March 2025 Summaries

LaunchDarkly's new AI Configs now support bringing your own model, unlocking more flexibility for supporting fine-tuned models and running models on local hardware. Ollama is an open-source tool used to run large language models locally. A tutorial demonstrates how to connect LaunchDarkly with Ollama using Python, creating a custom model AI config that tracks metrics such as latency, token usage, and generation count. The tutorial showcases the capabilities of reasoning models and provides guidance on tracking metrics, advanced targeting capabilities, and further reading resources for runtime model management.

The text discusses the challenges and complexities of deploying machine learning models, particularly when it comes to managing configurations, prompts, and runtime parameters. It highlights the need for a more flexible and scalable approach to deployment, which is where AI Configs come in. AI Configs provide a control plane for managing AI features at runtime, allowing developers to separate model configuration from application deployment and providing tools for safe runtime updates. The text covers key components of AI deployment, including model configuration, prompts, and runtime controls, as well as resource planning, best practices for deploying AI models in production, and strategies for monitoring and optimizing AI experiences. It emphasizes the importance of progressive delivery, monitoring metrics, and creating targeted experiences for different user segments to achieve long-term success with AI deployments.

This tutorial explains how to create a Flask application that uses LaunchDarkly feature flags to dynamically change the color of Google Maps markers. The application is built using Python 3.6 or newer, Visual Studio Code or another IDE, and the Google Cloud Console for a free API key. The tutorial covers setting up the environment, creating a feature flag in LaunchDarkly, configuring the Flask application, and integrating the LaunchDarkly SDK to evaluate the feature flag. The example demonstrates how to toggle the color of markers on a map by changing the value of the feature flag, showcasing the benefits of using feature flags for real-time experimentation and deployment of web applications.

With LaunchDarkly's expanded AI Configs, teams can more easily test, optimize, and manage AI-powered features in production. This is achieved through the introduction of two new capabilities: AI Experiments and AI Versioning. AI Experiments allow teams to run experiments to compare different prompts and model configurations, ensuring that changes improve quality without introducing unnecessary risk. Meanwhile, AI Versioning enables teams to track, compare, and revert AI configurations, reducing uncertainty around updating models and prompt snippets. These tools bring feature management and experimentation best practices to AI development, facilitating safer, smarter, and more efficient AI releases.

The integration between LaunchDarkly Experimentation and Snowflake enables Warehouse Native Experimentation, which allows teams to run and analyze experiments directly within their data warehouse. This eliminates compliance risks and data governance challenges by providing a single source of truth for experimentation data. With this integration, teams can leverage advanced analytics and AI-powered insights from Snowflake's platform, such as Elastic Compute and Cortex AI, to drive faster, data-backed decisions. The three pathways to run experiments include LaunchDarkly-hosted analysis, custom warehouse analysis, and Snowflake Native Warehouse Experimentation, each offering different levels of setup and integration. By using this integration, teams can optimize subscription conversions, experiment with high-fidelity data, and unlock new levels of experimentation velocity, accuracy, and scalability.

LaunchDarkly has helped its customers achieve significant results in their release processes, developer experiences, time and cost savings, customer churn reduction, and overall business outcomes. By using LaunchDarkly, teams have reported an average of 24.2% faster time to market for new features, a 16% average reduction in downtime, and an 8.86% average increase in freed-up developer time. The platform has also enabled users to reduce customer churn by 6% or more, with 46% of LaunchDarkly users reporting this achievement. Furthermore, LaunchDarkly has facilitated controlled, faster deployments, simplified feature flag management, and experimentation and A/B testing capabilities. These improvements have led to a ripple effect on the organization, resulting in releases attuned to customer needs, fewer release-related customer support tickets, stronger customer retention and satisfaction, and more meaningful business outcomes.

The tutorial demonstrates how to add a kill switch feature flag to a Flask application using the LaunchDarkly Python SDK. The kill switch allows for quick shutdown of features in the application, such as external API calls, to prevent disruptions or spam. The tutorial covers setting up the development environment, creating and configuring a kill switch flag, and integrating the LaunchDarkly client with the Flask application. It also provides examples of how to use the kill switch feature flag to enable or disable specific features in the application.

AlayaCare is transforming home and community care with an end-to-end platform that aims to improve patient outcomes. To scale software releases without disrupting critical healthcare service delivery, they've adopted a customer-friendly pace of change approach using LaunchDarkly. By standardizing their deployment strategy into "gentle deployment," AlayaCare categorizes releases as events or non-events and manages triage deployments to minimize disruption. The company has seen significant benefits, including reduced incidents, increased developer productivity, and boosted satisfaction ratings, with results such as a 15% decrease in time spent on coordinating releases and debugging issues. With LaunchDarkly, AlayaCare uses feature flags for better control and flexibility, focusing on parent-child flagging, standalone flags for minor updates, and proactive testing and monitoring. By tracking data and internal stakeholder surveys, the company aims to continue improving their release philosophy and sets a standard for software deployment in the healthcare space that prioritizes speed and safety.

At the end of February, we launched a new docs site, partnering with Fern to handle infrastructure and UX support. We combined our product and API docs into one site and made under-the-hood improvements, including updated root URLs, faster page load times, easier code snippet navigation, improved search results, and a guided feedback mechanism. The goal is to give the docs team more focus on writing high-quality content while Fern handles infrastructure and design support, benefiting LaunchDarkly developers and customers alike.

This tutorial demonstrates how to run large language models locally using Ollama and query the results from a Node.js application. It also shows how to create a custom model AI config with LaunchDarkly that tracks metrics such as latency, token usage, and generation count. The tutorial covers the benefits of running LLMs on local hardware, including enhanced data privacy, accessibility, and sustainability. It provides instructions on how to install Ollama, connect it to Node.js, and create a custom model AI config using LaunchDarkly. The tutorial also includes examples of how to use the custom model AI config to generate responses from different models and track metrics such as latency, token usage, and generation count.

** LaunchDarkly is a powerful experimentation tool that enables businesses to measure the impact of their releases on customer behavior. It allows engineers to control the deployment and release of software safely, making it an ideal platform for experimenting with new features. Unlike other tools that may bypass the engineering process or create silos between teams, LaunchDarkly integrates feature flagging and experimentation into a single platform, ensuring collaboration across product, business, and engineering teams. By tying business metrics directly to the change made in code, LaunchDarkly provides a direct signal between release changes and business outcomes, making it easier to prioritize features that delight customers. With its robust feature management capabilities, LaunchDarkly enables businesses to streamline workflows, reduce development resources, and focus on initiatives that truly impact customers.

Feature management powered by LaunchDarkly enables government agencies to implement policy updates rapidly and safely without having to wait for lengthy development cycles or full system redeployments. This approach allows agencies to deploy changes quickly, maintain security and compliance standards, test and validate updates in real-time, roll back changes instantly if issues arise, and focus on security and compliance with controlled rollouts. By using feature flags, agencies can empower themselves to work smarter, not harder, and respond to policy shifts with agility while continuing to serve the public effectively.

This tutorial teaches developers how to add a kill switch feature to their Sinatra applications using the LaunchDarkly Ruby SDK. The process involves creating a new Ruby project, setting up a simple Sinatra application, and integrating the LaunchDarkly Ruby SDK to manage external API calls. The tutorial also covers configuring Puma to ensure that LaunchDarkly flag changes take effect immediately. By following these steps, developers can create a flexible and responsive application that can quickly disable external dependencies during emergencies or unexpected issues.

The tutorial discusses the implementation of kill switches in a Java Spring Boot application using the LaunchDarkly SDK. A kill switch allows developers to quickly disable features in their application without causing additional disruption. The process involves creating a feature flag, integrating the LaunchDarkly client, and configuring the flag to control the behavior of specific code paths. The tutorial also covers how to integrate a third-party API and demonstrate the capabilities of the kill switches. By following these steps, developers can add kill switches to their Java Spring Boot application and improve the reliability and maintainability of their software.