{
  "id": 3340426,
  "title": "using Codex to build unity_access. Whats worked, what hasn't and tips for building your own accessible applications with AI.",
  "url": "https://urgent.news/2026/08/25/using-codex-to-build-unity-access-whats-worked-what-hasnt-and-tips",
  "topic": "ai",
  "section": "AI",
  "published": "2026-08-25T20:15:06.000Z",
  "source": {
    "name": "Dev.to",
    "slug": "dev-to",
    "url": "https://dev.to/retro-coder64/using-codex-to-build-unityaccess-whats-worked-what-hasnt-and-tips-for-building-your-own-2hkc"
  },
  "original_language": "en",
  "account": "Over the past week, author has been coding Unity Access, a plugin for the Unity Editor to make it more screen-reader compatible. Unity Access uses NVDA API and Unity Editor API to create a bridge between Unity and NVDA screen reader. Author used Codex for programming and wanted to write a post about it. Developing Unity Access taught author a lot about using Codex for accessible application development. One main lesson is that building accessible software with AI requires more than just telling the model to make it accessible. Accessibility needs to be specific and testable. This post will cover what worked, what didn't work, how to build accessible applications with AI, and tips.\n\nBefore diving into details, author wants to describe the tools and technology used. Codex with GPT-5.6 Sol was used for coding, providing a good balance between efficiency, cost, and quality. Markdown specification files were written using VS Code, which works well with NVDA screen reader. Testing was done using Unity with Universal Render Pipeline (URP) sample project. NVDA screen reader was used for reading displayed content. These tools allow independent development, running, and testing of code.\n\nHere are the things that worked well during Unity Access development:\n\n1. Setting clear boundaries in the root AGENTS.md file. This includes specifying allowed changes, APIs, files, change size, accessibility requirements, and completion criteria. Specific rules make the model more predictable, especially for accessibility-focused applications.\n2. Setting boundaries around the size of changes. Instructing AI to make small, direct changes prevents unnecessary file modifications and ensures changes are relevant.\n3. Standardizing functions for controls, such as buttons and text boxes, in a utilities folder. This prevents inconsistencies and allows shared improvements to benefit the entire plugin.\n4. Organizing code into related folders. This helps screen-reader users navigate the project and find relevant files.\n\nDeveloping Unity Access taught author valuable lessons about using Codex for accessible application development with AI.",
  "summary": "Over the past week, I have been vibe coding Unity Access , a plugin for the Unity Editor that makes the previously difficult-to-access editor more screen-reader compatible. Unity Access uses the NVDA API alongside the Unity Editor API to create a bridge between Unity and the NVDA screen reader. I have been using Codex for the programming, and it is for this reason that I wanted to write this…",
  "key_points": [
    "Set clear boundaries in AGENTS.md file for predictable AI behavior",
    "Standardize functions for controls to maintain consistency",
    "Organize code into related folders for easy navigation"
  ],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}