Coding with AI vs. without it: what the data says
A controlled experiment with 95 developers showed 55% more speed coding with AI. But AI without engineering judgment isn't enough. How we use it at Manivela.
There's a debate that no longer makes sense: whether or not to code with artificial intelligence. The data already answered it. What's still open, and what matters, is how you use it.
The hard number
In a controlled experiment, 95 developers had to build an HTTP server in JavaScript as fast as possible. Half used an AI assistant; the other half didn't.
The group that used AI finished 55% faster: 71 minutes versus 161. And not only that, they also completed the task at a higher success rate (78% versus 70%). The result is statistically significant, not a LinkedIn anecdote.
The fine print nobody tells you
This is where most people get it wrong. AI doesn't replace the engineer: it leverages them.
More recent studies show that, on large tasks over complex codebases, an inexperienced developer can even go slower with AI, because they accept code they don't understand, carry subtle bugs and end up debugging what the machine got wrong. Speed turns into technical debt.
The difference isn't the tool. It's who handles it.
AI accelerates writing code. Engineering judgment is what guarantees that code is correct, maintainable and holds up in production.
How we use it at Manivela
We're developers with years of craft in large-scale systems. We added AI as one more tool in the workshop, with clear rules:
- We pick the model for the case. There's no single best one for everything. We use Claude, ChatGPT, Grok, Llama or Gemini depending on the task: reasoning about architecture, generating code, processing data or analyzing documents.
- Nothing ships to production without human review. AI proposes, the engineer decides. Every line goes through review, tests and standards we don't compromise on.
- We use it to go faster, not to lower the bar. We deliver products sooner, with better architecture and better standards, because the time AI saves us on the repetitive we invest in what really matters: design, security and performance.
What that means for you: the same enterprise engineering rigor, but with the timelines and costs of a new era.
In short
Coding with AI is measurably and provably faster. But speed alone doesn't build software that still works two years from now. That comes from the team behind it.
If you have a custom software project, an integration or a SaaS in mind, let's talk. We'll tell you how we'd build it, with what stack and on what timeline.