AI and OpenTelemetry are shaping what comes next at Embrace

Observability should start where users expect reliability: on the screen. Users don’t encounter your infrastructure; they experience your code when they’re trying to make a payment, book a trip, or like a photo. Embrace collects and draws insights from production data to let every frontend and mobile developer create apps that fulfill users’ expectations.

But understanding the data we have or making use of it to actually resolve an issue isn’t necessarily simple. We have a lot of data, and remembering which attributes to queries, how to group certain granular data in more meaningful groups, or what data to focus on inside a flamegraph often makes the full power of our platform inaccessible to most developers.

With all of this data, how can we better leverage AI to help us discover the things we need to work on, explore the character of the data to get insights, and apply those insights with the help of AI-powered IDEs and Agents?

Today, I’m introducing Embrace’s unique approach to building an MCP server, using query translation to transform raw telemetry into actionable insights, and more. By leveraging AI across our product suite, we’re helping developers cut time-to-root-cause, highlight what matters most, and focus more on building rather than fixing.

Observability is most effective when it goes beyond data, and instead delivers insights. To bring order to the chaotic, multidimensional world of user-facing raw telemetry, Embrace standardized the information coming from our collection SDKs by committing to OpenTelemetry. Now every mobile app or web application on the Embrace platform speaks the same language and is designed for interoperability, extensibility, and choice.

OpenTelemetry is designed for precisely the highly-contextual, high-cardinality questions that we ask of mobile and web apps. The well-defined data shape, combined with Embrace’s full fidelity of automatically captured telemetry, means the analysis covers the highest-possible surface area of client activity. Using OTel, our opportunity is now to create tools that explain the interaction of technical operations and user activity from the perspective of the client:

Can we help development teams solve not only crashes and errors, but find out why cart checkout performance is slowing specifically in the Netherlands? Can we help customers learn why users stopped interacting with the app when switching to the latest Android major version?

Can we go deep enough in the data to reduce the blind spots for our customers, so they can see real business value from what they learn on mobile and web? Starting with the launch of User Journeys, we are growing our product suite with tools that surface enterprise-scale concepts like reliability, stability, and performance, aided by AI. By leaning into our expertise in user-facing apps, we have the ability to give a full picture of what happens on the client, and be opinionated about what tools are valuable for mobile and web developers.

Production observability data provides a unique perspective into the issues real users face, like odd edge case exceptions you didn’t encounter while testing, code that functions poorly or not at all when a device is on low power mode, or ANRs on Android devices that can’t be replicated in a lab. But resolving many of these requires code-context paired with the unique context provided by observability data. To that end, we are building a Model Context Protocol (MCP) server to let agents and AI-enabled IDEs speak directly with Embrace. Hooking their Claude Code, Copilot, or other agent up to Embrace will allow you to answer questions that can only be answered by production data.

Using MCP, rather than a proprietary LLM or agent, allows us to create a technology-agnostic communication loop for developers looking into their hangs, crashes, or to rapidly fix javascript exceptions. Developers are already working in their IDE of choice, so providing more context with Embrace’s telemetry will let the agent at their fingertips answer questions about their code that affect users.

Critically, we’re choosing to provide the most context we possibly can, powered by the high cardinality data and user flow context we capture and create around your users, rather than getting caught-up trying to be another AI-powered tool removing you from the tools you already know and use every day.

Embrace’s dataset is a reflection of the deep insight of our instrumentation, paired with the context we build around user journeys and behavior. That dataset can be huge, with many automatically and custom instrumented spans, context-rich logs, and attributes decorating that telemetry to provide rich context. Looking into a common client scenario – receiving a network response in order to trigger full rerenders of UI while simultaneously downloading files in the background – might stop developers from knowing where to start.

But being able to ask the right questions normally requires observability superusers to have a working understanding of all of the data and dimensions of their app. Is this feasible for new hires, teams segmented by module or business unit, or developers working in unfamiliar areas of their application? Code assistants and AI in the dashboard can ramp up these team members to explore cohorts and dimensions that they may not otherwise have familiarity with.

We are simplifying this question-asking experience in our dashboard to reduce the toil of issue discovery, so engineers can learn about their apps without requiring deep knowledge of the entire app system. Using AI, we can translate natural language questions like “why did checkout span performance degrade in the most recent version” into filters, charts, and workflows that show the entire picture.

This translation layer will let developers surface user-centric signals like regression, drop-offs, and unresponsive UI without having to learn a query language. Democratizing querying allows each member of the team to ask meaningful questions about the effect of technical performance on their apps.

AI has the opportunity to speed time-to-root-cause across the Embrace product suite. By mapping common patterns, sequences, and cohorts, AI can be an Embrace-specific assistant that helps engineers find user outcomes in raw telemetry.

What if anomalous performance traces could be instantly linked to new OS versions or concurrent slow network requests, and then surfaced in the Embrace dashboard? Or if flame graphs could automatically tell you that a third-party payment SDK was the common cause of hangs and needed to be wrapped to help users actually pay?

Using AI to learn and create insights from your data will let all developers find what matters. Highlighting the correct telemetry for teams will let them act on the insights in ways that they can trust and repeat, to dive deep into user cohorts, draw quick conclusions about production issues, and correlate user intention to technical signals across their apps.

By building in AI opportunities to make better sense of data, Embrace is helping teams graduate from monitoring to intelligent, user-focused observability, and we can’t wait to see what our users achieve with these new capabilities! Request a demo and talk to our team about how we’re using AI to help your engineers get the most out of user-focused telemetry.