Completing the Rewrite from Hell: Five Years of Technical Debt and How We Escaped
summary
A product called SDK bin launches October 2020 after a two‑person team builds a seven‑microservice system with Postgres, Pulumi on Azure Kubernetes Service, NuGet hosting and Stripe billing, and only three paying customers from 2018–2020. Early design causes cartesian explosion queries that burn $400/month DTUs and security/renewal bugs; tests are useless. From 2024–2025 the team applies a strangler fig refactor, moves Razor Pages to MVC, adopts CQRS, adds 17 ADRs, uses Playwright and .NET Aspire tests and AI assistance, and replaces billing with actor‑based sagas.
AI-written summary · about
chapters (22)
- 0:00SDK Cabin disaster intro
- 2:29Five phases of SDK Cabin
- 5:15SDK Cabin's flawed vision
- 8:26COVID and project derailment
- 10:37Generic repository anti-pattern
- 13:27Managerial failures
- 16:58Front-end Razor chaos
- 19:10Cartesian explosion costs
- 21:58Launch day failures
- 25:54Developer retention crisis
- 29:26Three years of SDK Cabin faults
- 31:13Zero test coverage issues
- 33:17Failed rewrite attempts
- 35:51Strangler fig refactoring
- 37:15Razor to MVC migration
- 39:02AI breaking anti-patterns
- 41:28Rebuilding authorization
- 43:34Billing system flaws
- 45:20Subscriptions as state machines
- 47:46Fixing test coverage with .NET Aspire
- 49:15SDK Cabin today and lessons
- 50:19Q&A
related talks
01h 01mThree Rewrites, One Startup - An Architect’s Journey
26mTackling tech debt
27mBig Rewrite Strikes Again
11mA Failed Project and the Technical Architecture That Saved It
37mClean Architecture: Stories of Pain and Joy From the Codebase
24mHow to prevent a rewrite
33mTaylor’s Guide to Big Rewrites
26mOne does not simply... rebuild a product