From Instructions to Vibes
In early 2025, Andrej Karpathy — one of the people who literally built modern AI — coined the term "vibe coding." It meant writing software not by typing code, but by describing what you want and letting AI figure out the implementation. It sounded like a novelty. Then 25% of Y Combinator's Winter 2025 batch turned out to be startups where the codebase was 95% AI-generated. A New York Times journalist with zero coding experience built a functional app by talking to Claude. Vibe coding wasn't a gimmick. It was a preview of how all knowledge work is about to change.
Welcome to vibe working.
What Vibe Working Actually Means
Vibe working is the shift from telling AI exactly what to do ("write a 500-word email about our Q4 results with a professional tone") to telling it what you're trying to achieve ("I need to convince a skeptical CFO to fund an AI pilot — the email should feel confident but not pushy, and it needs to acknowledge their concerns about ROI without getting defensive").
The difference is enormous. The first prompt gives AI a task. The second gives it a context, an audience, a relationship dynamic, and an emotional register. The output from the second is almost always better — because you're communicating intent, not just instructions.
This is possible now because AI models have gotten dramatically better at understanding what humans mean, not just what they say. Claude's context window hit 1 million tokens in beta. It can hold an entire project's worth of context in memory. It's not just processing your words — it's understanding your situation.
Why This Matters for Non-Technical Professionals
Here's what most AI coverage gets wrong: they frame every breakthrough as a tech story. Vibe coding was covered as a developer tool. But the real revolution is that the same principle applies to every knowledge worker.
If you're a marketer: instead of "write me 5 social media posts about our new product," try "our audience is exhausted by corporate AI announcements — write something that makes them stop scrolling because it's unexpectedly honest."
If you're in HR: instead of "write a job description for a senior analyst," try "we keep losing candidates at the final round because they think we're too corporate — write a JD that signals we're serious but human."
If you're a consultant: instead of "summarize this 200-page report," try "my client is a CEO who has 10 minutes and is already skeptical of consultants — what are the 3 things from this report that would change her mind?"
In each case, you're not giving AI more instructions. You're giving it better context. And the quality gap between generic instructions and rich context is getting wider every month.
The Skill That Matters Now
The new literacy isn't prompt engineering. It's intent articulation — the ability to clearly communicate not just what you want, but why you want it, who it's for, and what good looks like. This is fundamentally a human skill. It requires understanding audiences, reading emotional dynamics, and having taste.
People who've spent their careers doing nuanced, human-centered work — teachers, therapists, salespeople, designers, leaders — may be better at this than engineers. Because they've been articulating intent to other humans for decades. AI just became the newest audience for that skill.
Your Move
This week, try giving Claude a vibe instead of a spec. Don't tell it what to write — tell it what you're trying to accomplish, who the audience is, and how the output should feel. Compare the result to your usual approach. If it's better, you just discovered your new way of working. If it's not, refine the vibe — add more context about the relationship dynamics, the unspoken concerns, the emotional undertone. That refinement process IS the skill.
The best prompt isn't the most detailed. It's the one that communicates the most intent.