GitHub Analysis Shows Claude AI's Workflow Shift to Edit-First

# Claude AI Faces Criticism Over Changes in Engineering Task Performance A GitHub analysis by Stella Laurenzo, senior director of AI at AMD, states that changes to Claude AI have made it unusable for complex engineering tasks. The analysis, which gained widespread attention on social media, found that from late February into early March, Claude moved from a research-first approach of reading multiple files and gathering context before making changes to a more direct edit-first style.

com reported on the analysis and related statements.

The analysis found that the model reads less context before acting, makes more mistakes, and requires significantly more user intervention. It points to a rise in behaviors like stopping too early, avoiding responsibility, or asking unnecessary permission, linked to a reduction in thinking depth from late February into early March.

Stella Laurenzo wrote that Claude has regressed to the point that it cannot be trusted to perform complex engineering.

Anthropic Responds to Analysis Findings Anthropic’s Cherny commented that the analysis is likely misreading at least part of the data.

Cherny stated that the model’s reasoning hasn’t been reduced but that Anthropic had made a change so that the full reasoning trace of the model is no longer visible to the user. The response addresses the observed behaviors in the GitHub analysis.

Additional User Experiences with Claude Code Dimitris Papailiopoulos, a principal research manager at Microsoft, wrote on X that he had incredibly frustrating sessions with Claude Code in the past two weeks.

Papailiopoulos wrote on X that he set effort to max, yet Claude Code is extremely sloppy, ignores instructions, and repeats mistakes. These statements align with the timeline of changes noted in the analysis.

Background on Claude AI Adoption Claude AI has been widely adopted by individual developers and large enterprise engineering teams for complex, multi-step coding tasks since its debut.

The tool's use in such tasks provides context for the impact of the recent changes identified in the analysis. Fortune first convened The Smartest People We Know in 2001, bringing together CEOs and founders, builders and investors, thinkers and doers, though its direct relation to this event is not specified.