- AMD AI head says Claude Code lost performance in February 2026 update
- Claude Code “cannot be trusted” on complex tasks after thousands of coding sessions
- Anthropic says the effort has been reduced to medium, but perhaps teams and companies could achieve a high level.
Stella Laurenzo, director of AMD AI, claimed that Claude Code has become less effective since around February 2026, arguing that it “cannot be trusted to perform complex engineering tasks.”
Laurenzo’s criticisms are not without merit, based on the company’s analysis of more than 6,800 coding sessions, nearly 235,000 tool calls, and nearly 18,000 blocks of reasoning.
“All senior engineers on my team reported similar experiences/anecdotes,” Laurenzo wrote, noting that stop hook violations (where Claude quit early, dodged responsibility, or requested unnecessary permissions) increased from zero in early March to around 10 per day thereafter.
Article continues below
Claude Code’s situation is getting worse, warns AMD manager
In a GitHub post, user stellaraccident (also known as Stellar Laurenzo) identified a strong correlation between the introduction of thoughtful writing (redact-thinking-2026-02-12) and a drop in performance on complex tasks. AMD chief says extended reasoning can be “boosting” for advanced engineering work.
Laurenzo also observed a shift from search-focused behavior to editing-focused behavior, leading to lower quality code, less adherence to conventions, and overall reduced reliability for long sessions.
Anthropic has already responded to the research with a multi-faceted explanation. Boris from Claude Code explained that the redact-thinking-2026-02-12 setting only hides the reasoning from the UI and does not actually reduce the reasoning.
The company also introduced adaptive thinking with Opus 4.6, with this model dynamically deciding thinking time to improve performance and efficiency.
“Some people want the model to think longer, even if it takes more time and tokens,” Boris added. “To further improve intelligence, set effort=high via `/effort` or in your settings.json.”
With medium effort, or effort=85, now the default for users, Anthropic promised to test higher effort for Teams and Enterprise users so they “benefit from deep thinking even if it comes at the cost of additional tokens and latency.”
“I appreciate the depth of thought and care that went into this,” Boris also noted, crediting AMD’s Laurenzo for the analysis.
Follow TechRadar on Google News And add us as your favorite source to get our news, reviews and expert opinions in your feeds. Make sure to click the Follow button!
And of course you can too follow TechRadar on TikTok for news, reviews, unboxings in video form and receive regular updates from us on WhatsApp Also.
