Anthropic Economic Index: AI's impact on software development

Worauf sich diese Seite beruft, wörtlich, abgerufen am 12.10.2026:

• All conversations used in our analysis across Claude.ai and Claude Code were from April 6-13, 2025.bestätigt 10.11.2026
• 79% of conversations involved some form of automation, compared to 49% on Claude.ai.bestätigt 10.11.2026
• the primarily web-focused development languages of JavaScript and TypeScript together accounted for 31% of all queriesbestätigt 10.11.2026
• where developers of varying levels of experience describe their desired outcomes in natural language and let AI take the wheel on implementation details.bestätigt 10.11.2026
• Claude.ai conversations were specifically those from Claude.ai Free and Pro.bestätigt 10.11.2026
• This sample only includes Claude Code sessions powered by the first-party APIbestätigt 10.11.2026
• We excluded Team, Enterprise, and API usage that might show different patterns, particularly in professional settingsbestätigt 10.11.2026
• which distills user conversations into higher-level, anonymized insights.bestätigt 10.11.2026
• which Claude opted for in 5% of Claude.ai conversations and 2% of Claude Code conversations.bestätigt 10.11.2026
• We excluded this category from analysis and renormalized the resultsbestätigt 10.11.2026
• were also higher on Claude Code (43.8%, versus 27.5% on Claude.ai).bestätigt 10.11.2026
• were nearly twice as common on Claude Code (35.8% of interactions) as Claude.ai (21.3%).bestätigt 10.11.2026
• AI-assisted coding currently requires a lot of human reviewing and iteration relative to non-coding tasks, even when Claude does the bulk of the work.bestätigt 10.11.2026
• The HTML numbers for Claude.ai are likely inflated slightly because Artifacts leverage HTML.bestätigt 10.11.2026
• In a preliminary analysis, we estimated that 33% of conversations on Claude Code served startup-related work, compared to only 13% identified as enterprise-relevant applications.bestätigt 10.11.2026
• Startup work accounted for 32.9% of Claude Code conversationsbestätigt 10.11.2026
• whereas enterprise work represented only 23.8% of Claude Code conversationsbestätigt 10.11.2026
• We therefore treat these findings as more preliminary than the ones described above.bestätigt 10.11.2026
• our analysis tool made educated guesses based on incomplete information.bestätigt 10.11.2026
• Claude.ai usage does not include Claude For Work (Team and Enterprise plans) usage, which implies that enterprise numbers for Claude.ai specifically are likely undercountedbestätigt 10.11.2026
• We only studied what developers delegate to AIbestätigt 10.11.2026
• Our dataset likely captures early adopters.bestätigt 10.11.2026
• Two of the top five tasks were focused on user-facing app developmentbestätigt 10.11.2026
• each accounted for 12% and 8% of conversations, respectivelybestätigt 10.11.2026
• Such tasks increasingly lend themselves to a phenomenon known asbestätigt 10.11.2026
• to categorize a conversation as focusing onbestätigt 10.11.2026
• our results do show that even within automation, humans are still very often involvedbestätigt 10.11.2026
• The boundary between automation and augmentation becomes increasingly blurred with agentic tools like Claude Code.bestätigt 10.11.2026
• slightly below their 25.9% share on Claude.aibestätigt 10.11.2026
• Some classifications might therefore be incorrect.bestätigt 10.11.2026
• the quality of the resulting code, or whether these interactions effectively improved productivity or code qualitybestätigt 10.11.2026
• this self-selection could skew usage patterns towards more experienced or technically adventurous usersbestätigt 10.11.2026