AMD AI Director Publicly Criticizes Anthropic's Claude Code for Performance Regression, Calling It 'Unusable' for Complex Tasks

Apr 6, 2026

Stella Laurenzo, AMD's AI director, has publicly criticized Anthropic's Claude Code as "unusable" for complex engineering. An analysis of over 6,800 sessions revealed a 73% drop in thinking depth and a 70% decline in code reading since a March 2026 update. Anthropic attributed the changes to a new "adaptive thinking" feature. The incident highlights growing concerns over silent AI model updates that degrade performance and erode developer trust.

Claude chatbot capabilities

▪The analysis also found that API costs for the AMD team's usage increased 122-fold, from $12 per day to $1,504.
▪Anthropic advised users they can request longer thinking time by setting "effort=high" in their settings.
▪A separate bug since March 23, 2026, has been draining user usage quotas 10-20 times faster than normal across all paid tiers.
▪On March 31, 2026, Anthropic accidentally exposed the full TypeScript source code for its Claude Code CLI tool via a source map file on the npm registry.
▪Stella Laurenzo, AMD's director of AI, publicly stated that Anthropic's Claude Code is "unusable for complex engineering tasks."
▪Anthropic responded that the redaction feature only hides reasoning from the UI and does not reduce it, and that a new "adaptive thinking" feature was introduced.
▪An analysis by Laurenzo's team of 6,852 sessions showed Claude Code's median thinking depth dropped 73% and its code reading before editing fell 70% since February 2026.
▪Laurenzo attributed the performance drop to the early March 2026 deployment of version 2.1.69, which introduced "thinking content redaction."

Anthropic founding mission

▪Anthropic trains its Claude models on a "constitution" inspired by documents like the UN's Universal Declaration of Human Rights to be "helpful, honest, and harmless."
▪Anthropic was founded in 2021 by seven former OpenAI employees with a mission to create an AI startup focused on safety.

Claude versus ChatGPT competition

▪Six months prior to the criticism, Stella Laurenzo stated that Claude "stood alone in terms of reasoning quality and execution" but that competitors are catching up.
▪Anthropic is a major rival to OpenAI, though it initially focused on enterprise applications while OpenAI targeted the consumer market with ChatGPT.

Claude Code enterprise adoption

▪Anthropic has more than 1,000 enterprise customers who each pay over $1 million annually.
▪In the same week as the criticism, Anthropic announced a $21 billion TPU deal with Google and Broadcom.
▪Following the performance issues, Stella Laurenzo's team at AMD switched to a different provider for AI coding assistance.