Vibe Coding at One: From Meme to Movement

February 2, 2025. Andrej Karpathy fires off a tweet that would define the year: "There's a new kind of coding I call 'vibe coding,' where you fully give in to the vibes, embrace exponentials, and forget that the code even exists."

"I just see stuff, say stuff, run stuff, and copy paste stuff, and it mostly works... I'm mass mass mass producing code by just asking for it. I don't really care what the code is. I 'Accept All' always."

He described accepting all diffs without reading them. Copy-pasting errors with no comment. Code growing beyond comprehension. "Random changes until it goes away." His caveat? "Not too bad for throwaway weekend projects."

One year later, vibe coding went from meme to Merriam-Webster to Collins Dictionary Word of the Year. And at Soapbox, we've spent that year building the tools to make it work—not just for throwaway projects, but for real applications that matter.

This is the story of how we got here.

The Year of Vibes

The models got dramatically better. When Karpathy coined the term, GPT-4.5 scored 38% on SWE-bench. By August, GPT-5 hit 74.9%. Claude Opus 4.5 now exceeds 80%. Extended reasoning, agentic tool use, multi-agent systems—the "random changes until it goes away" approach started becoming something more sophisticated.

The vibes had gone mainstream.

Building for the Vibes

At Soapbox, we saw this coming. Our journey started in January 2025 when Block released Goose, an open-source AI agent framework. We saw the potential immediately: AI agents that could actually build software, not just autocomplete it.

Chorus: The Proof of Concept

In June, a group of us gathered for the Oslo Freedom Forum with an ambitious goal: build a Facebook-style groups app with an eCash wallet. The team included Calle from Cashu, JeffG from White Noise, Rabble and Liz from Nos, and us from Soapbox. Jack Dorsey advised.

We shipped in five days.

"When vibe coding, the human's role is not to write the code or to even care about the code. The role of the human is to be the visionary that imagines the end goal of the product, and prompts it into existence."

Chorus proved that vibe coding wasn't just for throwaway projects. It could ship real software for real activists at a human rights conference.

Shakespeare Takes the Stage

In July, we launched Shakespeare—an AI-powered app builder that runs entirely in your browser. No backend servers. Everything in IndexedDB. Pay anonymously with Lightning. AGPL licensed.

Within a week, we had 200+ users and TechCrunch coverage. But we were just getting started.

By November, we added image vision, MCP server integration, plugins, and configurable system prompts. Shakespeare wasn't just a vibe coding tool—it was becoming a full development environment that respected user autonomy.

OpenCode and the AI Hackathon

As Shakespeare matured, we started exploring how it could integrate with other tools in the AI coding ecosystem. OpenCode—a powerful terminal-based AI coding environment—caught our attention.

We added OpenCode Zen as an AI provider in Shakespeare, letting users tap into OpenCode's capabilities directly from our browser-based environment. But the real test came in January 2026 at the AI Hack for Freedom hackathon.

Our team built Pathos—a censorship-resistant platform for activists—in 48 hours using Shakespeare and OpenCode together. Venezuelan activists were using it within hours of launch. The combination of Shakespeare's browser-based simplicity and OpenCode's terminal power proved remarkably effective for rapid prototyping under pressure.

We're now working on tighter OpenCode integrations, exploring how these tools can complement each other. Shakespeare excels at accessibility and freedom; OpenCode excels at power-user workflows. Together, they cover more ground than either could alone.

Meet Quilly

But here's where it gets interesting. What do you do when you've built AI development tools and want to truly understand them?

You give the AI a seat at the table.

It sounds like science fiction, but it's surprisingly practical. Quilly wrote and published their own introduction blog post. The future arrived quietly.

Digital Spaghetti: Why Nostr Makes This Work

We call vibe coding "digital spaghetti" because you can throw code at the wall and see what sticks.

Build an app. Ship it. See if it resonates. If it doesn't, throw another version at the wall. Iterate rapidly, experiment freely, learn from what grows and what falls flat. Keep throwing until something sticks—then refine it.

Traditional software development punishes this approach. Technical debt accumulates. Rewrites are expensive. You're stuck maintaining code you don't fully understand.

But Nostr changes the calculus.

The protocol is the architecture. The app is just vibes. Throw it at the wall and see what sticks.

The Producer Mindset

There's a concept from music production that applies here: the producer isn't the one writing every note. The producer's job is taste, direction, knowing when it's done. They shape the song without playing every instrument.

Vibe coding inverts the traditional developer role in the same way. You're not the artist writing code—you're the producer shaping what the AI creates. Your job is vision, judgment, knowing when to accept the diff and when to ask for another take.

"I'm frankly exhausted by the end of the day... it's a deeply intellectual exercise."

Good vibe coding isn't passive. It's deeply creative work—just not the kind of creativity we traditionally associated with programming.

Looking Forward

Where does vibe coding go in year two?

The models will keep improving. Multi-agent systems will become standard. Autonomous coding sessions will stretch longer. The vibes will get more sophisticated.

Karpathy was right about one thing: this isn't too bad for throwaway weekend projects. But we've spent a year proving it's good for a lot more than that.

The vibes are strong. Come build with us.