The Need to Know: AI Vibe Coding

The Need to Know: AI Vibe Coding

Vibe coding is building software by describing what you want in plain language and letting an AI coding agent write, edit, debug, and extend the code. The human moves from manually writing every line to directing the system: defining the goal, reviewing the output, testing the result, and deciding what happens next.

The debate has split into two camps:

• The optimists see a software revolution. AI coding agents are making it possible for more people to build websites, applications, dashboards, automations, and internal tools without a traditional development team.
• The skeptics see a risk problem. People are shipping software they do not fully understand, trusting code they cannot properly review, and treating working demos like production systems.

Both sides have a good point.

The optimists are right that AI coding agents are no longer toys. As of 2026, tools like Claude Code, Cursor, Codex, and Copilot can produce real applications, connect databases, generate APIs, debug issues, write tests, and work across entire codebases. This is not just hobbyists building landing pages. Boris Cherny, creator of Claude Code, has said that 100% of his recent contributions to Claude Code were written by Claude Code. Spotify has also discussed its strongest developers moving away from manually writing code and using AI to drive development work.

Software creation is getting cheaper, faster, and more accessible.

That is the opportunity. But experienced developers are also right. Software engineering has never been only about generating code. A coding agent can build a login screen, but that does not mean the person directing it understands authentication. It can create a database, but that does not mean they understand data modeling, permissions, backups, recovery, or security. The system may work on the surface while becoming a codebase no human fully understands underneath.

That is the risk. The warning signs are already here: AI agents have deleted production databases, ignored instructions, and caused real operational failures. Even when nothing catastrophic happens, the habits can become dangerous. QA becomes “it seems to work.” Testing gets skipped. Security is assumed. The same tool that generated the code gets asked to verify that the code is safe. That is not quality control. That is misplaced trust.

The business takeaway is simple:

AI is reducing the cost of building software.
It is not reducing the cost of understanding the business.

Most business software problems are not just coding problems. They are workflow problems, process ownership problems, data visibility problems, and operating model problems.

The companies that win with AI coding agents will not be the ones that build the most software. They will be the ones that understand their operations well enough to build the right software.