Code Is Not Cheap: How to Multiply Your AI’s Output With Software Fundamentals

2026-05-21 · Source: AI Advances - Medium · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Intermediate, quick

Summary

Andrej Karpathy's concept of "vibe coding," introduced in February 2025, evolved into "agentic engineering" a year later, reflecting a growing understanding that AI-generated code requires significant human oversight. Karpathy observed that AI models are "jagged," excelling at complex tasks but struggling with basic ones. This shift is supported by data, with GitClear's 2025 study revealing that AI-coauthored pull requests contain 1.7x more issues than human-only ones. Furthermore, the percentage of copy-pasted code lines rose from 8.3% in 2021 to 12.3% in 2024. Despite these quality concerns, AI now generates 41% of all code on GitHub, with 4.7 million paid Copilot subscribers, underscoring AI's pervasive role while emphasizing the enduring need for software engineering expertise.

Key takeaway

For AI Engineers and Software Development Managers integrating AI coding tools, recognize that "agentic engineering" requires rigorous human oversight. Despite AI generating 41% of code on GitHub, AI-coauthored pull requests have 1.7x more issues. You must prioritize robust code review processes and invest in your team's fundamental software engineering skills to ensure quality and mitigate risks associated with "jagged" AI outputs.

Key insights

AI-assisted coding, while prevalent, demands human engineering expertise to mitigate quality issues from "jagged" model outputs.

Principles

• AI models exhibit "jagged" performance.
• Human oversight is critical for AI-generated code.
• Software fundamentals remain essential.

In practice

• Review AI-generated code for "jagged" errors.
• Prioritize human-led code quality checks.
• Invest in core software engineering skills.

Topics

• AI-assisted Coding
• Agentic Engineering
• Code Quality
• Software Fundamentals
• GitHub Copilot
• Andrej Karpathy

Best for: CTO, VP of Engineering/Data, AI Architect, AI Engineer, Software Engineer, Director of AI/ML

Related on AIssential

• Vibe coding and agentic engineering are getting closer than I'd like — The increasing reliability of AI coding agents blurs the distinction between casual "vibe coding" and professional "agentic engineering."
• Writing Code with Github Copilot and ChatGPT Together — III — AI code agents excel at routine tasks but require human oversight for novel or complex problems.
• Applying an Agentic Coding Tool for Improving Published Algorithm Implementations — Agentic coding tools can significantly improve published algorithm implementations across diverse domains within a single workday.
• AI Coding Is Powerful. That’s Exactly Why Fundamentals Matter — AI coding tools make fundamental engineering judgment and discipline more critical, not less.
• Slow down to speed up: so much has changed in 6 months’ time — Rapid AI agent adoption boosts individual dev output but risks software quality and organizational integrity without human oversight.
• 4 years of observing and living through the AI era — Generative and agentic AI tools offer unprecedented automation and creation capabilities, shifting the focus to human validation and decision-making.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AI Advances - Medium.