A/B Testing vs. Feature Flags: They're Not the Same Tool
Feature flags and A/B tests share infrastructure but answer different questions. A guide to telling them apart before your rollout data lies to you.
Read article →Export Your Feature Flags Before You Need To
Your flag configuration is production config that lives in exactly one place, on someone else's servers, with no backup. Here's a script that fixes that for LaunchDarkly in about four minutes, and what the export quietly leaves behind.
Read article →Your Feature Flag SDK Should Never Ask You to Restart Production
On July 10, LaunchDarkly's SDKs started 'giving up permanently' and the official remediation was to restart every affected application and Relay Proxy. That's not a recovery strategy, it's a design decision. Here's the design that makes restarts unnecessary, and how we built it.
Read article →Not Everything Is a Feature Flag
A field guide to telling feature flags, config values, and user settings apart, three things that look identical at the API boundary and behave nothing alike. Miscategorize them and you end up with 240 'flags', a dashboard nobody trusts, and a 3am page about a setting that should have been a database row.
Read article →Yes, You Could Vibe-Code Your Own Feature Flag Service. Here's the Pitch Anyway.
deathbyclawd.com isn't wrong, Claude really can one-shot a feature flag service in an afternoon. The problem isn't the first afternoon. It's the next two hundred. A friendly, no-insults case for paying us a small amount of money instead.
Read article →The Mobile Feature Flag Problem: When Your Rollout Plan Has to Survive an App Store
On the web, a flag flip reaches every user in seconds. On mobile, your rollout has to survive review queues, offline devices, persistent on-disk state, and the user who hasn't updated since 2024. A different problem, a different set of patterns.
Read article →Feature Flags for AI: Why Model Swaps, Prompt Changes, and Tool Rollouts Each Need a Different Pattern
Standard percentage rollouts don't translate cleanly to LLM features. Sticky bucketing, shadow evaluation, and capability gating are the three patterns AI engineers actually need, and why your latency, cost, and eval signals all behave differently.
Read article →Kill Switches: How Feature Flags Become Your Three-Second Incident Response
Progressive rollouts let you ship slowly. Kill switches let you stop instantly. A practical guide to the on-call engineer's most underrated tool, and how to design your code so it actually works at 3am.
Read article →The 6 Levels of Feature Flag Maturity (And Why Most Startups Get Stuck at Level 3)
From cowboy push-to-prod to a real flag system, a brutally honest map of how engineering teams actually evolve their release strategy, and how to know which level you should be at.
Read article →Why Feature Flags Are the Secret to Shipping Faster
Feature flags split deployment from release. Here is what that buys a team that ships daily, and where the practice starts to cost you.
Read article →