![AI software development]()

AI software development: It works, but it's finicky

AI will write code, but prepare to babysit it - and be sure you speak its language

The promise and reality of AI coding

This week on The Kettle, we predict that AI software development won't make you want to fire your devs anytime soon. Brandon Vigliarolo hosts Systems Editor Tobias Mann and Senior Reporter Tom Claburn to discuss the current state of AI software development, a.k.a. "vibe coding."

Serving as the core of the discussion is Tom's story from earlier this week on research that found telling an AI it's an expert software developer actually makes it turn out worse code and what that means for the use of AI as a software development tool.

Our take? Sure, AI can write code - even sophisticated code - but you still need expert developers around to fix its ever-present errors and failures. In other words, companies that try to reduce the size of their dev teams on an AI bet might be making a mistake.

The "vibe coding" phenomenon

Tell an AI to write you a poem and it'll do it, just in a way that requires a human touch to perfect; the same goes for writing code. The hosts explore how AI coding tools have evolved from simple code completion to generating entire applications, but with a critical caveat: the output often needs significant refinement.

Research reveals AI's coding limitations

The discussion centers on recent findings that challenge the assumption that AI performs better when prompted as an expert. Counterintuitively, research suggests that when AI systems are told they're expert developers, they may produce more complex but less reliable code compared to when they're given simpler, more direct instructions.

This finding has significant implications for how organizations should approach AI-assisted development. Rather than treating AI as a replacement for human developers, the evidence points toward AI being most effective as a tool that augments human expertise rather than substitutes for it.

The human element remains essential

The hosts emphasize that while AI can generate functional code, it still struggles with edge cases, error handling, and architectural decisions that require deep understanding of the problem domain. Expert developers remain crucial for:

• Debugging AI-generated code
• Ensuring code quality and maintainability
• Making architectural decisions
• Handling complex integrations
• Understanding business requirements

Practical implications for development teams

For organizations considering AI adoption in their development workflows, the conversation suggests a measured approach. Rather than downsizing development teams in anticipation of AI taking over, companies might be better served by:

• Using AI for prototyping and initial implementations
• Having developers review and refine AI-generated code
• Focusing human expertise on complex problem-solving
• Using AI as a productivity tool rather than a replacement

Listen to the full discussion

You can listen to The Kettle here, as well as on Spotify and Apple Music. The hosts dive deeper into specific examples of AI coding successes and failures, discuss the evolving relationship between developers and AI tools, and explore what the future might hold for software development in an AI-augmented world.

[The Register]: A leading technology news publication covering IT and enterprise computing topics