The AMD AI director has gone public with a complaint many Claude Code users have been voicing privately: the tool has gotten noticeably worse since a recent update, producing lazier output and weaker reasoning than before.
The criticism carries more weight than a frustrated developer's post. AMD runs serious AI engineering operations, and its AI leadership uses these tools in real workflows. When a director-level executive at a major chip company calls out a specific regression publicly, it's usually because the problem is consistent and reproducible, not a one-off bad session.
"Dumber and lazier" maps to two distinct failure modes. Dumber points to worse reasoning - the model making more mistakes, missing obvious steps, or producing less accurate code. Lazier is the more interesting accusation: it suggests Claude Code is intentionally doing less per prompt, writing stub functions instead of full implementations, stopping work earlier, or requiring more explicit instructions to complete tasks that previously ran on autopilot.
This pattern appeared with GPT-4 around late 2023, when users noticed similar degradation and speculated OpenAI was quietly adjusting the model to reduce inference costs (the compute required to generate each response). OpenAI never confirmed this. The same dynamic may be playing out at Anthropic now.
Anthropichasn't publicly acknowledged any behavioral changes to Claude Code. The company has been releasing updates to Claude Sonnet, the model that powers Claude Code, and any of those updates could alter how the model approaches coding tasks - sometimes helping one use case while hurting another.
For developers relying on Claude Code for actual development work - multi-step autonomous tasks, not just code snippets - behavioral regression is a serious problem. The pitch for an agentic coding tool is that it keeps going without constant supervision. If the model has been tuned to do less without being asked, that directly undermines the reason to use it over a simpler autocomplete tool.