‘AI Code Is Insane Trash’ - David Gerard (Video)

“I think we’re getting there… AI doesn’t work very well and it’s sort of selling itself on the future promises.”

In this 15 July 2026 published video, host Isaac Pound talks with technology journalist and host of Pivot to AI, David Gerard joins The Tech Report to talk about the fundamental problem with the AI development business model, selling it at a price people can and will purchase.

The phrase “AI code is insane trash” stems directly from a recent viral critique by tech commentator Gerard. It captures a growing wave of frustration among software engineers over the real-world limitations of generative AI. 

Gerard argues that AI-generated code is often “insane trash” due to fundamental structural flaws, diminishing LLM performance, and the creation of “slop” that requires excessive debugging. He contends that the promised 10x productivity gains from AI are a mirage, citing flat industry-wide output metrics and a growing backlash against AI-driven development. While AI code generation tools are massive productivity boosters for basic tasks, relying on them blindly introduces heavy technical debt. 

Large language models are hitting diminishing returns. Getting slight accuracy improvements now requires an exponential increase in computing resources. AI agents excel at isolated snippets but struggle with massive, existing corporate codebases. Without a unified understanding of the project, they duplicate logic and hallucinate non-functional elements. Software leaders note that up to 80 percent of unvetted, raw AI code is effectively garbage. Studies show roughly 45 percent of AI-generated code introduces predictable security flaws or logic bugs. Senior engineers increasingly report that reviewing and debugging poorly written AI code is far more exhausting than simply writing it themselves from scratch.

Many working developers argue that the ‘trash’ sentiment is a skill issue rather than an inherent tool failure. Senior engineers view AI as a junior assistant. If a developer lacks strong architectural and logical skills, they will use AI to generate messy, broken code faster. If a skilled developer provides rigid guardrails and treats AI as an automation layer for repetitive tasks, the output remains high-quality.

The industry is realizing that AI shifts the engineering workload. Typing syntax is automated, making human system design, security auditing, and critical thinking more vital than ever.

https://youtu.be/EwLW11Ucnps