AI-Native
Development
at REX

A structured approach to making teams and codebases AI-native across the organization. From individual developer setup to team workflows to organizational scale.

Improvement Layers Dev Lifecycle Agent Skills AI Ecosystem Teaching Strategy

Improvement Layers

Three layers to improve. Start with the individual. Enable the codebase. Then the team ships differently. Each layer enables the next.

Layer 1 Individual Developer
"Help the person use AI effectively."
Environment
Terminal setup VS Code Claude Code
Prompting
Context engineering Communicating intent Prompt patterns
Skills
Installing skills Skills registry Custom skills
Tools
Azure models GitHub Copilot MCP servers Foundry
Layer 2 Project / Codebase
"Make the codebase ready for AI to work in."
AI Context
AGENTS.md CLAUDE.md Domain docs
Code Quality
Pre-commit hooks Linting Prettier
Testing
TDD AI guardrails Validation
Security
npm audits Dependency hygiene Vulnerability scanning
Layer 3 Team Workflow
"How does the team ship with AI in the loop?"
PR Workflow
AI-generated descriptions PR templates Conventional commits
Code Review
AI review guidelines Domain-aware review CI integration
Automation
CI + AI pipeline Guardrails Quality gates
Ways of Working
Dev lifecycle Ownership model Knowledge sharing

The Development Lifecycle

Nine steps from idea to production. Quality Assurance is the biggest phase because shipping broken code with AI is worse than shipping it without.

Prepare
1
Input
Gather context, requirements, domain knowledge
AI enables
Summarizes docs, gathers context from codebase and documentation
2
Clarity
Define scope, success criteria, acceptance tests
AI enables
Identifies edge cases, gaps in requirements, missing acceptance criteria
3
Plan
Architecture, approach, task breakdown
AI enables
Proposes architecture, breaks down tasks, identifies dependencies
Build
4
Execute
Build with AI, write code, iterate
AI enables
Writes code, suggests implementations, you guide direction and review
Quality Assurance
5
Test
Unit, integration, e2e tests pass
AI enables
Generates test cases, validates coverage, catches regressions
6
CI Pipeline
Lint, type check, security scan, build
AI enables
Automated checks run with AI-configured rules and quality gates
7
AI Review
Automated code review, quality validation
AI enables
Reviews for correctness, security vulnerabilities, and domain fit
8
Human Review
Team review, approval, knowledge transfer
AI enables
Summarizes changes, highlights risk areas, generates PR descriptions
Ship
9
Ship
Deploy, monitor, validate in production
AI enables
Monitors deployments, alerts on anomalies, validates production behavior

Agent Skills Roadmap

Skills organized by layer. Core skills should be adopted first. Extended skills add depth once the foundation is in place.

Layer 1
Individual
code-discipline Core
Behavioral guardrails for disciplined coding. Surface assumptions, write minimum code, make surgical changes, define success criteria.
research Core
Research-first workflow. Gathers verified, current information before answering or acting on any topic.
environment-setup Core
Interactive guide for AI dev environment: terminal, VS Code, Copilot, Claude Code, Foundry models.
prompt-patterns Core
Context engineering techniques. How to communicate intent, provide context, and get high-quality AI output.
debugger Extended
Systematic 6-step root cause analysis. Teaches structured debugging thinking with AI assistance.
decision-helper Extended
Decision frameworks: SWOT analysis, decision matrices, ICE scoring. Structured thinking for technical choices.
Layer 2
Project
setup-ai-context Core
Questionnaire-driven generation of AGENTS.md, CLAUDE.md, and domain knowledge files for any project.
ai-readiness Core
Audits a project for AI tooling coverage. Scores AI-friendliness (hooks, linting, tests, context files), provides fixes.
git-commit Core
Conventional commits from diff. Consistent format, scoped messages, standardized across all teams.
pr-description Core
Auto-generate PR title and body from branch diff. Follows team conventions and templates.
code-reviewer Core
Reviews code for security vulnerabilities, performance issues, and correctness. Catches what linting misses.
technical-writer Extended
Generates API docs, READMEs, and architecture documentation. Keeps project docs in sync with code.
doc-updater Extended
Keeps docs/ files current as codebase evolves. Detects drift between code and documentation.
security-best-practices Extended
Web security patterns. OWASP checks, dependency audits, vulnerability scanning, fix recommendations.
rex-domain-knowledge Extended
Domain skill for REX. What AI agents need to understand about us: glossary, projects, architecture, conventions.
Layer 3
Workflow
code-review-setup Core
Set up AI review guidelines, review skills, and domain-aware review for a project's review process.
dev-workflow-guide Core
Interactive guide through the Input-to-Ship lifecycle. Teaches Ways of Working with AI at each stage.
project-planner Extended
Work breakdown structures, dependencies, milestones, and estimation. Structured project planning with AI.
documentation-and-adrs Extended
Architecture Decision Records. Captures why decisions were made, not just what. Institutional knowledge.
self-improving-skills Extended
Meta-skill that reviews and improves existing SKILL.md files. Teams iterate on their own skills over time.
Creating Skills is Simple
Use the skill-creator, follow the AgentSkills 2 specification, and build with frontier models (Claude, GPT) for the highest quality output. The process is lightweight: discuss the need, define the scope, then generate.
1. Discuss and decide 2. Use skill-creator 3. Follow AgentSkills 2 spec 4. Build with frontier models 5. Test, iterate, share

The AI Ecosystem

Four layers of the AI stack. Your configuration shapes how interfaces use models that run on infrastructure. Know what each does and how they connect.

Developer You configure and use CONFIGURES USES Configuration Lives in your codebase AGENTS.md / CLAUDE.md Skills & MCP Servers Rules & Hooks Domain Knowledge Interfaces Tools on your machine Claude Code (CLI) GitHub Copilot (IDE) Web Chat (Claude.ai, ChatGPT) CLI Tools (Aider, OpenCode) SHAPES INSTRUCTS CALLS Infrastructure Azure Foundry Hosts and routes AI models OpenAI Models Claude Models GitHub Copilot Reasoning depth · Context window · Speed · Cost Access & Auth Controls and governance API Keys SSO Model Permissions Usage Tracking
Models
The intelligence layer
Choose by task
  • Claude (Anthropic)Deep reasoning, code generation, long context (200K). Best for complex tasks, planning, review.
  • GPT (OpenAI)General purpose, broad knowledge. Good for quick answers, content generation, translation.
  • Gemini (Google)Multimodal, massive context (1M+). Strong for document analysis, image understanding.
  • Model selection mattersContext window, reasoning depth, speed, and cost all factor into which model fits which task.
Interfaces
How developers interact
Tools at hand
  • Claude Code (CLI)Terminal-first agent. Reads your codebase, runs commands, writes code. Full project context.
  • GitHub Copilot (IDE)Inline suggestions and chat in VS Code. Autocomplete, explain, fix. Fastest feedback loop.
  • Web ChatClaude.ai, ChatGPT, Gemini. Quick questions, brainstorming, exploration before coding.
  • CLI ToolsOpenCode, Aider, and other terminal tools. Different strengths for different workflows.
Configuration
Shaping AI behavior
Per-project setup
  • AGENTS.md / CLAUDE.mdProject-level instructions. Code style, architecture decisions, domain context. AI reads these every session.
  • SkillsReusable, shareable AI workflows. Audits, setup wizards, review processes. Install and invoke.
  • Rules and HooksAutomated enforcement. Pre-commit linting, commit message validation, code formatting on save.
  • MCP ServersBridge AI to external services. Databases, APIs, CI/CD, documentation. Extend what AI can reach.
  • Domain KnowledgeTeaching AI your domain. Business logic, system architecture, team conventions, API contracts.
Infrastructure
What powers it
Platform layer
  • Azure OpenAIEnterprise GPT access with data residency, compliance, and rate limits you control.
  • Azure FoundryHosts and routes models (OpenAI, Claude, Copilot). Connects interfaces to approved models.
  • Access and AuthAPI keys, SSO, model permissions. Who can use what, and how usage is tracked.

Teaching Strategy

We go deep on core topics and cover specific concepts through lighter sessions. Hands-on delivery, shareable artifacts, learning by doing.

Deep-Dive Topics

  • 01 AI-Native Development: the full picture across all three layers
  • 02 Context Engineering: communicating intent and providing context to AI
  • 03 The AI Development Lifecycle: from input to ship
  • 04 Making Codebases AI-Ready: AGENTS.md, hooks, testing, quality
  • 05 Building and Sharing Agent Skills across REX teams

Lightning Talks

  • 01 Setting up your AI development environment
  • 02 Understanding LLM models: what makes them different
  • 03 Writing AGENTS.md and CLAUDE.md that actually work
  • 04 Pre-commit hooks and code quality for AI
  • 05 AI code review: setup, results, and pitfalls
  • 06 Your first week with Claude Code and GitHub Copilot
  • 07 From legacy codebase to AI-ready
  • 08 How we migrated SLS to CDK using AI and skills (KP RMS team)
How We Deliver
Workshops
Hands-on building sessions. Set up environments, configure projects, create skills together in real time.
Show and Tell
Live demos of tools, workflows, and results. See what's working, share discoveries, learn from each other.
Tutorials
Step-by-step written guides and walkthroughs. Follow at your own pace, reference later.
Boilerplates
Starter templates and project scaffolding. Get a new project AI-ready from day one.
Scaffolding Tools
Generators that set up AI context, hooks, and configuration for existing projects.
Video Recordings
All talks and workshops recorded and shared. Catch up async, revisit concepts, onboard new team members.