Why 'Vibe Coding' Is Software's Instagram Moment

There's a pet portrait app doing $100,000 a month. It generates Renaissance-style portraits of your dog or cat—Baroque dukes and Flemish noblewomen, shipped as physical prints. The entire premise sounds ridiculous because it is ridiculous. It's also real, profitable, and represents something I've been watching develop over the past few weeks: software creation has stopped being work and started being play.

Nate Jones, who covers AI strategy, calls this "vibe coding"—using AI to build software through natural language. The term itself sounds deeply unserious, which is precisely the point. This isn't about optimizing developer productivity or disrupting enterprise SaaS. It's about what happens when the friction cost of building software drops low enough that people start making things for fun.

I've watched enough tech cycles to recognize when something genuinely shifts. The smartphone made everyone a photographer. Instagram made sharing those photos frictionless. Together they created an entire ecosystem of casual creators who never would have touched a DSLR. What Jones describes—and what I'm seeing in corners of Twitter and Discord—looks like software's version of that same moment.

The Gate That Finally Opened

For most of software's history, building something meant crossing a gap that took years of specialized training. You had an idea, but between "I wish this existed" and "I made it exist" lay data structures, APIs, deployment pipelines, and a thousand other concepts that required serious commitment to learn. Most ideas died in that gap not because they were bad, but because the activation energy was too high.

AI coding tools have existed since early 2024, but they were still work. You fought with context windows, babysat the AI through confusion, debugged failures that made no sense. The friction was high enough that you needed to be serious about what you were building.

What's changed in recent weeks isn't any single breakthrough—it's that several improvements stacked together reached a threshold. Models hold context longer. Agentic patterns matured. Builder platforms got more reliable. "The friction has now dropped enough that building software has stopped feeling like work and started to feel like play," Jones says. "And play produces very different things."

That pet portrait app—called Fable—wasn't born from market research or a validated business hypothesis. Jones describes it as a "wouldn't it be funny if" story. Someone was playing. They built the joke. The internet turned out to have demand for it.

Software Vision Versus Software Skills

Here's the interesting filter: when you tell people they can now build any app they want, most say "cool" and never think of anything to build. Not because they lack creativity, but because most people's problems aren't software-shaped, and most don't notice when they are.

Jones borrows a concept from parkour called "parkour vision"—the trained ability to see walls as surfaces you can run along, gaps as spaces to jump through. Programmers develop "software vision" the same way. They look at a repetitive task and think "I can script this." Everyone else just keeps doing it manually.

The people taking to vibe coding aren't necessarily technical. They're people who already have software vision or develop it quickly. If you've ever spent an afternoon building a complicated spreadsheet to solve a problem, you probably have some of this disposition. If you've duct-taped together automations in Zapier or bent a tool to do something it wasn't designed for, that's the mindset.

What's genuinely new is that this vision—this ability to spot software-shaped problems—now matters more than coding ability. The skill gap between "I see the problem" and "I built the solution" has collapsed to something crossable in a weekend.

The Two Ways This Goes Wrong

Jones is honest about failure modes, and they're worth understanding because they're not the ones people expect.

The first is moving so fast you never stop to think. When building becomes instant, the bottleneck shifts to knowing what you actually want. "It's very easy to prompt before you figure that out," Jones notes. You generate piles of features that don't fit together, end up hip-deep in a project that serves no clear purpose. The build-test-iterate loop is so fast it becomes intoxicating for its own sake.

The discipline here is old-fashioned: pause and describe what you want before you start prompting. Know why you're building it, what success looks like. The tools will happily turn vague intentions into working code, but it might not be your idea of working code.

The second failure mode is confusing "works on my laptop" with "ready for users." AI has compressed the cost of creating software toward zero. A working prototype takes minutes or hours. But AI doesn't compress the cost of owning software in production. Someone still has to answer for it.

For personal projects—your greenhouse automation, your friend group bot—this doesn't matter. The stakes are low and imperfection is fine. But if you build something that users depend on, even for silly dog pictures, the gap between prototype and production-grade is still real. Security researchers have found that roughly 10% of apps built on popular vibe coding platforms have vulnerabilities—databases exposed to the public internet, API keys visible to anyone who looks.

Jones points out that platforms like Lovable are running the Shopify playbook: start with "you can vibe code anything" and help users grow up over time. The bridge toward production is extending—authentication, security, scaling infrastructure—but it's not all the way there yet.

What Actually Matters Now

The valuable skill, Jones argues, isn't coding anymore. It's specification. Experienced developers already know this. They know how to break problems into pieces, what questions to ask (what happens when a user isn't logged in, what if the database is slow). Beginners tend to prompt vaguely and accept whatever the AI generates.

"You don't need to be a professional developer to walk across this bridge," Jones says, "but you do need to develop enough intuition to specify clearly and evaluate with a degree of critical thinking."

That's a much smaller gap than learning to code from scratch. It closes faster with practice. You need to build intuition for software-shaped things—start small, notice what goes wrong, develop a sense for what questions to ask.

I find this framing useful because it's neither hype nor doom. It doesn't claim developers are obsolete or that everyone will become a builder. It notices that something has shifted—building software has become accessible enough to support a hobbyist ecosystem—and describes what that shift enables and where it breaks.

History Doesn't Repeat But It Rhymes

I remember when desktop publishing was going to eliminate graphic designers. When digital photography was going to destroy professional photographers. When websites builders were going to replace web developers. The pattern is consistent: when creation tools become accessible, the professional tier doesn't disappear. It gets joined by an amateur tier that makes different things—more playful things, weirder things, occasionally things that turn out to have real commercial value.

What Jones describes—designers building personal dashboards that show moon phases and vacation countdowns, retirees automating greenhouse irrigation, someone making a browser extension that logs every article mentioning cat grooming—these aren't businesses. They're projects built for the satisfaction of building, used by the maker and maybe a few friends.

This is new for software. Hobbyist programmers technically existed before, but "I code for fun" was a niche identity, not a casual weekend activity. For most of software's history, building required enough specialized knowledge that it was primarily professional.

Jones keeps returning to three conditions that haven't been simultaneously true before: building software is inherently satisfying (that's always been true but was previously gated), the internet has nearly infinite demand for interesting things (also true but expensive to discover), and the cost of building something is approaching zero for hobby-scale software (genuinely new, and getting newer).

Put those together and experimentation becomes cheap enough to be playful. You can try the idea. If it doesn't work, you didn't lose much time. If it does, maybe the internet loves your dog pictures.

I don't know if this becomes culturally significant or stays niche. What I notice is that the tools are going mainstream quickly, and "I built a little app for that" may soon become as common as "I made you a spreadsheet." That's not revolutionary, exactly. It's just different—the kind of different that accumulates in ways that are hard to see until they're everywhere.

—Mike Sullivan, Technology Correspondent