A security engineer with more than a decade in software and security is pushing back against the growing use of "vibe coded" as an insult, saying his own AI-assisted development with Claude Code follows a disciplined, reviewed process.
Developer workflow · AI-assisted coding
A security engineer defends "vibe coding" with a three-gate process
He says every change to his Claude Code projects is planned, reviewed by a second AI agent, and tested by him before it ships. His account is a self-description, with no published audits or measured defect rates to check it against.
The workflow, in order
Each change passes through three checks before it is accepted.
Step 1
Plan
The change is planned before any code is written.
→
Step 2
AI review
A second AI agent reviews the written code.
→
Step 3
Human test
He tests the change himself before it is accepted.
The case for the process
Work is planned before it is written, not prompted in a loop.
A second agent checks every change for problems.
A security-minded engineer tests each change himself.
An outside developer is said to have reviewed and understood the work.
The open questions
The account is the engineer's own description of his workflow.
No independent audit or published code is included.
No measured results, such as defect rates, are given.
Critics say unreviewed AI code can add security flaws and upkeep costs.
The larger question
Can layered human and machine checks make AI coding agents dependable?
Layered checks
Plan, machine review and human testing together.
Speed gains
Faster output may carry risks that show up only in production.
What would settle it
More documented examples and results from real projects.
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article...