Tabnine launches its code review agent

AI code assistant company Tabnine is launching a code review agent today that aims to help developers stick to their organization’s best practices and standards. It allows organizations to codify these rules either by providing the agent with documentation or by pointing it at a set of “golden code repos.”

The agent will passively review the code as the developer works in the IDE, flag issues as it spots them, and offer fixes when able.

As Tabnine President and CMO Peter Guagenti told me, many companies — and the developers they employ — aren’t even aware of the rules and best practices in their own documentation.

While one of Tabnine’s core tenants has always been to customize the agents based on an organization’s needs, the company also partnered with companies like Redis to collect some of their best practices and pre-trained its models on that. Starting today, Tabnine is opening this up to other vendors who want to provide their rules, too.

“Database companies are a great example of this,” he said. “Each one has its own pattern. Each one has its own ways of working that have good and bad outcomes — and when it has a bad outcome, they blame the product, not the code, right? So we think this is a great opportunity for AI to correct the behavior and actually make the products more successful.”

Developers can turn these pre-trained rules on and off as needed.

As for how correct the suggestions are, Tabnine argues that its review agents read the code like a human would. That also means that if the code is really obscure, it might miss an issue. But unlike human reviewers, it will read every line of code and not just skim.

It’s surely no coincidence that Tabnine is making today’s announcement on the first day of GitHub’s Universe conference. GitHub’s Copilot, after all, surely has the majority of the brand awareness among AI coding tools. But Guagenti doesn’t seem to worry too much about competing with GitHub.

“We think the business will stratify,” he said. “We think folks like Cursor and others, they’ll eat up the bottom of the market because a lot of people don’t want to write code. We think Copilot has to go for the fat middle — it has to be 80% for as many people as possible. However, that wasn’t [Tabine co-founders] Dror [Weiss’s] and Eran [Yahav’s]vision. Their vision was: how do I make the top 1,000 engineering teams in the world more productive, more successful, and write better applications? That vision is now starting to really come together.”

Subscribe for the industry’s biggest tech news