Need to migrate from one database to another without downtime? TL;DR: Dual writes can keep old and new databases aligned during a live migration, but partial failures are where the real complexity starts. bencane.com/posts/2026-0...

Glue Services: Part Two — Data Synchronization TL;DR: When old and new systems must run in parallel, synchronization glue services are often what make safe migrations possible. bencane.com/posts/2026-0...

When modernizing legacy systems, don’t be afraid to build glue services. TL;DR: Don’t let legacy integrations dictate your modern architecture—use glue services to isolate the old world from the new one. bencane.com/posts/2026-0...

Coding agents can’t see your architecture diagrams—fix that. TL;DR: If your architecture diagrams are images, agents can’t reason about them—make them code instead. bencane.com/posts/2026-0...

Building resilient systems isn't about preventing failures. It's about ensuring that when failures occur, the entire system isn't impacted. americanexpress.io/cell-based-a...

A few people have asked if I'm going to start a newsletter. After dismissing the idea for a while, I finally gave in. madflojo.substack.com It’s the same content I share here—distributed systems, architecture, leadership, and the occasional hot take—delivered to your inbox. Hit subscribe!

Most teams put low-level architecture in the wrong place—if they document it at all. TL;DR: Put low-level architecture in your codebase—because that's where both engineers and agents actually need it. bencane.com/posts/2026-0...

Your coding agent is missing one thing: architectural context. TL;DR: Give your coding agent access to your architecture decisions—without that context, it’s just guessing. bencane.com/posts/2026-0...

Who would have thought that the most important problem to solve for AI-assisted software engineering would be CI/CD pipelines? 🤔

One of the easiest ways to break a gRPC service in production is health-checking the wrong listener.
 TL;DR: If production traffic uses gRPC, your primary health checks should too—an HTTP listener being healthy doesn’t mean your gRPC service is. bencane.com/posts/2026-0...

Weighted load balancing has saved me more times than I can count. TL;DR: Weighted load balancing lets you shift traffic by percentage, making migrations, canaries, and production changes far safer than all-at-once cutovers. bencane.com/posts/2026-0...

Cut a new release of madflojo/tasks over the weekend. Mostly focused on maintainability improvements, cleanup, and hardening the codebase a bit more. Shout out to the friendly neighborhood LLMs for taking up most of the grunt work. github.com/madflojo/tas...

YOLO Is a Terrible Strategy for Validating Production Changes. TL;DR: Validate production changes in layers—canary, shadow traffic, synthetic traffic, and smoke tests—so customers don’t become your QA team. bencane.com/posts/2026-0...

Deterministic routing is one of the most effective ways for distributed systems to reduce consistency problems at scale. tl;dr: Route requests for the same data to the same place to scale systems while reducing stale reads and coordination overhead. bencane.com/posts/2026-0...

I was looking up a configuration for HA Proxy and came across an article that had all the answers I needed. But it seemed familiar somehow… Oh, wait, I wrote it 10 years ago. 😅

Running the same microservice in multiple platforms can sometimes be better than centralizing it. TL;DR: Reuse the codebase locally to reduce coupling, latency, and the impact of failures. bencane.com/posts/2026-0...

Recently, I published a few of my personal agent skills that I’ve been using for a while now. 📋 Review This Review a codebase, docs, or even blog content. Get a quality score and recommendations. github.com/madflojo/rev...

Every time a credit card is swiped, a correlation problem has to be solved in real time. In low-latency systems, you can’t wait for each request to finish. I wrote a deep dive comparing how payment systems and modern distributed systems handle this. 👉

If you’re migrating a system without shadow traffic, you’re taking on unnecessary risk. tl;dr: Traffic mirroring lets you send real production traffic to a new system without affecting users. It’s one of the safest ways to validate behavior before a cutover. 👉 bencane.com/posts/2026-0...

Agent Skills aren’t just for actions; they’re for institutional knowledge tl;dr: Most institutional knowledge lives in docs no one reads. Agent Skills let you capture that knowledge (frameworks, patterns, constraints) so agents can discover and apply it automatically. 🔗

Saved prompts are dead. Agent Skills are the next step. 
tl;dr: Instead of manually calling prompts, agents can now discover and execute skills based on intent. Skills bundle prompts, scripts, and context into reusable capabilities. 
👉

Generating code faster is only valuable if you can validate every change with confidence. AI tools are making it dramatically easier to produce code. But if you can generate changes faster than you can test them, you eventually create more software than you can safely operate.

I’ve been experimenting with capturing how I want my Go code to look in an agent skill: github.com/madflojo/go-... I decided over the weekend to break it out into its own project and share it with others. It’s very opinionated, but I’ve found it successful so far.

When you go to production with gRPC, make sure you’ve solved load distribution first. One of the biggest gotchas when teams roll out gRPC isn’t performance. It’s load balancing.

Migrating the American Express Payments Network Twice with Zero Downtime. Modernization isn’t real until the new system takes production traffic. Here’s how we approached migrating from our legacy payments network to our new cloud-ready platform. americanexpress.io/migrating-th...

You may be building for availability, but are you building for resiliency? Many teams design for availability. Far fewer design for resiliency.

If you're into hearing about complicated migrations of mission-critical payments platforms, you should check out my talk tomorrow at Monster Scale Summit. At AMEX, we didn't just migrate the Payments Network once; we did it twice, zero downtime. www.scylladb.com/monster-scal...

When your coding agent doesn’t understand your project, you’ll get junk.

You can have 100% Code Coverage and still have ticking time bombs in your code. 💣