OpenAI Brings Git AI Founders In-House to Measure Codex Performance

The measurement capabilities that Git AI has provided to the broader market are now moving inside OpenAI's walls. The company announced Friday that it has brought aboard Aidan Cunniffe and Sasha Varlamov, the two founders behind Git AI, an open-source utility designed to measure how much code AI systems generate and to evaluate the performance and financial efficiency of coding agents.

In a LinkedIn post made late Friday evening, Cunniffe disclosed that both he and Varlamov would be joining the Codex team at OpenAI. Their role will center on equipping organizations with improved visibility into how well coding agents function and what business value they deliver. Cunniffe characterized the move with the phrase: "Same mission, bigger stage."

While specifics about how the integration will unfold remain limited, Thibault Sottiaux, a member of OpenAI's technical staff working on ChatGPT and Codex, confirmed via X that the organization intends to leverage Git AI's capabilities to help enterprises grasp Codex's real-world impact. "We'll make it easier for businesses to see where Codex is making a difference when working through problems for individuals and teams," Sottiaux stated.

Cunniffe elaborated on the partnership in a blog post released Friday, emphasizing that both organizations share a commitment to evaluating how well AI coding tools perform and deliver value. "OpenAI shares our belief that users should have the data to compare model performance and understand the ROI of every token they spend," he wrote.

https://x.com/thsottiaux/status/2098569976143806918?ref_src=twsrc%5Etfw

How Git AI Works

Git AI functions as an open-source extension for Git that monitors code generated by artificial intelligence. The system attaches metadata to each line of code, recording which agent, model, and prompts created it. This metadata persists even after code undergoes typical Git operations like commits, merges, or rebases.

The broader purpose is to give development teams visibility into several key metrics: what percentage of AI-generated code actually ships to production, how frequently that code requires rework, and where computational resources and API tokens are being consumed. Organizations can then use this information to evaluate different coding agents and models, and to determine whether their spending on these tools translates into genuine productivity gains.

Git AI currently works with a wide range of coding agents and models. The supported list includes OpenAI Codex, Anthropic's Claude Code, Cursor, and Google's Gemini CLI, along with background agents such as Codex Cloud, Claude Web, Cursor Agent, and Devin.

Once installed, supported agents automatically report their modifications to Git AI. From a developer's perspective, the workflow remains straightforward: they continue working with their preferred coding tools and Git as normal. Git AI then provides a breakdown showing which portions of the codebase came from humans versus AI, and traces individual lines back to their source agent.

Background

Cunniffe has navigated this trajectory before. In 2018, he launched Optic, a developer tools company focused on open-source utilities for API development. Atlassian acquired the company in 2024, and Cunniffe subsequently took on a role as principal product manager at Atlassian, overseeing go-to-market strategy for developer tools.

Git AI emerged while Cunniffe was still employed at Atlassian. He and Varlamov began developing the project during the summer of 2025 as a side endeavor, driven by a fundamental question: what proportion of their code was actually being written by AI, and what became of that code afterward?

By November, when the first major release arrived, Cunniffe was publicly referring to Git AI as a "weekend project" that had expanded significantly. He departed Atlassian in January 2026 to establish Git AI as a standalone venture, creating a commercial entity built on top of the open-source foundation.

Now, less than a year into that commercial venture, both Cunniffe and Varlamov are transitioning to OpenAI. The implications for Git AI as a business remain uncertain, though the departure of both founders suggests the commercial operation may be discontinued while the open-source project persists.

Cunniffe indicated that OpenAI intends to maintain support for the open-source initiative. "We'll keep investing in open source, while also giving enterprises the data they need to build effective software factories," he stated.

Whether Git AI can maintain its independence across multiple models while operating under OpenAI's stewardship will be a significant development to monitor going forward.

Source: The New Stack