Dev & EngARTICLE

Writing code with AI is a path with no turning back, but it requires care to ensure reliability

Article · Dev & Eng
Reading
2 min
In short

Only 29% of developers trust AI-generated code. At Brid, the solution was to invert the logic: humans write and lead, while AI tests and audits every line.

In recent months, the debate about artificial intelligence and software development has focused on one question: "how many programmers will AI replace?". But there's another question, perhaps more important, that has started to gain ground: "does AI write trustworthy code?". Recent data released by Keyhole Software helps shed light on exactly this point.

In the United States, 92% of developers already use AI coding tools daily. So far, so good: adoption is real and irreversible. The problem comes next: only 29% trust the accuracy of the generated code. And the effects showed up fast: a 41% increase in bug rates after adopting these tools, and 91.5% of applications with AI-written code show traceable vulnerabilities.

These numbers don't surprise me. They simply confirm, on a global scale, what we discovered in practice, within our own engineering team: AI is an excellent reviewer and a terrible author. Precisely because it's a tool, AI cannot be the author of anything; rather, it can co-create under the permanent supervision of the human programmer, who performs the final validation.

AI as a quality reviewer

At Brid, a company based in Londrina, Brazil, specialized in agribusiness solutions, we experienced this lesson firsthand. In the new version of Agrofortis, our CRM platform focused on agribusiness, we made an engineering decision that made all the difference: using artificial intelligence not just to write code, but to test it. AI acts as a rigorous quality reviewer, capable of examining every line before the product reaches the user. The result is a safer, more stable application with a significantly reduced error rate.

The gain also shows up in timelines. Tasks that used to take months are now completed in one or two weeks. The agility is real, but it only becomes value when the order is right: humans create and lead, the machine audits. Reversing this logic, letting AI write and sending the result straight to the user, means trading speed for future damage. The bug that AI "wrote" today will be tomorrow's incident.

The developer remains responsible

That's why I disagree with the apocalyptic tone that dominates discussions about layoffs in the industry. AI didn't come to replace the developer; it came to expose the difference between those who master software engineering and those who merely type. The person responsible for the code remains the developer, with name, surname, and reputation at stake. Technology enhances this work; it doesn't eliminate it.

Translated from the Brazilian Portuguese original · Read the original

View profile →