A study presented at the AIware 2026 conference analyzes current methods for configuring agentic AI coding tools, such as Claude Code, Cursor, and GitHub Copilot. Based on an analysis of 2,853 GitHub repositories, researchers found that agent management relies primarily on context files, with the AGENTS.md file potentially becoming a new industry standard.
What Happened
An analysis of over 2,800 repositories showed that the primary way to manage AI agents during the development process is through the use of context files. Meanwhile, complex mechanisms such as specialized skills and hierarchical subagents are used extremely rarely and are mostly limited to simple text instructions without the ability to execute programmatic code.
Context
There is a noticeable gap between the marketing promises of agentic AI and real-world usage practices. While the industry strives for automation, current workflows remain largely manual, relying on text prompts rather than deep programmatic integration of agent capabilities.
Why It Matters for the Industry
The study lays an empirical foundation for standardizing interaction between developers and AI agents. It points to the need to move from fragmented text instructions to unified configuration formats, which would simplify the orchestration of multiple agents over a single project and lower the barrier to entry for creating specialized tools.
Why It Matters for Users
Developers are encouraged to start structuring their repositories more consciously by implementing context files (e.g., AGENTS.md). This will increase the predictability of current AI assistants and better prepare workspaces for effective interaction with agentic tools.
What Is Not Yet Known / Limitations
There are differing views on the implications of such standardization: while developers see opportunities for optimization, security and intellectual property specialists point to the emergence of new risk vectors and complexities regarding access control.
Sources
Author
Look at AI, Editorial Staff