Graham McNicoll

Product development is forever changed, and the pace of the change is incredible. We've seen this with our own product, too. Engineering teams are using agentic coding tools such as Claude Code and Cursor to build features, manage flags, and run experiments directly via GrowthBook's MCP server and API. Experimentation and flag management always required a human. A person wrote the hypothesis. A person read the result and decided what to ship. I am watching that assumption come apart, as AI agents take over. And yet when an agent is shipping products and features at volume, the old failure modes do not go away. They scale with it. A premature decision does not care whether a human or an agent caused it. Neither does a flag nobody ever cleaned up. Speed without guardrails is just a faster way to ship a mistake. So we built GrowthBook 4.4 around this new mode of working. The platform should work the same way whether a human or an agent is driving. A few of the things that come with that: - Ramp schedules that stage a rollout with guardrail metrics attached, so a release can hold or roll back on its own instead of waiting for someone to remember the next step. - Approval workflows that apply to agents exactly as they apply to people. An agent can draft a change. A human still approves it. Same gate, same audit log, no special casing. - Stale flag detection over the API and MCP, so an agent can surface the dead flags in your codebase and clean up the references in one pass. That is the technical debt most teams defer forever. - A programmable experiment lifecycle, with templates that carry your org's standards, bandits without the sticky bucketing setup, and an API to conclude a test and record the decision. - An AI Data Analyst that answers questions about your metrics in plain language. It is aware of the metrics and fact tables already defined in GrowthBook, so it works from your trusted definitions across your warehouse. The through line is the same one GrowthBook has always had. We built it for teams that ship with discipline, that care about measuring what they release and being able to roll it back. 4.4 extends that to the agents now working alongside those teams. Same rigor, whether a person or an agent runs the loop. The full 4.4 breakdown is on growthbook.io