Snyk Uses Cortex AI to Provide Natural Language Access to Vital Insights

The cybersecurity landscape is forever changing, with new, more advanced risks emerging by the day. Tackling these risks requires new, more advanced security measures to match. Since 2016, Snyk has been providing the world’s developers with application security at scale, delivering a suite of tools that help identify vulnerabilities and embed mitigation measures within code. 

“Our goal is to make software more secure,” says Randall Degges, Head of Developer and Security Relations at Snyk. “We have lots of products that plug into different parts of the development cycle. This includes a static analysis testing service, tools for examining open source dependencies and finding vulnerabilities. We also use generative AI to rewrite chunks of code and semi-autonomously remediate vulnerabilities.”  

Data is integral to the service Snyk offers. Over the past ten years, the company has built a vast resource of data related to code bases and vulnerabilities that it uses to detect and fix its clients’ weaknesses. Its team of security analysts is constantly adding to this intelligence, analyzing every publicly licensed code base across the world to identify new issues. 

As Snyk has grown, taking on larger customers and facing greater demands for reporting, its data team realized its legacy platform would have to be upgraded. Since an initial migration in 2021, it’s been using Snowflake’s AI Data Cloud to provide a central store of data and deliver rapid, robust insights. Now, using Cortex AI, Snyk is making its vast resources of data available to even more users through an AI chatbot.

When generative AI capabilities became more advanced and opened up a whole new world of opportunity for tech-based organizations, Snyk saw an opportunity to improve its internal processes. The idea was to help go-to-market teams access the most useful information buried within the vast amounts of data the company holds. 

“Snyk, as a product, is very technical. The customer questions that teams are fielding are very complicated,” says Degges. “We have extensive internal documentation, so our idea was to feed that into a chatbot that would allow teams to get quicker answers to their queries, benefiting us and our customers.”

Using Cortex AI, Snyk built an AI chatbot that allows users to query data in natural language through Slack, the company’s primary internal communications platform. The chatbot uses Cortex Search to retrieve data from Snowflake and an inference layer to interpret the questions users ask. The result is instant answers to complex queries, without any coding required from users. 

“We’ve built a daily ingestion process using an application that runs across all of the unstructured data in Snyk,” says Oliver Armstrong, AI Systems Engineer at Snyk. “That includes Slack threads, documentation, blogs, articles, Snyk Learn lessons, and more. We’d already chosen Snowflake to help us centralize and store all our data in one place, so using it for this allowed us to keep all of our data within the same boundaries and ensure security.”

Originally, Snyk had planned to build this functionality in-house, but using Snowlake and Cortex proved an easier, cheaper, and faster solution. 

“Snowflake’s vector store structure makes it very easy to do filtering. The search service is easy to configure, and the API call is easy to use,” says Armstrong. “Cortex was also very robust and easy to use. It would have taken weeks to build this ourselves, which didn’t make sense when the capability we needed already existed.” 

Snyk’s AI chatbot now provides answers to approximately 2,500 questions a month, including those related to products, sales information and customer queries.

Snyk’s AI-powered chatbot saves employees hours they might normally spend searching for the right information to serve customers — especially people without technical data skills . The result is a far more efficient working environment, more satisfied employees and, above all else, a better experience for Snyk clients.

“People say it always helps them find the right documentation, which is really important when on customer calls or trying to walk people through solutions,” says Armstrong.

Ultimately, this faster access to information also helps Snyk improve its own products and services. Previously, product managers would spend a huge amount of time sifting through customer tickets and trying to triangulate information to learn what they should prioritize across customer issues, produce requests and future roadmaps. Now, AI can do the hard work for them, resulting in an estimated time saving of 1,250 hours a month. 

“With the Cortex Agents, we can return answers in less than a minute,” says Armstrong, “Whereas before teams would have been working across multiple sources of data to gain that insight.”

The success of the tool also helps improve the company’s internal documentation. Snyk regularly reviews data to see how its AI chatbot is used and what questions are asked most often. With those insights, the company is able to update internal resources in line with common queries, and that means better answers are then added to the bot.

Following the success of its chatbot, Snyk has begun to experiment with further use cases. Part of the company’s service offering is an educational platform called Snyk Assist, which provides developers with the latest best practice guidance on security measures. This platform now has its own chatbot, where data pulled from Snowflake is used to provide answers to its enterprise customers. 

“The platform provides a lot of really fun ways of learning,” says Armstrong. “But a lot of the time, when people are actually doing their jobs, they just want the fastest possible solution. Using data from Snowflake, we can provide context about our customers and their issues, and give them really comprehensive, in-depth technical answers — which means they can get back to coding more quickly.”

Snyk’s ambitions don’t end there. Plans are in place to provide the same functionality for more internal teams, and even customize the chatbot so it reacts differently to the needs of different users and their roles.