Crime Pays but Botany Doesn't: The Vibe Coding Trick That Gets Results

Crime Pays but Botany Doesn't: The Vibe Coding Trick That Gets Results

If you've tried vibe coding — building software by talking to an AI instead of hand-writing code — you know the painful pattern. Every project starts the same: a clean sidebar, a card grid, a purple gradient. The AI seems to be stuck on a single "modern SaaS" template.

That changes when you feed the model something genuinely strange. "Crime pays but botany doesn't" is one such phrase. It sounds like a joke from a punk-rock lecture, yet it works as a prompt ingredient. It forces the AI out of its average-output rut.

In this article, I'll explain what vibe coding is, why "Crime Pays but Botany Doesn't" is a useful creative constraint, and how you can use it in your own micro-apps. No hype — just a practical trick you can test in the next ten minutes.

Vibe Coding in 2026: A Quick Refresher

The term "vibe coding" was coined by Andrej Karpathy, co-founder of OpenAI, in a February 2025 X post. He described it as "fully giving in to the vibes" — let the AI write the tedious parts while you focus on intent and iteration. As of 2026, this has become a standard workflow: developers use AI assistants like Cursor, GitHub Copilot, and Claude to generate functions, components, and entire micro-apps from natural-language descriptions. The concept is well documented on Wikipedia.

But there is a catch. These models are trained on a large slice of the internet. Their default output is the statistical average of every popular tutorial, blog post, and code sample. If you ask for a habit tracker, you get the blend of a million identical habit trackers. This is the "average output problem" — and it's the reason many vibe-coded projects look and feel the same.

Why "Crime Pays but Botany Doesn't" Breaks the Pattern

The phrase is the title of a long-running YouTube series hosted by botanist Tony Santorella — an eponymous channel that explores urban ecosystems with an anti-establishment twist. Syntax-wise, it's a paradox: crime is illegal and should not pay; botany is scientific and should be valued. Yet the phrase implies the opposite.

In a language model, this tension produces what prompt engineers call semantic distance. The model must pull from crime vocabulary (evidence, suspects, illegal activity) at the same time as botany vocabulary (plants, species, ecosystems). The clash prevents the model from collapsing into "nature app" clichés and forces it to generate less probable — and therefore more distinctive — tokens.

This isn't magic. It's a direct consequence of how transformer models assign probabilities over vocabularies. If you want unique output, you need a unique context. "Crime Pays but Botany Doesn't" is an excellent context because it is both a meme and a thesis: it gives the model a topic and a moral tension to resolve through design and copy.

Case Study: A 30-Minute Vibe-Coding Experiment

To test this principle, I ran an experiment while writing this article. The problem: I had to prototype a landing page for a fictional nonprofit that identifies edible wild plants in cities. The audience was teenagers and young adults — a demographic that typically ignores green nonprofits.

The solution was to apply the "Crime Pays" aesthetic. I opened Claude's web interface and entered the following prompt:

You are a botanist who used to work as a forensic evidence photographer. Build a single-page web app called Crime Pays but Botany Doesn't. It should have a suspect board for invasive plant species. Use case file numbers, most wanted sections, and an anonymous tip line. Dark theme with yellow crime-scene tape accents. Output clean HTML/CSS/JS.

The model generated an evidence-wall layout: a MOST WANTED grid of invasive plants, each with a profile photo, known aliases, and a list of charged offenses. One plant was "charged with strangling native asters." The dark theme and yellow accents gave it an immediately recognizable brand. The result was nothing like a typical plant-identification app.

The prototype wasn't production-ready. The text contrast was low, and the tip form had no backend. But within 30 minutes, I had a working demo with a strong voice. A second iteration fixed the contrast and added placeholder API calls. The workflow was: prompt, review, correct, repeat — a textbook vibe-coding session.

If you want to turn such a prototype into a full product with dynamic content, you'll probably need to call an LLM API from your backend. ASI Biont supports connecting to the Anthropic API via API — more at asibiont.com/courses.

How to Apply the Trick Yourself

You don't need a specific AI tool to use this pattern. The principle is to add a concrete and contradictory domain instead of an adjective. Here's a quick comparison:

The prompts in the third column produce far more specific interfaces because they give the model structural constraints, not just a mood.

When Not to Use This Trick

A provocative phrase is a tool, not a silver bullet. If you are building a standard internal CRUD app for a law firm, a "crime pays" theme is inappropriate. Use this technique when your product needs an opinionated brand — a personality that stands out. Do not use it as a substitute for technical quality: AI-generated code still requires review, especially for accessibility, security, and error handling.

Also, remember that the specific phrase is not universal. "Crime Pays but Botany Doesn't" carries cultural baggage. The mechanism is conflict between domains, not the words themselves. You can swap in "finance but entomology" or "engineering but mycology" — as long as the tension is real and meaningful to your audience.

The Takeaway

The phrase "Crime Pays but Botany Doesn't" started as a joke about career incentives. In the hands of a vibe coder, it becomes a prompt-engineering device that produces memorable software. Crime may pay, but botany — and with AI, even botany-based side projects — can win attention.

So next time the AI returns a generic result, feed it a paradox. You don't need a prompt-engineering degree; you just need one weird hook. The vibe will follow.