VIBEPASS: Can Vibe Coders Really Pass the Vibe Check?
Evaluates whether natural-language-to-code workflows produce software that meets real-world quality standards. Benchmarks vibe-coded applications against professional baselines.
Read paper →Resources
Research, articles, and tools for AI-assisted teams.
Curated from academic research, practitioner blogs, and community signals. Everything here links to real, published content.
Research Papers
Academic research on AI-assisted software development from ArXiv.
SWE-CI: Evaluating Agent Capabilities in Maintaining Codebases via Continuous Integration
Benchmarks how well coding agents maintain existing codebases through CI pipelines. Tests agents on real-world maintenance tasks rather than greenfield generation.
Read paper →Evaluating Agentic Optimization on Large Codebases
Measures how AI agents perform when optimizing code across large, real-world repositories. Directly relevant to the blast radius concerns in HyperAgility.
Read paper →CodeTaste: Can LLMs Generate Human-Level Code Refactorings?
Blind comparisons of LLM-generated refactorings against expert human samples. Asks whether "human-level" is the right benchmark when AI operates at different scale.
Read paper →Vibe Code Bench: Evaluating AI Models on End-to-End Web Application Development
Benchmark for evaluating how AI models handle complete web application development from natural language descriptions to working applications.
Read paper →Review Beats Planning: Dual-Model Interaction Patterns for Code Synthesis
Research showing that review-oriented interaction patterns outperform planning-oriented ones for code synthesis. Supports HyperAgility's emphasis on review processes.
Read paper →Asymmetric Goal Drift in Coding Agents Under Value Conflict
Studies how coding agents deviate from intended goals when facing conflicting constraints. Directly related to scope drift and hallucination in underspecified tasks.
Read paper →Building Effective AI Coding Agents for the Terminal
Scaffolding, harness design, context engineering, and lessons learned from building production AI coding agents. Practical architecture guidance.
Read paper →XAI for Coding Agent Failures
Transforms raw execution traces into actionable insights when AI coding agents fail. Explains why agents fail, not just that they failed.
Read paper →Lore: Repurposing Git Commit Messages as a Knowledge Protocol for AI Coding Agents
Uses git history as structured knowledge to improve AI coding agent context. Treats commit messages as a protocol for preserving architectural intent.
Read paper →Nonstandard Errors in AI Agents
Categorizes error patterns unique to AI agents that don't fit traditional debugging models. Useful for understanding why AI-generated code fails differently.
Read paper →Trust Over Fear: How Motivation Framing Affects AI Agent Debugging Depth
Examines how the framing of system prompts affects how deeply AI agents investigate bugs. Trust-based framing produces more thorough debugging than fear-based.
Read paper →Code Fingerprints: Disentangled Attribution of LLM-Generated Code
Methods for attributing which parts of a codebase were generated by AI vs. written by humans. Relevant to ownership and accountability tracking.
Read paper →CONCUR: Benchmarking LLMs for Concurrent Code Generation
Evaluates how well LLMs handle concurrency when generating code. Tests a category of bugs that AI-generated code is particularly prone to.
Read paper →Test-Driven AI Agent Definition (TDAD)
Compiles tool-using agents from behavioral specifications. A formalization of spec-driven approaches to AI agent development.
Read paper →Articles & Analysis
Practitioner writing on AI-assisted development from across the web.
Code Quality in the Age of Coding Agents
Michael Timbs on what happens to code quality when AI agents generate most of the codebase. Confirms the patterns HyperAgility was designed to address.
Read article →A "High Blast Radius": Amazon Probes Surge in Outages Linked to AI Coding Tools
How even Amazon's stringent CI/CD pipelines couldn't prevent AI-assisted coding from causing production outages. Process needs to evolve alongside tooling.
Read article →Programming with Coding Agents Is Not Human Programming with Better Autocomplete
Argues that AI-assisted development is a fundamentally different mode of work, not just faster typing. The mental model needs to shift entirely.
Read article →Beyond Vibe Coding: Using AI as an Engineering Collaborator
Moving past "vibe coding" toward structured collaboration with AI. Practical patterns for using AI as a thinking partner, not just a code generator.
Read article →Code Review Is Broken — Here's What We Can Do
A practitioner's take on why traditional code review processes are failing under the volume and velocity of AI-generated changes.
Read article →How Far Can You Push Vibe Coding and Still Have Control?
Practical exploration of scaling agentic development while maintaining control over the codebase. Where does the approach break down?
Read article →Open Source Projects Need to Adapt to the AI Coding Era
How AI-generated contributions are changing the dynamics of open source maintenance, review burden, and community norms.
Read article →How Coding Agents Work
Simon Willison's guide to agentic engineering patterns. A clear explanation of the architecture behind the tools changing how we write software.
Read guide →Community Discussions
Practitioner conversations on the problems HyperAgility addresses.
Ask HN: How Do You Manage PR Review Fatigue from Coding Agents?
Community discussion on the growing burden of reviewing AI-generated pull requests. Real experiences from teams dealing with review overload.
Read discussion →Are Developers Trusting AI-Generated Code Too Much?
Discussion on the confidence gap: developers shipping code they generated with AI but don't fully understand. The core problem HyperAgility tries to solve.
Read discussion →"When each engineer ships 10+ PRs a day, the entire traditional SDLC collapses"
Evan Boyle on how agentic coding breaks the assumptions behind traditional software delivery. This is the core tension HyperAgility addresses.
View thread →"Vibe coding is easy. Engineering is still hard."
Geoffrey Huntley on the gap between generating code and building reliable systems. Speed without process is just faster failure.
View thread →Tools & Projects
Open source tools addressing AI-assisted development challenges.
Vibecheck: Lint for AI-Generated Code Smells
Detects common quality issues in AI-generated JavaScript, TypeScript, and Python code. Catches patterns that traditional linters miss.
View on GitHub →AgentProbe: Real-Time Observability for Coding Agents
Monitoring and observability tooling for AI coding agents. See what your agents are doing, where they're failing, and why.
View on GitHub →Sugar: Cross-Project Memory for AI Coding Agents via MCP
Persistent memory system that gives AI coding agents context across projects through the Model Context Protocol.
View on GitHub →Twill: Score Your GitHub Repo for AI Coding Agents
Analyzes your repository and scores how well-structured it is for AI-assisted development. Identifies areas that will cause agents to struggle.
Try it →Foundations
The thinking that HyperAgility builds on.
Manifesto for Agile Software Development
The original Agile Manifesto. Many of its values still hold, but the constraints it was designed around have shifted. HyperAgility builds on what remains relevant.
agilemanifesto.org →The Twelve-Factor App
Principles for building software-as-a-service apps. The emphasis on explicit contracts, disposability, and dev/prod parity aligns with AI-assisted architecture.
12factor.net →HyperAgility on GitHub
The source for this guide, the manifesto board, and all community contributions. Open issues, submit PRs, or start a discussion.
View repository →Know a resource that belongs here?
This page is maintained in the open. Submit a link, paper, or tool by opening an issue or PR on GitHub.
Suggest a resource