Ask HN: Do you have any evidence that agentic coding works?

Article summary

The author has been trying to use agentic coding, but has not seen net-positive results, citing issues with code quality and technical debt. They question whether agentic coding can produce high-quality code without human review. The author shares their experience with using Codex to build an iOS app, which resulted in a significant amount of time spent fixing bugs and introducing guardrails. The author is skeptical about the long-term viability of agentic coding without human oversight.

Main themes

• Agentic coding effectiveness
• Code quality and technical debt
• Human review and oversight
• AI-assisted development
• Software development tradeoffs

What commenters say

• Agentic coding can be net-positive, but it requires human review and oversight to ensure code quality.
• The benefits of agentic coding, such as increased speed, may outweigh the costs of technical debt and lower code quality.
• Agentic coding is not a replacement for human developers, but rather a tool that can assist with certain tasks, such as boilerplate code.
• The limitations of agentic coding, including its inability to plan ahead or design complex systems, make it unsuitable for large or complex projects.
• The use of agentic coding requires a careful balance between the benefits of increased speed and the risks of technical debt and lower code quality.
• Human developers need to understand the underlying system and theory of mind to effectively use agentic coding tools.
• Agentic coding can be useful for certain tasks, such as testing, but it is not a panacea for software development and requires careful evaluation and review.
• The effectiveness of agentic coding depends on the specific use case and the skill level of the human developer using the tool.