Cancel the (Issue) Noise

For those who like to get straight to the point, here’s a quick summary of the latest noise-reducing improvements to the Issues experience:

• AI Grouping – available for early adopters, this feature significantly improves issue creation by leveraging advanced embeddings models to create semantic fingerprints of errors. Early results show a 40% reduction in issue volume for projects that have enabled this feature.

• ML Priority Alerts – available to early adopters on the Business Plan, this feature uses a machine learning algorithm to filter out trivial and unactionable problems, reducing alert volume by an average of 35%.

We’re feeling pretty good about these features. Might even go out on a limb and say they’re some of the most significant improvements to our error grouping and alerts ever. Need more convincing? Keep reading.

AI Grouping: Smarter, More Effective Issue Creation

Sentry has historically relied solely on heuristics-based fingerprinting to aggregate errors into issues. The complexity of modern JavaScript and new frameworks reduces the effectiveness of this approach. With the introduction of AI Grouping, we’ve taken a major step forward in reducing duplicate and single-event issues.

AI Grouping replaces traditional, rule-based fingerprints with “semantic” fingerprints, crafted by a model that understands the context and meaning behind errors.

This model creates a detailed representation of each issue, allowing it to intelligently determine whether new errors stem from the same underlying problem. This approach excels in handling the nuances of code refactors, version updates, and function recursion, areas where static heuristics can fall short.

Currently, AI Grouping is enabled by default for early adopters with JavaScript, Ruby, and Python projects. If you’re interested in trying it out, simply head to your organization settings page and toggle the Early Adopter switch.

ML Priority Alerts: Focus on What Really Matters

Alert fatigue is real, especially in large, complex projects. While AI Grouping improves how we create issues, the fact that Sentry’s default alert rules notify on every new issue created means that you will get alerts for everything - including the debug and info-level issues that don’t need to be addressed right away. ML Priority Alerts is designed to help you focus on the issues that matter by filtering out noisy alerts stemming from less important issues.

This feature integrates AI Grouping’s semantic understanding of an issue with a classification model that identifies non-actionable or low-priority errors. Since issues are now classified into different priority levels, the updated default alert rule with priority-based conditions enables us to proactively filter out low-priority issues from triggering alerts. To ensure that minor issues don’t escalate into major problems, we’ve also introduced Escalating Issues detection, which triggers alerts when significant spikes in issue volume are detected.

More Coming Soon

AI Grouping and ML Priority Alerts are just the beginning. We’ve been hard at work making improvements across the board, many of which are already live with more on the way.

From quality-of-life enhancements like removing archived issues from metric alerts to defaulting on new Inbound Filters to help cut down on the noise in frontend projects, we are making sure you and your team can easily fix the critical problems first. That said, we’re working on a few more capabilities you might not hate:

• AI Autofix, currently in closed beta, AI Autofix has shown significant progress since its initial announcement. It’s already capable of autonomously resolving over 60% of real-world software issues, and we’re gearing up to move into open beta soon
• A revamped issues search experience, making it easier than ever to query for the issues you care about and save individualized views for easy access
• In addition to static and % change metric alerts, Sentry will soon support anomaly-based metric alerts. This detection methodology is particularly beneficial for data with seasonal trends and requires minimal configuration

We’re actively working on these features and a few more. If you want to live on the edge, we can turn some of them on for you sooner rather than later. Find us on Discord, and let us know what you’re interested in or what you think of the improvements.