Why ship behind a flag?
Four moments when a feature flag saves you a Saturday.
A feature flag is a switch in your code. The new feature ships off by default. You flip it on for a slice of users, watch what happens, then expand β or roll it back on the SDK's next check-in if something breaks. No re-deploy, no panic β see the callout below on how the check-in timing works.
Most teams build flags themselves, badly, three times before they admit they need a real one. Here are the four moments when the cost-of-not-having-a-flag shows up.
1. The risky release
You're shipping a checkout rewrite. It works on staging. It worked in code review. It will probably work in production. "Probably" is the word that costs you weekends. Ship it behind a flag, flip it on for 5%, watch your conversion rate for an hour. If it holds, expand. If it doesn't, you flip it back in the time it takes to refresh a dashboard.
2. The A/B test you keep meaning to run
Sheepit experiments are flags with two variants and a metric attached. Same code path, same SDK, same audit log. You don't need a second tool, a third dashboard, or a stats consultant β just pick a flag, pick a goal, split traffic, wait for significance.
3. The paid feature
Pro-tier features are the same switch, gated by a different rule. Free users get false. Pro users get true. Trial users get true for 14 days. Internal users get true forever. All of that is one flag and a handful of rules.
4. The beta group
You're showing the new dashboard to your three friendliest customers before everyone else sees it. Create a cohort ("emails ending in @acme.com"), attach the flag, done. When you're ready to expand, you change the rule β not the code.
Not ready yet? Create it as a draft
A draft flag isn't sent to any app: every SDK returns the default compiled into your code. Codegen still emits its key (marked as a draft), so you can write the code, set up its rules, prepare a rollout, and publish when you're ready β the rollout starts after you publish, so give its steps dates after your publish date. Publishing is one-way: once a flag is live, retire it by archiving it.