Vibe Coding Just Grew Up—And Nobody Knows What It Is Yet

Vibe coding just turned one year old. In tech years, that means it's now completely unrecognizable from what it started as.

This week's product announcements from Perplexity and Replit aren't just iterative updates—they're signaling a phase change in how we think about building software. Or maybe more accurately, they're revealing that we still don't really know what we're building toward.

The Everything Machine Problem

Perplexity launched Computer for Enterprise this week, building on their February announcement of Perplexity Computer. The pitch is ambitious: an "AI everything machine" that creates and executes entire workflows that can run for hours or months. You don't tell it what to do—you describe an outcome and let it figure out the rest.

The system breaks goals into tasks and subtasks, spinning up agents and sub-agents to handle everything from research to document generation to interacting with connected services. When it hits a problem, it's supposed to just solve it.

Then they announced Personal Computer—essentially their version of what the community has been calling "OpenClaw." It's an always-on system running on a Mac Mini, with persistent access to your files, apps, and sessions. You can control it from any device, anywhere.

The Mac Mini integration is telling. Apple accidentally stumbled into an AI strategy by making a cheap, always-on box that every AI company now wants to colonize. It's almost funny.

When a Coding Platform Stops Being About Code

Replit's Agent 4 takes a different but parallel approach. Paul Graham tweeted a couple weeks ago that Replit founder Amjad Masad showed him something representing a "real paradigm shift." Agent 4 is apparently that thing—Graham said it "generalizes the idea of vibe coding beyond what people usually think of as coding."

What does that mean in practice? Agent 4 is a collaborative canvas where individuals, teams, and agents work together. But it's not just building websites anymore—it's sites, slides, videos, and whatever else you can describe in plain English.

The interface has evolved too. Instead of one chat bar and one artifact field, you get direct manipulation tools, annotation capabilities, and the ability to edit specific parts in natural language. You can run multiple tasks in parallel rather than sequencing them. Multiple people can work on the same file simultaneously.

Latent Space captured the significance well: "Replit is unrecognizable from the coding-with-some-AI-tacked-on platform that Replit was just two years ago." Now that coding is "approximately solved," these platforms are moving up the stack into broader knowledge work.

What's Actually Changing

Both announcements point to several converging trends. First, we're seeing blended user experiences—no more choosing between chat interfaces or traditional input methods. Everything exists on extensible canvases with multiple interaction modes at once.

Second, persistent context. Perplexity's Personal Computer gives agents 24/7 access to your systems, which solves some of the memory problems that have limited what agents can do. It also raises every security concern you're already thinking about.

Third, multi-agent orchestration. These aren't single agents anymore—they're teams of agents that get spun up dynamically based on what you're trying to accomplish.

Fourth, multiplayer mode is becoming standard. Perplexity Computer integrates with Slack. Replit Agent 4 supports real-time collaboration. The solo developer superhero narrative is giving way to tools built for how teams actually work.

Interestingly, Perplexity is charging enterprises based on usage rather than seats. Their reasoning: generating a video costs vastly different amounts than generating a memo, so why pretend all users cost the same? It's a recognition that variable workloads need variable pricing.

In a VentureBeat article, Perplexity's head of business Dmitri Shevelenko said: "With no hyperbole, the introduction of computer inside Perplexity was the single biggest productivity unlock in our entire history as a company."

That's notable because Computer started as an internal tool—a Slackbot for Perplexity's own employees before they considered launching it publicly. They're now competing against Microsoft's Copilot and Google's Gemini workspace integrations with a product that was originally just meant to help them ship faster.

The Pivot Problem

Here's what strikes me about both announcements: they represent companies that are comfortable completely reinventing what they are. Two months ago, people were writing Perplexity off. Now there's excitement again—not because they made a better search product, but because they were willing to say "search doesn't matter, this does."

The traditional startup wisdom was that you pivot on your way to product-market fit. Once you find it, you execute. But these announcements suggest that nimbleness and pivoting might just be permanent operating conditions now. The pace of change in this space doesn't allow for the luxury of settling into a stable identity.

Replit went from a coding platform with some AI features to a full productivity suite for digital creation in roughly two years. That's not iteration—that's transformation.

What We Don't Know

Andre Karpathy tweeted something relevant this week: "Expectation: the age of the IDE is over. Reality: we're going to need a bigger IDE. It just looks very different because humans now move upwards and program at a higher level. The basic unit of interest is not one file but one agent."

He's right that we're still programming—just at a different level of abstraction. But what's the right interface for programming teams of agents? What are the debugging tools? The version control? The testing frameworks?

Perplexity Computer and Replit Agent 4 are making ambitious bets on what these interfaces should look like. But they're also admitting, implicitly, that we don't actually know yet. These are glimpses, not destinations.

The security questions around persistent system access remain largely unsolved. Everyone pitches their version as "secure," but giving any system—AI or otherwise—full access to your files and applications is a meaningful architectural decision with meaningful risks.

The governance questions around multi-agent systems are similarly open. When an agent team running for hours or months encounters an ambiguous situation, who decides what happens? What are the human checkpoints? What work should require explicit approval versus autonomous execution?

The pricing models are still being figured out. Perplexity's usage-based approach makes sense given variable costs, but it also makes budgeting harder for enterprises used to per-seat predictability.

And perhaps most fundamentally: what is vibe coding actually becoming? A year ago, it meant describing what you wanted and having code appear. Now it means... what? Describing what you want and having any digital artifact appear? Managing teams of persistent agents that do knowledge work? Something else entirely?

The one-year anniversary of vibe coding might be less about celebrating what it's become and more about acknowledging how little we understand about where it's headed. These tools are getting powerful fast. What we build with that power—and what it costs us to build it—remains very much in question.

—Dev Kapoor, Open Source & Developer Communities Correspondent