The Primeagen has flipped on AI

Overview

The speaker argues that prominent developers are increasingly embracing “vibe coding” and AI-assisted coding, which often produces large amounts of “slop” code. They claim this shift is alarming because the core problems with AI haven’t improved.

Key Claims

Notable reversals toward AI coding

The speaker points out that respected figures who were previously skeptical are changing their stance, especially:

• DHH: described as becoming strongly pro–vibe coding
• ThePrimeagen: said to have started endorsing AI-assisted development across multiple recent videos/streams

“Vibe coding” normalizes untrustworthy code

The speaker’s central concern is that AI-generated agents and code can’t be trusted. They argue this leads to:

• Low-quality output
• Code that becomes increasingly hard to maintain over time

Long-term dependence on AI companies

The speaker claims that once AI-generated code makes projects too unruly to edit manually, developers will eventually be forced to keep relying on AI tooling—resulting in:

• Dependency on large AI providers
• Reduced independence, framed as “renting” coding ability from major corporations

Local models won’t solve the independence problem

They argue local models are unlikely to become a broadly viable alternative due to:

• Restrictions or “nerfing”
• Prohibitive hardware costs

So most people won’t realistically be able to run AI tooling offline or affordably.

Hallucinations are inherent and unsolvable

The speaker asserts that LLM hallucinations are fundamental to how the technology works, meaning they can’t be eliminated. As a result, the core risk remains.

Proposed Alternative Approach

The speaker suggests using LLMs sparingly—similar to a search engine—but only for information that can be verified. Otherwise, they prefer traditional, more reliable resources such as:

• Books
• Documentation
• Stack Overflow (offline)

Framing: Openness, Autonomy, and FOSS

The speaker connects the issue to the FOSS/Linux “independence” ethos, arguing that AI adoption undermines autonomy.

Additional Concern (Briefly Mentioned)

They also briefly note concerns about:

• Stolen code
• The broader “murkiness” of AI training and outputs

However, they state the main focus is on coding reliability and independence.

Presenters / Contributors Mentioned

• ThePrimeagen
• DHH
• Teeg (mentioned as collaborating/showcasing)
• Claude (referenced as part of vibe-coding workflow)
• Jonathan Blow
• Casey Muratori
• Zed (credited for the GPUI project mentioned)