires talking to users and actually listening, even when their feedback contradicts your vision. It requires looking at usage data and having the honesty to admit that the feature you spent a week on isn't being used. It requires the discipline to kill your darlings and the conviction to keep going when growth flatlines. Vibe coding makes iteration faster in the mechanical sense. You can ship updates more quickly. But it doesn't help you decide which updates to ship. That decision-making muscle, the ability to synthesize qualitative feedback, quantitative data, market trends, and your own product intuition into a coherent direction, is entirely human.

Testing goes beyond unit tests

When people say "testing" in the context of vibe coding, they usually mean automated tests, the kind AI can generate. But the testing that matters most for a product's success is the kind that happens outside the codebase. Does this pricing page actually convert? Does this onboarding flow retain users past day seven? Does this feature increase engagement or just add complexity? Does this positioning resonate with the buyer, or just with the builder? This kind of testing is messy, slow, and deeply contextual. It requires designing good experiments, interpreting ambiguous results, and making judgment calls with incomplete information. AI can help with analysis, but the strategic thinking that surrounds it remains stubbornly human.

The real skill stack

Here's what I think the post-vibe-coding world actually rewards:

Notice that none of these are technical skills. They're the skills that have always separated successful products from side projects. Vibe coding didn't change that. It just made it impossible to ignore.

The optimistic read

None of this is meant to discourage building. Vibe coding is genuinely wonderful. The fact that anyone can turn an idea into a working prototype is one of the most democratizing shifts in the history of technology. More people building means more experiments, more creativity, and more chances for something great to emerge. But if you want to go beyond the prototype, beyond the demo, beyond the "look what I built this weekend" tweet, you need to develop the skills that AI can't replicate. You need to learn how to distribute, market, sell, polish, iterate, and test. Anyone can vibe code. Not everyone can do what comes after. And that's where all the value is.

References