Why not just rewrite the system?

Rewrites rarely land faster than 12–18 months and you carry the old system while building the new one. Debt compounds on both. We targeted deploy safety, observability, and infra cost—the three areas with immediate impact on MTTR, change failure rate, and margin—so the business could keep shipping.

Do we need Istio to do canaries and circuit breakers?

No, but a mesh like Istio makes it consistent. You can start with NGINX annotations for canaries and app-level circuit breakers (e.g., `resilience4j`). The key is automated rollout/rollback and health-based gating.

How do we avoid AI-generated ‘vibe code’ in the future?

Keep AI in the loop but enforce patterns: architecture decision records (ADRs), shared libraries, strong linters/formatters, and CI gates. Run periodic `AI code refactoring` sprints to consolidate helpers and remove dead code. Treat Copilot/ChatGPT as interns: review everything.

What’s the minimal tool stack to replicate this?

ArgoCD for GitOps, your feature flags of choice (LaunchDarkly/OpenFeature), Prometheus + Grafana for SLOs, OpenTelemetry for tracing, Terraform for infra guardrails. If you’re on a PaaS (Fly.io, Render), you can still apply the same principles with their equivalents.

Case-studies · Nov 17, 2025 · 9 minute read

The Rewrite We Didn’t Ship: 90 Days of Tech-Debt Paydown Dropped MTTR 90% and Cut Cloud Spend 24%

A high-growth startup was stuck in incident purgatory and AWS sticker shock. We killed three classes of debt, avoided a risky rewrite, and bought back 30% of engineering time in one quarter.

Back to all posts

The Rewrite We Didn’t Ship: 90 Days of Tech-Debt Paydown Dropped MTTR 90% and Cut Cloud Spend 24%

Key takeaways

Implementation checklist