Specialty · AI Risk · Saturday, May 23

The 'Vibe Slop' Crisis Is Coming — And Why The Infrastructure Trade Still Works

Two AI insiders who helped launch the agentic-AI craze are now warning their creations are producing bad code. Anthropic's own Claude Code is called 'one of the most broken pieces of software I've ever used' by a top AI insider. What it means for the AI book.

WSJ Saturday B2 · Reframed for the CW book

The Saturday paper's B2 lead is a striking piece on what the industry is starting to call the "vibe slop" crisis: AI-generated code that compiles, runs, but is silently broken. The two AI insiders quoted include one who has used Anthropic's Claude Code extensively. His verdict: "Claude Code is one of the most broken pieces of software I've ever used in my entire life." He cites flickering on-screen graphics, feature creep, and a prodigious appetite for memory.

The piece quotes Sundar Pichai's claim that 75% of all new code at Google is generated by AI (up from 50% last fall) and Mark Zuckerberg's prediction that AI will be writing and reviewing most of Meta's internal AI code by end of 2026. The vibe-slop argument is that these productivity claims hide the maintenance, security, and code-review costs that show up later.

Why The Capex Trade Still Works

Here's the crucial framing for our book: the vibe-slop crisis is not the same as an AI-capex collapse. The compute spend is real, persistent, and accelerating — even if the code quality is degrading. In fact, vibe-slop probably increases compute demand for two reasons:

Code review cycles get longer. Every vibe-coded change needs more compute to test, scan for security issues, and regression-test. That's more inference, more GPU hours, more Nvidia (NVDA) revenue.
Enterprise adoption requires guardrails. Anthropic, OpenAI, and Google are all building "AI checking AI" tools — Codex, Glasswing, Sentinel. That's a whole new vertical of compute spend that didn't exist a year ago.

The Saturday piece quotes Rohan Varma, who leads OpenAI's Codex team: "If you assume it will work out of the box, it probably won't." He adds that human engineers are ultimately responsible for any AI-generated code that goes into production. That's a tacit admission that the AI-coding productivity story has a ceiling, but it doesn't break the infrastructure thesis.

Three Names That Get A Tailwind From This

The cybersecurity sleeve in particular benefits from the vibe-slop wave:

CrowdStrike (CRWD) — at a 52-week high $659.55 on Friday. Every codebase that ships vibe-coded changes needs more endpoint protection.
Palo Alto Networks (PANW) — same playbook on the network side.
Fortinet (FTNT) — cheaper exposure to the same trend.

And on the picks-and-shovels side: Nvidia (NVDA), Advanced Micro Devices (AMD), Arm Holdings (ARM, +16% to a fresh 52-week high Thursday), Taiwan Semiconductor (TSM), GlobalFoundries (GFS), Broadcom (AVGO). The agentic AI capex cycle is broadening, not narrowing — even when the code being generated is questionable.

What Could Actually Break The Thesis

The trade has two real risks: (1) an enterprise-scale incident where a vibe-coded change causes a publicly visible failure (a hospital, a bank, a critical-infrastructure provider) that triggers regulatory pushback, or (2) Anthropic / OpenAI losing enterprise customers because the code quality story damages their reputations. Both are tail risks. We're watching them, but we're not trading them yet.

Book Impact · What It Means For The Portfolios

The vibe-slop crisis is bullish for our cybersecurity sleeve (CRWD, PANW, FTNT) and neutral-to-positive for the picks-and-shovels AI book (NVDA, AMD, ARM, TSM, GFS, AVGO). Code that's silently broken needs more inference compute to test, review, and remediate — that's more compute spend, not less. Watching for any enterprise-scale incident that could trigger regulatory pushback as the only real downside catalyst.

Book a 15-Minute Portfolio Review →