Slide 1

Slide 1 text

AWS USER GROUP JAKARTA Spec-Driven Development with Kiro From Vibe Coding to Production-Ready Software Petra • Tech Lead, SawitPRO December 2025

Slide 2

Slide 2 text

Today's Journey 1 The Problem with Vibe Coding Why rapid AI coding creates technical debt 2 Spec-Driven Development A structured approach to AI-assisted coding 3 Kiro's Three-Phase Workflow Requirements → Design → Implementation 4 Live Demo See spec-driven development in action SESSION INFO 45 Minutes 35 min content + 10 min Q&A Questions welcome anytime!

Slide 3

Slide 3 text

CONTEXT The Evolution of Software Development Traditional SDLC 1. Requirements ↓ 2. Design ↓ 3. Implementation ↓ 4. Testing ↓ 5. Deployment The Problems ✗ Slow and rigid process ✗ Docs become outdated fast ✗ Handoffs lose context ✗ Expensive to change late ✗ Devs skip steps under pressure ✗ Agile helped but fragmented docs AI-Enhanced SDLC ✓ AI generates specs from prompts ✓ Docs auto-sync with code ✓ Context preserved in specs ✓ Fast iteration, keep structure ✓ Process is the workflow ✓ Human steers, AI executes Key insight: AI doesn't replace SDLC—it makes the structured approach fast enough to actually use.

Slide 4

Slide 4 text

THE CHALLENGE What is Vibe Coding? DEFINITION Natural language prompts → AI generates code → Ship it fast Great for: •Quick prototypes •Proof of concepts •Learning and exploration But for production... Lost Context Like a 100-message Slack thread with no summary Unclear Decisions Why did we build it this way? Nobody knows. Technical Debt Code that works but nobody understands No Traceability Can't link code back to requirements Skips the traditional SDLC: requirements → design → implement → test

Slide 5

Slide 5 text

" Successful developers were already planning before prompting. They didn't just vibe—they thought first. — Insight from Kiro team's user research The developer steers, the AI authors

Slide 6

Slide 6 text

THE SOLUTION Spec-Driven Development Transform natural language requirements into structured specifications, then let AI implement with full context and traceability. VIBE CODING "Build me a todo app with authentication" → AI generates code immediately → Works, but why? How? What next? SPEC-DRIVEN "Build me a todo app with authentication" → AI generates requirements first → Then design, then trackable tasks 📋 Clear Requirements 🏗 Documented Design ✅ Trackable Tasks 🔗 Full Traceability

Slide 7

Slide 7 text

THE TOOL Meet Kiro WHAT IS IT? An agentic IDE built on Code OSS (VS Code base) with spec-driven development as a first-class feature. KEY POINTS •Not an AWS service—standalone IDE •Works with any tech stack •Free during public preview •Powered by Claude Sonnet FEATURES Specs Structured artifacts for requirements, design, and tasks Agent Hooks Event-driven automation triggered by file changes Steering Files Project context, coding standards, preferences MCP Integration Connect to docs, databases, APIs kiro.dev • Generally available December 2025

Slide 8

Slide 8 text

THE WORKFLOW Three Phases to Production 1 Requirements User stories + Acceptance criteria → 2 Design Architecture + Technical decisions → 3 Implementation Trackable tasks + Code generation You can iterate at any phase. Refine requirements, update designs, reorder tasks.

Slide 9

Slide 9 text

PHASE 1 Requirements Natural language → Structured user stories YOUR INPUT "I want to build a task management app where users can create, edit, and delete tasks with due dates" EARS NOTATION WHEN [trigger] THE System SHALL [response] KIRO GENERATES USER STORY As a user, I want to create new tasks so that I can track my work items. ACCEPTANCE CRITERIA WHEN the user clicks "Add Task" THEN the system SHALL display a task creation form WHEN the user submits a valid task THEN the system SHALL save it and display confirmation WHEN the task has a due date THEN the system SHALL validate date format EARS = Easy Approach to Requirements Syntax • Industry standard

Slide 10

Slide 10 text

PHASE 2 Design Requirements → Architecture and technical decisions KIRO ANALYZES Your existing codebase structure Technology stack in use Architectural patterns Dependencies and integrations OUTPUT INCLUDES •System architecture diagrams •Data schemas •API contracts •Technology recommendations EXAMPLE DESIGN OUTPUT ## Technical Design ### Architecture React frontend + Node.js API + PostgreSQL ### Data Model Task { id, title, description, due_date, status, created_at, user_id } ### API Endpoints

Slide 11

Slide 11 text

PHASE 3 Implementation Design → Trackable tasks with code generation GENERATED TASK LIST ✓ Set up project structure ✓ Create database schema → Implement task CRUD API Build React components Add validation and tests Execute one at a time — review, approve, checkpoint WHY THIS MATTERS Traceability Every line of code links back to its requirement Checkpoints Pause, review, and resume at any task Reordering Change task order based on priorities Property-Based Testing Auto-generates tests from your specs

Slide 12

Slide 12 text

BEYOND BASICS Power Features for Teams ⚡ Agent Hooks Event-driven automation triggered by file changes. Auto-format code, generate tests, update docs, scan for security issues. 🔌 MCP Integration Connect to external tools via Model Context Protocol. Access AWS docs, GitHub repos, databases, and custom APIs. 📋 Steering Files Configure project context, coding standards, and preferred workflows. Kiro remembers your preferences across sessions. 🧪 Property-Based Testing Specs become testable properties. Generate hundreds of test cases automatically to verify code matches requirements.

Slide 13

Slide 13 text

TEAM BENEFITS SDLC and Team Collaboration TRADITIONAL SDLC RESTORED Requirements → Design → Implement → Test Kiro automates this flow while keeping humans in control DOCUMENTATION THAT EVOLVES Specs stay in sync with code. No more outdated docs or Confluence pages that nobody reads. TEAM COLLABORATION Single Source of Truth Everyone works from the same specs—PMs, devs, QA Clear Boundaries Concurrent development without conflicts—each task is well-defined Easier Onboarding New team members understand why code exists, not just what it does Code Review Clarity Review against specs, not just "does it work?" Audit Trail Every decision documented—great for compliance and handoffs

Slide 14

Slide 14 text

LIVE DEMO Let's Build Something Watch the spec-driven workflow in action 1 Start with a prompt → 2 Review specs → 3 Execute tasks 💡 Follow along at kiro.dev

Slide 15

Slide 15 text

PRACTICAL GUIDANCE When to Use What? Vibe Coding Quick and exploratory BEST FOR •Quick prototypes •Proof of concepts •Learning new tech •Personal projects WATCH OUT FOR •Growing complexity •Team handoffs Spec-Driven Structured and maintainable BEST FOR •Production features •Team collaboration •Complex business logic •Compliance requirements THE TRADE-OFF •More upfront time, less rework

Slide 16

Slide 16 text

GET STARTED TODAY Try Kiro Now 1 Download from kiro.dev Available for macOS, Windows, Linux 2 Sign in Google, GitHub, Builder ID, or AWS SSO 3 Create your first Spec Click + under Specs panel, describe your feature PRICING Free tier: 50 interactions/month Pro: $19/mo (1,000) • Pro+: $39/mo (3,000) RESOURCES 📚 Documentation kiro.dev/docs 🏆 Hackathon $100,000 in prizes • kiro.devpost.com 💬 Community Discord, GitHub Discussions 🚀 Startup Offer 1 year Pro+ free for qualifying startups

Slide 17

Slide 17 text

Questions? Let's discuss! KEY TAKEAWAYS 📋 Plan first 🎯 Steer the AI 🔗 Stay traceable kiro.dev