You're Already a Vibe Architect (You Just Don't Know It Yet)

A new piece of research from NNGroup just dropped, and if you're a designer who's been quietly building things with AI tools in the evenings or tinkering with automations for your team, you need to read this. Because it describes you — or at least who you're becoming.

The NNGroup team studied a small group of non-developers who are building surprisingly complex AI-powered systems without any formal engineering background. They're calling these people vibe architects. And spoiler: a product designer at a financial institution building personal software tools on weekends is literally one of their study participants.

Let's unpack what this means for you.

What Is a "Vibe Architect," Exactly?

A vibe architect, as NNGroup defines them, is someone who builds complex, proactive agentic AI systems — that's AI that takes actions and makes decisions on your behalf, not just answers questions — using mostly instinct developed through experimentation, YouTube rabbit holes, and Reddit threads.

These aren't people who understand the code under the hood. They're people who've developed a feel for how to get things done with AI tools like Claude. They're designing how information flows, how tasks get handed off between AI agents (think: multiple AI "workers" coordinating), and how entire team workflows get restructured. Often without being entirely sure how it's working.

One participant in the study had built a Claude-based "operating system" for his whole team — replacing meetings, sharing information, and helping with decisions. Another was running a marketing startup on a multi-agent coordination system. These aren't toy projects.

Your Design Instincts Are a Superpower Here

Here's what's fascinating about the NNGroup findings: the thing that makes vibe architects effective isn't technical knowledge — it's systems thinking and pattern recognition built through time spent with the tools.

Sound familiar? That's basically what UX designers do all day.

You already think about information flows, user journeys, and how different pieces of a product connect. When you start applying those instincts to AI-powered systems — asking "how should this agent hand off to that one?" or "what does this workflow need to trigger?" — you're doing vibe architecture. You're just using design language instead of engineering language.

The research suggests that the most effective vibe architects aren't necessarily those who understand the technical details. They're the ones who can clearly articulate what they want the system to do, which is exactly the skill you've been sharpening.

The Real Risks: Decay and Blind Trust

Now for the part we'd be doing you a disservice to skip over. The NNGroup research doesn't just celebrate vibe architects — it also flags some real, practical pitfalls.

Systems decay. One participant described how his team's Claude-based workflow would hold up "for a few weeks, and then the decay hits." Connections expire, context drifts, and things stop working. Maintaining these systems becomes an ongoing time tax. If you're building for your team — not just yourself — factor in the cost of upkeep from the start.

Instincts aren't always accurate. One participant had developed a habit of "resetting" her AI chats by copy-pasting entire conversations into a new prompt, believing it saved tokens (a token is roughly a unit of text the AI processes — more tokens mean higher cost and slower responses). It actually didn't help. The point isn't to mock her — it's a reminder that the mental models we build through experimentation can be subtly wrong, and worth questioning.

Blind acceptance is a real risk. Multiple participants in the study admitted to clicking "Accept" on AI permission requests without reading them. One said she reasoned that the AI only had access to one folder, so it was probably fine. Another reportedly developed carpal tunnel from repeatedly clicking accept. This is worth sitting with: when your AI agent asks for permission to take an action, taking two seconds to actually read what it's asking for is worth the habit.

You Don't Need to Prompt-Engineer. But You Do Need to Direct.

One genuinely encouraging finding from the research: the latest AI models are good enough at understanding natural, messy, stream-of-consciousness instructions that many vibe architects don't carefully engineer their prompts at all. Some even use voice dictation apps to just talk at their AI tools.

The "engineering" that does matter, the research suggests, is higher-level: choosing the right tools, deciding on the right approach, and knowing what questions to ask. One participant's move of regularly asking Claude "What would make this world-class?" — and then acting on the suggestions — is a great example. That's a design director's instinct applied to AI collaboration.

What to Take Away

The NNGroup research is early and based on a small sample (7 participants, most from design and product fields), so take specific conclusions with appropriate skepticism. But the broad signal is strong: non-engineers are building real, complex, impactful systems with agentic AI tools, and designers are among the most naturally suited to do this well.

The skills you already have — systems thinking, articulating intent, sweating the details of how things flow — transfer directly. What you're adding is a new medium to design in.

So if you've been hesitating to go deeper with tools like Claude because you feel like "that's an engineering thing" — the research suggests otherwise. You're not an engineer learning to design. You're a designer learning to build. That's a different thing, and it's entirely within reach.

Start experimenting. Log what works. Question your instincts occasionally. And maybe read the permission requests before you click accept.