LaunchDarkly expands into AI ops

AI Configs matter because they let LaunchDarkly sell into the fast growing AI operations budget, not just the software release budget. The same control loop that decides which users see a new feature can now decide which prompt, model, temperature, or routing path an AI request uses, and can change that decision instantly without a redeploy. That makes LaunchDarkly a live traffic controller for production AI, not just a flag vendor.

• The product is concrete, not conceptual. Teams define prompts, model choices, and rules in the dashboard, then the AI SDK evaluates them at runtime. A free user can be routed to a cheaper model, a regulated user can get stricter instructions, and 10% of traffic can test a new model version with rollback if quality drops.
  1 sacra 2 launchdarkly
• Online evaluations push this into a separate monitoring budget. LaunchDarkly made Online Evaluations generally available in February 2026, adding built in judges for accuracy, relevance, and toxicity on live model outputs. That moves the product closer to AI quality control and closer to observability and MLOps spend than classic feature flag spend.
  1 sacra 4 launchdarkly 5 launchdarkly
• The competitive opening is that many rivals still start from a different buying center. Statsig is strongest where product analytics and experimentation lead the sale, while Harness gained Split to bundle flags into broader delivery tooling. LaunchDarkly is using AI Configs to deepen its position with platform and release teams that already own production safety and rollback.
  1 sacra 6 sacra 7 prnewswire

This points toward LaunchDarkly becoming the control plane for high change software systems, where code releases and AI behavior are managed in one place. As more production traffic depends on model selection, prompt tuning, and continuous output scoring, AI Configs should expand LaunchDarkly from release tooling into a larger runtime governance layer.