Vibe coding: can you really build software just by describing it?
The honest answer — yes, within limits that matter a lot if you run a real business.
Yes — you can describe software in plain language and have AI write it, and it actually works. Tools built this way run real businesses today. The honest limit is that code you do not read and do not understand carries real risks the moment it handles something that matters: customer data, money, or anything you would be liable for if it broke.
Where the term comes from
Andrej Karpathy — a founding member of OpenAI and former Director of AI at Tesla — used the phrase "vibe coding" in early 2025 to describe a pattern he had noticed: writing software by feel, in natural language, accepting AI output without carefully reading it. He was not attacking it. He was naming something real.
The idea spread because it matched what many people were already doing. You describe what you want. The AI writes the code. You run it, see if it works, describe the next piece. You might not read a single line of the output.
What it genuinely unlocks
The barrier to building something small has dropped to near zero. That is a real change, and it is useful for a small business in Northwestern Ontario in specific ways.
- Prototyping an idea before paying a developer to build it properly — a working demo in hours, not weeks.
- Internal tools only your team uses: a simple inventory tracker, a form that emails you, a spreadsheet that calculates automatically.
- Automating a small repetitive task: pulling data from one place to another, formatting a report, sending a scheduled reminder.
- Testing whether a concept is worth building — spin up something rough, see if it solves the problem, then decide whether to invest.
None of these require you to understand the code deeply. They require you to understand your own problem clearly enough to describe it — a skill most owners already have.
Where it creates real problems
The risks are specific, not theoretical. AI-generated code can contain bugs it does not flag, handle data in ways you did not intend, or have security gaps a developer would catch on review. When it runs an internal spreadsheet, that is a minor inconvenience. When it handles a customer’s payment or stores their personal information, it is a liability.
The standing rule: do not ship code you do not understand — or that no one on your team can read — into any system that touches customers, handles money, or stores personal data. AI-generated code still needs a human who can read it to review it before production. This is not about distrusting AI; it is recognizing that the AI does not know what it does not know, and neither do you if you have not read what it wrote.
Maintainability is the other quiet problem. Code you did not write and do not understand is hard to fix when something breaks six months later — and something will break. For anything you plan to run long-term, you want someone who understands the codebase.
A practical frame for NWO businesses
Think of vibe coding as a drafting tool, not a finished product. Use it to move fast on low-stakes experiments, to scope what is possible before spending money, and to automate small things that do not need to be bulletproof.
When an experiment proves out and you want to build it properly — or when it needs to handle real customer data — bring in a developer to review, rewrite, or rebuild from your working prototype. The prototype is genuinely useful input; it is just not the production system. The tools most people use for this are Cursor (an AI-first code editor), Replit (browser-based, no install), and direct prompting through Claude or ChatGPT.
The honest version of the story: vibe coding lowers the cost of experimenting dramatically — it does not remove the need for careful engineering on systems that matter. Experiment freely. Ship carefully.
Andrej Karpathy introduced the term "vibe coding" in a post on X in early 2025. Karpathy is a founding member of OpenAI and former Director of AI at Tesla. No statistics are cited; the framing reflects the documented behaviour of AI code-generation tools.
Frequently asked
What is vibe coding?
Building software by describing what you want in plain language and letting an AI write the code — often without closely reading what it produces. The term was popularized by AI researcher Andrej Karpathy in early 2025.
Can a non-technical business owner use vibe coding?
Yes, for low-stakes internal tools, quick prototypes, and simple automations. You describe the goal, the AI produces working code, and you test it. The barrier to getting something running has dropped dramatically.
Where does vibe coding become dangerous?
Any system that touches customer data, payments, personal information, or anything you could be held liable for. Code you do not understand and cannot review is code you cannot safely trust at that level.
Do I need to know how to code to try this?
Not to get started — tools like Cursor, Replit, and Claude let you describe what you want and iterate in plain English. But knowing enough to read and sanity-check the output matters more as the stakes go up.
Get the weekly Signal
The AI, funding, government, and tech moves that matter for Northwestern Ontario — one email a week, source-linked, read by a human before it reaches you.
One email a week from Thunder Bay AI. Unsubscribe anytime.