4 Ways Machine Learning and AI Will Be Used to Improve Code

By Brian Rue, CEO and co-founder of Rollbar

The way software is built has changed immensely in recent years. For many companies, it's a more continuous process with frequent - sometimes daily - releases, a microservices architecture, and cloud-native applications. Businesses want to move quickly so they can stay competitive and keep pace with delighting customers. One thing that hasn't kept pace, though, is how developers deal with code errors and bugs.

The typical process is reactive and manual. Best case scenario is the team gets an alert that something's wrong. But many times, issues aren't surfaced until enough users report problems.

Either way, organizations need to investigate such problems to figure out why the issues are happening and the best way to resolve them. This can take days or weeks.

Wouldn't it better to minimize the effort to resolve bugs, giving that time back to developers, and getting ahead of issues before they impact users? The good news is that's already starting to happen thanks to machine learning and AI being introduced into error response tools and processes. Here are the four ways this will be adopted by best-in-class dev teams in 2021:

Predict Errors During Pre-Production

Catching and resolving error quickly is great. Preventing them from happening is even better.

Machine learning can analyze the entire history of your code and how it's changed over time to understand what errors you've had in the past. It can use that to predict when errors will happen again based on similar patterns of code - so you don't make the same mistake twice.

Right now, it will let you know where it thinks the problem will occur. Further down the line, we'll start to see machine learning and AI working together to suggest on code changes to fix it.

Anticipate the Severity of Errors in Production

Predicting issues as you're writing new code is great, but errors still happen.

Maybe you have a new integration that causes something else to break. Whatever it is, you want it fixed quickly and easily. But you probably also have a queue of issues, new integrations, and other updates that your team needs to address. So, prioritization for dev teams is paramount. That's where machine learning can help.

With the initial information on a new error, machine learning can correlate that against other information about your product and users, or anything else you let it analyze. With that information, you can decide what to do: take immediate action to prevent something serious, let it continue because it's minor, make a small correction until you can dig in later, etc.

For example, maybe the issue is only affecting Android users, and they make up 20% of your user base. That's significant enough that you want to fix it quickly. But maybe you have something else that's affecting 95% of your users. Knowing that helps you prioritize what you should be focused on next. In the year ahead, more dev teams will benefit from this knowledge.

Enjoy Better Error Classification

Development teams already have tools and systems to alert them about problems. That's why many are on call just like medical professionals - to deal with issues immediately.

But instead of a single alert, you get more noise than you can handle. A hundred alerts at once can be overwhelming. The first thing you'll want to do is see whether all the alerts were caused by the same thing. Getting accurate information about how to group errors can be helpful.

Grouping solutions, or engines, already exist. But they're built on hard-coded definitions. So, if a new bug pops up that was never defined in the engine, such solutions won't recognize it.

However, a solution built on machine learning will continuously update its definitions. That's how we built our grouping engine at Rollbar. Better error signals also open up the possibilities on how to automate the next steps to take.

Benefit from Proactive Decision Making

If developers can trust the error signals they get, because they're getting better information from their grouping engine, they can move much more quickly to minimize the impact to users.

If AI sees a major error that's affecting 95% of users on iOS, the AI solution can automatically revert to the previous known version of that code that worked properly.

Our newly launched AI-Assisted Workflows are already doing this. Based on criteria users set up, progressive deployments can be halted if an error is found, or feature flags can be automatically turned off for the same reason.

We're going to start to see even more ways to automate many of the manual tasks when responding to and remediating errors. But 2021 is going to be the year that AI and machine learning begin to become table stakes for dev teams. The organizations that invest the most in these solutions will be the best poised to get ahead of the competitive pack.

Some of the ways that machine learning and AI are being used may not seem like true game changers. But we're only at the beginning of this evolution in improving code. As teams start to adopt these solutions in 2021, it'll build the foundation to go even further in the future.

##

About the Author

Brian Rue is CEO and co-founder of Rollbar, the leading Continuous Code Improvement platform, helping developers to proactively discover, predict, and remediate errors with our real time AI-Assisted Workflows. Rollbar also has an Automation-Grade Grouping engine, which is powered by machine learning to reduce noise in error signals to delivering better error classifications, which is the foundation of our automated workflows. Simply put: Rollbar wants to help developers build software quickly and painlessly.