Customer Support Tickets That Code

Software developers are starting to combine the two most successful uses of large language models: coding and customer support.

Specifically, they’re using AI to automate the process of turning customer complaints about software issues or feature requests into new code. Previously, that could be a long and tedious process, which involved sending customer support tickets to a company’s support team, directing those issues to the relevant engineer, and coding up and testing a fix to the issue. 

AI cybersecurity startup Semgrep, for instance, routes customer complaints about bugs in its app or requests for new features to Anthropic’s Claude model, founder and CEO Isaac Evans said. Claude handles the fix or codes the new features in a sandboxed, isolated version of Semgrep’s app. After a human engineer tests the changes, Semgrep merges them into the real Semgrep app, which primarily helps developers scan their code for bugs and vulnerabilities, he said. Right now, Semgrep is using this approach for a handful of customer support tickets per month, but Evans expects that number to grow quickly, he said.