Available Languages: English | Español | Deutsch | Français | 日本語 | Português

Active Development - Updated November 26, 2025

coding-with-ai.dev

Awesome AI Coding Techniques

Practical techniques for coding with AI - Community-driven and practitioner-tested

This resource organizes techniques for working with coding assistants by development stage (from requirements and planning through review and refactoring).

The techniques draw from practitioners including Simon Willison, Armin Ronacher, Indragie Karunaratne, Orta Therox, and the Anthropic team.

Community-maintained and living. Contributions welcome.

🚀 Live site: coding-with-ai.dev

📝 Contributing: See CONTRIBUTING.md to share your techniques and experiences

Table of Contents

• Requirements & Planning
• UI & Prototyping
• Coding
• Debugging
• Testing & QA
• Review & Refactoring
• Cross-Stage Techniques

Requirements & Planning

Set Up Memory Files

Create context files that persistently guide tools about your project's structure, standards, and preferences.

Community adoption: 81% essential • Vote on coding-with-ai.dev (n=85)

"Think of AGENTS.md as a 'README for agents': a dedicated, predictable place to provide the context and instructions to help AI coding agents work on your project." — agents.md community

Tool Implementations:

For new projects: Run /init in your project root to create a starter CLAUDE.md.

For existing codebases: Run /init and Claude will analyze your project structure, dependencies, and configuration files to automatically generate essential information for working effectively in your codebase.

Claude examines: • package.json - Build scripts, dependencies, project metadata • Configuration files - .eslintrc, vite.config.js, tsconfig.json • Project structure - Component patterns, folder organization • Documentation - README.md, existing rules files

The generated CLAUDE.md includes: • Essential commands - npm run dev, npm test, npm run build • Technology stack - Frameworks, libraries, tools identified • Architecture overview - Component patterns, state management, routing • Key conventions - Code style, file organization, testing approach • Common gotchas - Build issues, configuration quirks, workflow notes

Memory file hierarchy: • ~/.claude/CLAUDE.md - your personal coding preferences (global) • ./CLAUDE.md - project team standards (project-specific)

Quick editing: • /memory - full editor interface • # - quick shortcut to add notes

Pro tips: • Review and customize the generated content for your specific project needs • Add gotchas you discover: "Never edit files in /generated/", "Always restart after config changes" • Link to project docs: @docs/deployment.md, @architecture.md • Iteratively improve - when you find yourself repeating instructions to Claude, add them to CLAUDE.md • Share with your team by committing CLAUDE.md to version control

Create AGENTS.md at project root: Cursor reads this file (also supports legacy .cursorrules) for consistent project guidance.

The AGENTS.md file should include: • Essential commands - npm run dev, npm test, npm run build • Technology stack - Frameworks, libraries, tools in your project • Code style guidelines - Naming conventions, preferred patterns • Architecture overview - Component structure, API routes, file organization • Common gotchas - Build issues, workflow requirements, restrictions

Real-time context (@-mentions): • @codebase - pull in relevant files from your entire project automatically • @docs - reference your project documentation
• @git - understand what you've changed recently • @web - get the latest patterns and examples from the internet

Project rules hierarchy: • Global rules in .cursor/rules directory • Project-specific rules in AGENTS.md • Path-specific rules with gitignore-style matching

Pro tips: • Combine static rules (AGENTS.md) with dynamic context (@-mentions) • Use @codebase when you need Cursor to understand the full project context • Keep AGENTS.md focused on project-specific conventions and gotchas • Let @-mentions handle the heavy lifting for code understanding

Create AGENTS.md at project root: Codex automatically reads this file at the start of every session.

The AGENTS.md file should include: • Essential commands - npm run dev, npm test, npm run build • Technology stack - Frameworks, libraries, tools in your project • Code style guidelines - Naming conventions, preferred patterns • Architecture notes - Component structure, file organization • Security rules - Environment variables, input validation requirements • Project gotchas - Common mistakes, build quirks, workflow notes

Monorepo support: • Put AGENTS.md in each package directory • Codex uses the closest one to your working directory • Get package-specific guidance automatically

Visual context (Codex's unique strength): • Drag and drop screenshots directly into your chat • Include UI mockups and design files • Share architecture diagrams and flowcharts • Perfect for implementing designs or explaining complex systems

Pro tips: • Keep your AGENTS.md updated as your project evolves • Add common mistakes you want to avoid: "Never edit files in /generated/" • Include build dependencies and setup requirements • Document any special deployment or testing procedures

Ask to Plan First

Tell the assistant to outline steps, risks, and quick tests before touching code so you can review and adjust the approach.

"If you want to iterate on the plan, it helps to explicitly include instructions in the prompt to not proceed with implementation until the plan has been accepted by the user." — Indragie Karunaratne

Tool Implementations:

Hit Shift+Tab to drop into Plan Mode so it only reads and drafts. Use the shared planning prompt, iterate until it looks right, then exit Plan Mode when you green-light implementation.

Click the Plan toggle in Cursor so it stays read-only while you iterate. Have it list steps, impacted files, risks, and quick tests, then exit Plan Mode to open the diff once you green-light implementation.

Remind Codex to keep planning separate from implementation: list steps, risks, and quick tests, pause for your review, then let it implement and inspect the diff once approved.

Plan with High-Capacity Model

When gathering requirements or drafting specs, temporarily switch to a higher-capability model or extended reasoning mode so it can read, synthesize, and propose a plan before coding.

Tool Implementations:

Run /model and pick opus (or another higher tier) when scoping requirements so it can reason deeply, then use Plan Mode if you want it to stay read-only until you approve edits.

Run /model and pick gpt-5-codex high for spec work that benefits from Codex's coding bias, or choose gpt-5 high when you need broader reasoning. Step back to your usual tier once the plan is approved.

Spec-Driven Development: Iterate Until Working

Iterate on specifications in Markdown until the assistant generates working code - treating specs as the source of truth rather than writing code directly.

"The workflow involves iterating on specifications in Markdown files, asking AI to compile into code, running/testing the app, and updating the spec if something doesn't work as expected. Developers should treat specifications as living documents, constantly updating and refining them to guide AI code generation with increasing precision." — GitHub Engineering

Choose Boring, Stable Libraries

Deliberately pick well-established libraries with good stability that existed before model training cutoff dates for better LLM-assisted code generation.

Community adoption: 64% essential • Vote on coding-with-ai.dev (n=43)

"I gain enough value from LLMs that I now deliberately consider this when picking a library—I try to stick with libraries with good stability and that are popular enough that many examples of them will have made it into the training data. I like applying the principles of boring technology—innovate on your project's unique selling points, stick with tried and tested solutions for everything else." — [Simon Willison](https://simonwillison.net/2025/Mar/11/using-llms-for-code/#