LaunchDarkly

LaunchDarkly operates a feature management and experimentation platform for software development teams. Its foundation is feature flags: developers install one of the company's SDKs, create a flag in the LaunchDarkly UI, wrap a feature in that flag key, and then turn functionality on or off, target segments, or run percentage and progressive rollouts at runtime without redeploying code. The platform adds targeting and segments, automated rollbacks, feature monitoring, release pipelines, flag dependency management, custom roles and policies, and flag-change approvals for enterprise governance. [2][5]

The company now positions the product as a "runtime control layer" spanning two product areas: CodeControl (feature flags, observability including error monitoring, logs and traces, session replay and auto-remediation, plus experimentation with A/B/n testing, multi-armed bandits, holdouts and warehouse-native metrics) and AgentControl (online and offline evaluations, adaptive triggers, agent observability, LLM traces and playground, custom judges, prompt snippets, agent definitions, approvals and RBAC). Documentation covers AgentControl configs, LLM observability, warehouse-native experimentation with external data warehouses, data export, and AI SDKs for Python, Node.js and Java. [0][3][5]

Operationally, the platform is delivered through a global flag delivery network using 100+ CDN points of presence, with a streaming architecture the company says propagates configuration changes in under 200 milliseconds worldwide, multi-region resiliency, and a 99.99% enterprise uptime SLA. The company cites SOC 2 Type II, ISO 27001, ISO 27701 certification, Data Privacy Framework alignment with GDPR and CCPA, and compliance with FedRAMP and HIPAA, and documents deployments for China, Pakistan, the European Union and federal environments. [2][3][5]

Founding story

Harbaugh and Kodumal were classmates at Harvey Mudd College and stayed in touch as they advanced in separate engineering careers, both encountering the risk of all-at-once code launches. As a product director at TripIt, Harbaugh used feature flagging - splitting a release into components and rolling them out in stages - and found teams were repeatedly rebuilding homegrown feature workflows. They founded LaunchDarkly in 2014 to provide centralized feature management, and Harbaugh has described the difficulty of landing the first ten customers. [4][6][7]

Business model

LaunchDarkly sells a SaaS developer platform on a self-service and enterprise basis; the site offers a free trial and sign-in flow alongside "book a demo" enterprise motions, published pricing and billing/usage pages in documentation, and an enterprise tier with governance features and a 99.99% uptime SLA. A third-party profile notes the company has also introduced professional services for onboarding. [0][2][3][5][6]

Subscription software; documentation includes dedicated pricing and billing-and-usage sections, and the product is offered via free trial, paid plans, and enterprise agreements with SLAs and governance features. [2][3][5]

Traction

The company reports more than 5,500 customers, roughly a quarter of the Fortune 500, 570+ employees, and 45 trillion feature flag evaluations daily (a platform page cites 50T+ evaluations per day). Earlier figures show growth over time: Forbes reported more than 2,000 customers and 20 trillion flags on peak days in 2021, and a third-party profile referenced over 4,000 customers. [4][5][6][7]

Latest developments

Recent product direction centers on AI: AgentControl for governing agents in production, LLM observability and instrumentation, adaptive flag and config triggers, new handler-based Python, Node.js and Java AI SDKs (with legacy AI SDKs moved to maintenance mode), auto-remediation and self-healing workflows, ClickHouse data export, a Vercel Marketplace integration, count-distinct warehouse metrics, and MCP server tooling - documented in changelog entries dated August 2026. The site also promotes a virtual event on August 26 covering six new features. [0][3][5]