L1: IAG Booking (Iberia Pilot) Overview

Capstone Case Study - IAG Booking Rewrite (Iberia Pilot) Overview

In Modules 1 through 11, we explored the theoretical and mechanical application of Claude within the Agentic Software Development Life Cycle (SDLC). This capstone lesson synthesizes those concepts by analyzing a massive, real-world enterprise modernization effort: rewriting a legacy airline booking engine.

1. The Legacy Challenge (The Monolith)

Airline booking systems are notoriously complex, often relying on decades-old mainframes, monolithic C++/Java architectures, and rigid Global Distribution Systems (GDS) like Amadeus or Sabre.

The Iberia Pilot Problem Statement:

Technical Debt: The legacy booking engine used heavily coupled SOAP APIs and XML payloads, making mobile app integration painfully slow and error-prone.
Brittle Business Logic: Thousands of undocumented pricing rules, regional tax codes, baggage allowances, and loyalty tier constraints were hardcoded across hundreds of legacy files.
The Goal: Pilot an AI-driven rewrite of the core Passenger Name Record (PNR) creation and seat-selection flows. The target architecture was modern microservices (Node.js/TypeScript, GraphQL) and React-based frontends, executed without disrupting active revenue streams.

2. Architectural Strategy: The Hub-and-Spoke Deployment

Instead of handing Claude a massive prompt to "rewrite the booking system," the AI Architecture team deployed the strict Task Decomposition and Chaining models discussed in Module 3.

The Orchestration:

The Legacy Analysis Agent (Claude 4 Opus): Tasked purely with reading the legacy codebase and extracting the mathematical business rules into structured JSON. It did not write new code; it acted as the extraction engine.
The BRD Traceability Agent: Cross-referenced the extracted JSON rules against the modern Business Requirements Documents (BRD) to ensure no edge cases (e.g., infant ticketing rules or complex multi-city routing) were dropped (Module 8).
The Microservice Generators (Claude 3.5 Sonnet): Multiple parallel subagents tasked with scaffolding the new TypeScript microservices based on the strict CLAUDE.md engineering standards (Module 5).

3. Execution via the Message Batches API

Rewriting an enterprise booking engine requires processing thousands of files. Attempting this with synchronous API calls would have triggered rate limits and caused pipeline timeouts.

The Batching Solution (Module 9):

The architecture team utilized Anthropic's Message Batches API. They packaged thousands of legacy Java and C++ files into a single .jsonl payload. Over the course of 24 hours, Claude processed the batch asynchronously, outputting decoupled, modern TypeScript functions and their corresponding OpenAPI specifications. This bypassed real-time rate limits and reduced the token processing cost by 50%.

4. Continuous Verification and Shadow Testing

In airline ticketing, a calculation error of a few cents multiplied by millions of passengers results in catastrophic revenue loss. The "Tautological Testing" trap (Module 7) had to be avoided at all costs.

The Agentic Validation Loop:

Contract-First Testing: Claude generated strict unit tests based only on the legacy API contracts and the extracted BRD, not the newly generated implementation code.
Shadow Traffic Routing: Before the Iberia pilot went live, the architecture team used the new endpoints to process "shadow traffic"—copying live user requests from the old system and running them through the new AI-generated system in parallel.
The Coverage Feedback Loop: If the legacy system quoted €450 for a flight but the AI-generated service quoted €448, the APM telemetry triggered a Triage Agent (Module 10) to autonomously identify the missing tax variable, patch the new code, and re-run the pipeline.

5. UI Modernization and Design System Enforcement

A backend rewrite is useless if the frontend cannot consume it. Iberia's mobile and web booking interfaces required a pixel-perfect, highly responsive update.

Figma-to-Code Pipeline (Module 6):

The UX team created the new booking flow in Figma.
Using the programmatic JSON extraction method, the designs were fed to Claude alongside the enterprise CLAUDE.md Design System constraints.
Claude autonomously generated accessible, responsive React components, mathematically mapping the Figma spacing and hex codes to the established UI tokens.
A specialized UI Auditor Agent ensured WCAG compliance, mathematically guaranteeing keyboard navigability for complex components like the seat selection map.

Summary of Impact

The IAG (Iberia) Booking Rewrite pilot serves as the definitive proof of concept for Agentic SDLC automation. By architecting strict boundaries, enforcing requirement traceability, and orchestrating Claude 4 Opus and Claude 3.5 Sonnet in parallel, the enterprise reduced a multi-year legacy modernization roadmap down to a matter of months. They achieved mathematical parity with the legacy monolith while successfully transitioning to a scalable, modern cloud architecture.