Company: LaunchDarkly
CEO: Edith Harbaugh
Website: https://launchdarkly.com/
About LaunchDarkly
LaunchDarkly’s journey
After experiencing software launches that turned into stressful late nights, LaunchDarkly CEO and co-founder Edith Harbaugh knew there had to be a better way. Drawing on her experience at companies including Atlassian and TripIt, she saw software teams relying on fragile in-house tools and manual processes that made deployments risky and experimentation difficult. LaunchDarkly was founded in 2014 to separate deployment from release, giving teams the ability to safely test, control and iterate on software in production.
Twelve years later, LaunchDarkly helps some of the world’s biggest companies gain greater control over software launches and deeper insight into how users experience their products. Trusted by 25% of Fortune 500 companies, LaunchDarkly employs more than 450 people and evaluates 45 trillion feature flags daily.
The AI era has made that mission more urgent. Generative AI is creating a new class of software behaviour, with AI agents building, testing and deploying at machine speed. Traditional DevOps tooling was not designed for the dynamic and unpredictable nature of large language model-based applications. Teams often resorted to embedding prompts and models into code, spreadsheets or manual workflows, making iteration slow, risky and difficult to govern.
A new solution was needed: one that gave developers runtime control and reliability, product managers and designers safe experimentation environments, and governance and risk teams the auditability and oversight they require.
Enter CodeControl and AgentControl
LaunchDarkly’s AI Configs, released at the end of 2024 and made generally available in April 2025, gave teams a control plane for GenAI applications. It enabled them to manage prompts, models, experiments, fallbacks and performance without redeployment.
The company is now building on that foundation with CodeControl and AgentControl. CodeControl gives engineering teams control over code in production, allowing them to separate deployment from release, progressively roll out changes, target experiences precisely and respond instantly when issues arise. This is increasingly important as AI accelerates the volume of code reaching production and places pressure on review and quality assurance processes.
AgentControl applies the same principle to AI agents. It gives teams real-time control over agent behaviour in production, enabling them to configure agents across teams and frameworks, benchmark quality before changes reach live traffic, release updates safely, observe performance with trace-level visibility and intervene instantly when an agent drifts or underperforms.
Challenges faced
LaunchDarkly has had to evolve at the pace of a rapidly changing market. AI-native software development has made quality, safety and performance far less predictable than in traditional applications. As a result, the company has adapted proven principles such as feature management, experimentation, observability and rollback to an environment where agents, prompts and models behave dynamically.
Our impact
LaunchDarkly is building the runtime control layer for this future, helping teams control code, control agents and create safer, more trustworthy AI-powered products at scale. It is still early days, but the direction is clear: CodeControl and AgentControl will help shape how modern software teams build with AI, enabling them to ship continuously, learn instantly and remain in control of what reaches users.
Want to find out what TechRound can do for your business?
