New: outcomized services. Pay per result, with a monthly cap you set. See pricing →
Engineering

What is Feature Flags?

Feature Flags: Switches in code that turn features on or off for specific users or accounts without a new deploy, used for safe rollouts, tests, and quick rollback.

A feature flag wraps new code in a condition that can be changed at runtime. The feature ships to production turned off, then is enabled for internal users, a few beta customers, a percentage of traffic, or everyone. If something goes wrong, you switch it off instead of rolling back a deploy. Tools like LaunchDarkly, Statsig, PostHog, and Unleash manage flags at scale, and simple setups can use a config table.

Flags are especially useful for AI features, where behavior is harder to predict: you can route a small share of traffic to a new model or prompt, compare evals and user metrics, and expand gradually. The cost is complexity. Old flags that are never removed pile up into a different kind of technical debt.

Why it matters when you build or buy AI

Feature flags let you ship often without betting the whole product on each release. For an MVP, a handful of simple flags is plenty; you do not need a paid platform on day one. Make it a rule that each flag has an owner and a removal date, so your code does not fill up with dead branches.

Example

A new AI reply-drafting model is enabled for 10 percent of support conversations behind a flag. When its CSAT matches the old model over two weeks, the flag goes to 100 percent and the old path is deleted.

Need senior engineering leadership?Architecture, review, and hands-on building by the month.

Related terms

More in Engineering

Next step

Book 20 minutes, or send a brief

A former CTO replies. You leave with a written plan and a fixed estimate.

No spam. One reply within one business day. Or skip the form and book a call.

Thank you. Your brief is on its way. Grab a time now so we can talk it through.

Prefer LinkedIn? Connect with Harish.

Book a callFree lead test