Problem Brief — AI-First Engineering Team

The original brief that shaped this case study.

The Problem

Company: PhoenixDX — AI-first engineering hub, Ho Chi Minh City
Mission: Re-architect mission-critical enterprise systems using AI-augmented engineering

An enterprise platform serving ~40,000 global users needs to be modernized.
The platform covers: travel booking, event management, workforce allocation, digital payments, communications, and reporting.

Current state: Legacy .NET monolith — single database, single deployment, no events, no tests
Target state: .NET 8 microservices, event-driven, per-service databases, React 18 frontend

The Constraints

These are real constraints, not artificial difficulty. They force real trade-offs.

Constraint	Why it matters
5 engineers	Cannot grow the team — must do more with the same headcount
9 months	Hard deadline — scope must fit or be explicitly deferred
Zero downtime	40K users across timezones — no maintenance window exists
Payment frozen (Phase 1)	PCI compliance risk — Payment module cannot be touched initially

The Central Question

Can a 5-person team execute a 9-month platform modernization that would normally require 10+ engineers?

The hypothesis: yes — but only if AI is treated as engineering infrastructure, not a tool individuals use ad hoc.

This means:

Investing Month 1 in AI toolchain setup before writing any product code
Assigning AI multipliers to capacity planning (not guessing)
Building governance for what AI can and cannot own
Designing architecture to be AI-legible (ADRs, contracts, clear boundaries)

What Needed to Be Produced

Six formal outputs — the backbone of this case study:

Output	Core question answered
4.1 Target Architecture	What does the target system look like, and why?
4.2 Migration Strategy	How do we get there without breaking anything?
4.3 Failure Modeling	What can go wrong, and what's the recovery plan?
4.4 Trade-Off Log	What are we deliberately not doing, and why?
4.5 Assumptions	What must be true for this plan to hold?
4.6 AI Usage Declaration	What did AI produce, and what did humans validate?

See docs/03-deliverables/ for each output.

Target Tech Stack

Layer	Technology	Key reason
Backend	.NET 8 Microservices	Migration path from existing .NET monolith
Frontend	React 18 + Shared Design System	Incremental adoption, vast ecosystem
Gateway	YARP	.NET-native, Strangler Fig routing built-in
Messaging	Azure Service Bus	Managed, zero ops overhead
Database	Per-service Azure SQL	One technology, team already knows SQL Server
IaC	Bicep	Azure-native, no state file management
Containers	Azure Container Apps	Zero Kubernetes ops for 5 engineers
Observability	OpenTelemetry + Serilog	Vendor-neutral, .NET-native
AI Tooling	Cursor Pro + Claude Code + CodeRabbit	The 2× multiplier stack

Domain Modules (Legacy → Microservices)

LEGACY MONOLITH                    TARGET (9 months)
────────────────────               ─────────────────────────────
┌──────────────────┐               ✈  Travel Booking     (M3)
│  Travel Booking  │  ──extract──► 📅  Event Management  (M4)
│  Event Mgmt      │               👥  Workforce          (M6)
│  Workforce       │               📧  Communications     (M7)
│  Communications  │               📊  Reporting CQRS     (M7)
│  Reporting       │
│  ─────────────── │               💳  Payment — stays in monolith
│  Payment         │  ──ACL──────► (Anti-Corruption Layer bridge)
└──────────────────┘
       │
  Single SQL Server
  ~200+ tables
  Cross-module FKs

Assessment Brief