LaunchDarkly

What Is LaunchDarkly?

LaunchDarkly is the category-defining feature flag platform. Engineering teams use it to decouple deploy from release — ship code dark, flip it on for internal users, then for beta cohorts, then for everyone. It has since expanded into experimentation, using the same flag infrastructure to run A/B tests and measure lift.

Also Known As

• LD (engineering shorthand)
• Feature flag platform / feature management platform
• Release management tool
• "The flag system" in many engineering orgs

How It Works

An engineer wraps a new payment flow behind a LaunchDarkly flag: "new-checkout-v2". They deploy to production with the flag off. In the LaunchDarkly UI, a PM enables the flag for internal emails, then for 1 percent of traffic, then 10, then 50, then 100 — with kill-switch rollback available at every step. For experimentation, they attach a metric (completed checkouts) and let LaunchDarkly's stats engine compare the two cohorts.

Best Practices

• Use short-lived flags for rollouts and delete them within 30 days of 100 percent release — stale flags accumulate and become a maintenance burden.
• Separate "release flags" (temporary) from "permission flags" (long-lived entitlements). Mixing them leads to flag sprawl.
• Require a kill-switch plan for every flag that touches payments, auth, or external APIs.
• Keep flag targeting rules in code review, not just in the UI — flag config is production config.

Common Mistakes

• Letting flags live forever. "We'll clean it up later" becomes 400 stale flags in your codebase.
• Using LaunchDarkly purely for experimentation without leveraging release management — that's over-paying for what a dedicated experimentation tool does better.
• Flipping flags in production without a revert plan.

Industry Context

LaunchDarkly is dominant in SaaS/B2B, especially at series-B-and-up companies where engineering velocity and safe releases matter more than marketing-led CRO. Ecommerce/DTC uses it less — they lean toward VWO or Optimizely Web. Lead gen rarely needs it. The common pattern at mature orgs: LaunchDarkly for engineering-led feature rollouts, plus a separate experimentation-first tool for growth tests.

The Behavioral Science Connection

Feature flags are, at their core, a reversibility mechanism. Kahneman and others have shown that humans make better decisions under reversibility — we're willing to try more things when we know we can undo them. LaunchDarkly operationalizes that insight at the engineering layer, which is why it tends to increase deployment frequency in teams that adopt it.

Key Takeaway

LaunchDarkly is the enterprise default for feature flagging and progressive delivery, with experimentation as a complementary layer — pick it when safe release velocity is the top priority.