February 2025 Summaries

When deciding to roll out a new feature, it's challenging to observe the impact on key metrics without splitting the world into parallel universes. Instead, experimentation helps approximate this by randomly selecting groups of users and delivering different experiences to them, allowing for observation of results. This approach is valuable in various scenarios, including stakeholder disagreements, replicating another product team's win, when UX research is inconclusive, sales team pressure, engineer-driven ideas, efforts between MVP and full feature, tech licensing decisions, directive from above, developing personas, and big, impactful ideas. Experimentation provides a way to test hypotheses, gather evidence, and make data-driven decisions, ultimately driving business outcomes and informed product development.

The text discusses LaunchDarkly's new AI configs feature, which allows developers to deploy multiple AI models in a single application without redeploying code. The tutorial guides users through setting up an AI config for the DeepSeek R1 model and comparing its results with those of another model, GPT-4o. Users learn how to create and manage AI configs, including defining variations, setting percentage rollouts, and monitoring metrics. The feature aims to reduce risks associated with deploying AI models in production by decoupling configuration changes from deployment.

As an engineer, I used to equate innovation with risk, but Guarded Releases changed my approach, shifting from reactive chaos to proactive control. Deployments now feel deliberate and confident, rather than a gamble. The new way allows me to focus on building without fear of everything breaking, and has saved countless hours spent in firefights, strengthened our team, and put customer experiences first. I've found that innovation can be safer, seamless, and fully controlled with Guarded Releases, enabling me to push boundaries without fear, knowing my releases are secure.

We're excited to announce the introduction of Warehouse Native Experimentation, a new collaboration between Snowflake and LaunchDarkly that unifies feature management and experimentation with data analysis within Snowflake's AI Data Cloud. This allows teams to design, run, and analyze experiments using trusted datasets, unlocking deeper insights and making critical decisions more quickly. The integration removes barriers to experimentation by connecting trusted business data with tools used to run experiments, ensuring meaningful insights and accurate measurement of experiment impact. With Warehouse Native Experimentation, teams can run experiments in LaunchDarkly and make decisions based on enhanced results powered by Snowflake's AI Data Cloud, using trusted, organization-wide metric data from Snowflake.

Companies are building and releasing new software more quickly, often at the expense of performance or reliability. To mitigate this, enterprises must adopt strategies focused on safety, reversibility, and continuous improvement to ensure sustainable success. A progressive release strategy, where updates are rolled out to a small group of users first, allows for quick fixes with minimal disruption. This approach also enables feature-level observability, which captures relevant data during critical releases, reducing noise and allowing teams to concentrate on urgent issues. Additionally, the importance of reversibility is highlighted, as systems should allow immediate reversal of changes when issues arise. Finally, a culture of experimentation and continuous learning supports innovation, reduces risk, and provides valuable insights for ongoing refinement, helping to ensure that each release builds on the last and drives continuous improvement.

The tutorial demonstrates how to integrate DeepSeek's free r1 model into a Python application using LaunchDarkly AI Configs. The process involves setting up a developer environment, building the Python AI config application scaffolding, and retrieving AI config values within the application. The tutorial also covers generating DeepSeek AI results in the Python application and tracking metrics on the DeepSeek AI Config. The goal is to create an AI-powered educational chatbot that helps students with homework and material understanding, using the DeepSeek model's free r1 reasoning model.

Event Explorer is a powerful new tool for tracking, verifying, and investigating events sent to LaunchDarkly, simplifying metric creation and tracking while providing greater visibility into event data. It addresses the limitations of Live Events by offering historical data insights and analytics, allowing users to analyze past events and make informed decisions. With Event Explorer, users can gain full visibility into event data, distinguish between active and inactive events, quickly view linked metrics or create new ones, search for events, monitor trends and origins, filter by environment, and get better visibility into the data they're sending to LaunchDarkly. The tool is available to all customers on all plan types and can be accessed through the public documentation or directly in the LaunchDarkly dashboard.

The LaunchDarkly team overcame challenges with traditional A/B testing tools by adopting a feature flags-based approach to experimentation, eliminating the need for DOM-based JavaScript injections and resulting in faster load times, more reliable experiments, and improved user experience. This shift enabled them to run multiple experiments simultaneously, increasing testing velocity and leading to significant improvements in conversions, cost per lead, and website performance. By streamlining experiment setup and execution, the team was able to focus on strategic initiatives, fostering a culture of continuous improvement and innovation.

The tutorial provides a comprehensive overview of using feature flags in Flask applications. It explains the benefits of using feature flags, such as testing and iterating on different configurations at runtime, and how to integrate LaunchDarkly into a Flask application. The tutorial covers the creation of a new feature flag, setting up the development environment, building the Flask application, navigating the LaunchDarkly dashboard, creating another flag, adding audio and video to the Flask webpage, and exploring what's next for building Flask applications with feature flags. Throughout the tutorial, it provides step-by-step instructions, code snippets, and explanations of key concepts, making it a valuable resource for developers looking to incorporate feature flags into their projects.