We vibe code all the time at Hello Spark. It’s also a small part of what we do.
That might sound contradictory, so I want to explain what vibe coding is, why it’s useful, and why it isn’t the same as building software a business can depend on.
Where the term came from
Andrej Karpathy coined the term in February 2025. He described it as a way of coding where you “fully give in to the vibes, embrace exponentials, and forget that the code even exists” (Karpathy on X). IBM now describes it more broadly as prompting AI tools to generate code rather than writing it manually (IBM).
In the original sense, vibe coding means you aren’t really reading the code. You describe what you want, look at the result, and keep going.
Why it’s useful
That speed is valuable early in a project. Instead of spending weeks discussing a problem, we can build working software around a real process and put it in front of the people who do the work.
I want to be clear about the word “working.” This isn’t a mockup or a slide showing what AI could do. It runs on the client’s actual data and does the job. That’s how we learn what we misunderstood, which exceptions nobody mentioned, and what the team actually needs.
Why it isn’t production software
Getting software to work once is the easy part. Production software has to handle what happens when things go wrong.
A system a business depends on needs answers to questions like these.
- What happens when a document is missing a page or comes in a format nobody expected?
- Who is allowed to see which records?
- What happens if a step fails at 2 a.m.?
- How is data backed up and recovered?
- How will anyone know the system is still working correctly six months from now?
A demo doesn’t show any of that work, but it’s most of the job.
How we bridge the two
We use vibe coding to learn as much as we can, as fast as we can. Then we throw that first version away and start over.
The second build draws on what the first one taught us, plus the experience Michael and I bring from a combined 20 years of leading engineering teams and advising businesses on software (About Hello Spark). In this build we make deliberate decisions about architecture, permissions, error handling, security, testing, and monitoring (Custom Development).
We also keep AI to the parts of the system where its judgment helps. AI handles messy input. Exact work, such as totals, dates, and record matching, runs in plain code. Expensive or hard-to-undo actions wait for a person.
What this means if you’re hiring someone
A lot of people can show you a fast demo right now. Ask what happens after the demo. Ask who reviews the code, how failures are handled, and who owns the software. If the answers are vague, the software probably will be too. Our list of questions to ask before hiring an AI development agency goes further.