See what’s new in GrowthBook 5.2: Interleaving and Human in the Loop

GrowthBook 5.2 makes advanced experimentation simpler. Evaluate new ranking algorithms 50–100x faster than with a standard A/B test. With Human in the Loop, AI agents work on their own while a person approves the risky changes. And you can target millions of users with a CSV upload, with no payload bloat. In this video, the engineers behind 5.2 demo each feature live.

Chapters

• 0:00 Intro
• 0:09 Interleaving experiments, Watch here
• 3:49 Human in the Loop for agents, Watch here
• 6:03 Experiment page redesign, Watch here
• 7:26 Metrics with live preview, Watch here
• 9:08 New data onboarding flow, Watch here
• 11:48 Remote saved groups, Watch here
• 13:31 Wrap-up

Topics

• Interleaving experiments (private beta): test ranking and recommendation algorithms 50–100x faster than a standard A/B test. Enterprise customers can reach out to join.
• Human in the Loop for agents: decide what an AI agent can do on its own. Risky actions, like stopping an experiment or publishing to production, wait for a person to approve them in GrowthBook.
• Experiment page redesign: inline editing, feature flags built into the flow, and a launch checklist that’s always visible.
• Metrics with live preview: see results, SQL and sample rows before you save a metric.
• New data onboarding: connect your warehouse with clearer guidance, or start with the managed warehouse or Event Forwarder.
• Remote saved groups: target millions of user IDs from a CSV without making your SDK payload bigger.
• Also in 5.2: warehouse usage dashboards, experiment diagnostics, a free Event Forwarder tier, and contextual bandit improvements.

Get the full rundown in the 5.2 release post, or join the conversation on GitHub Discussions.