39,535 Claude.md Files Reveal Agent Adoption Patterns

A comprehensive analysis of 39,535 Claude.md and AGENTS.md configuration files from GitHub repositories has revealed significant insights into how developers are building and deploying AI agents in production environments. The study, officially released on November 17, 2026, represents the largest empirical examination of agent configuration practices to date.
What the Data Shows
The analysis examined configuration files from public repositories across GitHub, focusing on the standardized formats developers use to define agent behaviors, capabilities, and constraints. Claude.md files typically contain system prompts and behavioral instructions, while AGENTS.md files define tool access, permissions, and integration patterns.
Key findings reveal that 73% of repositories using these configuration files include explicit safety guardrails, while 61% define custom tool integrations. The most common configuration pattern involves combining natural language instructions with structured JSON schemas for tool definitions, suggesting developers prefer hybrid approaches that balance flexibility with precision.
Common Configuration Patterns
The study identified several recurring patterns in how developers structure agent instructions:
- Domain-specific prompts tailored to industries like healthcare, finance, and customer service
- Multi-stage reasoning frameworks that break complex tasks into discrete steps
- Explicit error-handling protocols and fallback behaviors
- Integration patterns for connecting agents to databases, APIs, and enterprise systems
Repositories with mature agent implementations typically include version control for prompt templates, automated testing suites, and detailed documentation of expected behaviors. This suggests the developer community is treating agent configuration with the same rigor as traditional software engineering practices.
Release Date and Methodology
Officially released on November 17, 2026, the analysis scraped public GitHub repositories over a three-month period. The researcher employed pattern matching and natural language processing to categorize configuration approaches, identifying both standardized practices and novel experimental patterns emerging in the community.
The dataset excludes private repositories and enterprise deployments, meaning actual agent adoption may be significantly higher than the 39,535 files analyzed. Organizations using Claude in production often maintain proprietary configuration files in private repositories, suggesting this study captures primarily open-source and experimental implementations.
Adoption Trends and Developer Behavior
The growth curve of Claude.md and AGENTS.md files shows exponential adoption since early 2026, with repository creation rates doubling quarter over quarter. Geographic analysis reveals strong adoption in North America and Europe, with emerging growth in Asia-Pacific markets. The most active sectors include software development tools, customer support automation, and content generation platforms.
Notably, 42% of repositories using these configuration files also include integration code for multiple AI models, indicating developers are building model-agnostic agent frameworks. This polyglot approach suggests the community anticipates a multi-model future where agents dynamically route tasks to specialized models based on requirements.
What This Means
This analysis provides the first large-scale empirical evidence of how developers are translating AI agent concepts into production systems. The standardization around configuration file formats signals maturity in the agent development ecosystem, while the diversity of implementation patterns reveals ongoing experimentation. For teams building AI agents, these findings offer a data-driven baseline for evaluating their own configuration practices against community norms. The rapid adoption curve suggests agent-based architectures are transitioning from experimental prototypes to mainstream development patterns, with implications for how organizations will structure AI capabilities in enterprise applications throughout 2027 and beyond.
on Emergent today





