OpenAI Codex Head Thibault Sottiaux Announces Welcome to Git AI Team Members Aidan and Sasha
OpenAI Codex Team Welcomes Git AI Founders
Thibault Sottiaux, head of OpenAI Codex, announced the welcome of Git AI team members Aidan and Sasha; this open-source tool measures the contribution of coding agents to codebases, and both parties will continue to maintain open-source and increase investment, helping enterprises understand Codex's role in individual and team problem-solving.
Git AI co-founder Aidan Cunniffe confirmed the integration into the Codex team on the same day. The product is a Git extension: it tags each line of code written by agents to specific agents, models, and prompts without altering workflows or adding repository hooks, retaining attribution even after rebasing, merging, compressing, and cherry-picking. The company claims the tool is running on hundreds of thousands of developer machines, with clients ranging from growth companies to Fortune 500 firms, used to compare model performance, token efficiency, and agent return on investment. Collaborators include Cursor, Cognition, Google, Factory, OpenAI, and GitHub, involving hooks, telemetry modes, and attribution standards.
Sottiaux currently oversees both Codex and the core ChatGPT product line, reporting to Greg Brockman. Codex operates an open-source command line with a multi-model interface, competing with Anthropic's closed-source Claude Code. In recent years, OpenAI has recruited talent from programming assistants like Cline to supplement Codex and has acquired other editor plugin teams. Git AI emphasizes cross-agent and local-first approaches, avoiding data lock-in with a single model vendor.
The buyer aims to prove that agent modifications can translate into output for engineering managers; the seller is OpenAI, integrating the attribution layer into Codex. The driving force is that enterprise procurement looks for return on investment, not just new model releases. Talent and open-source repositories are entering the Codex framework, with subscription and token revenues flowing back to agent usage. Beneficiaries are buyers who can compare various agents on the same Git history; under pressure are assistants that only report generated line counts without providing auditable attribution.
Source: Public Information
ABAB AI Insight
OpenAI is not just buying another completion plugin; it is acquiring the ledger layer of "who wrote this line of code." Enterprises will not place orders solely based on model scores; they want to see if agents reduce rework and how many tokens they consume. Git AI integrates attribution into Git history, allowing return rates to be calculated per commit. Continuing open-source is to preserve the standards already installed on hundreds of thousands of machines, avoiding the perception that once integrated into OpenAI, it only serves to account for Codex.
The capital path is talent integration into the product line. Codex expands from about forty people to the core product organization, through recruiting competitor teams and acquiring attribution tools, rather than developing a parallel logging system with Git. Aidan's side has already discussed hooks and telemetry with Cursor and GitHub; if OpenAI continues to maintain the standards, other agents will need to write data in this format, or they will not appear on enterprise dashboards.
In comparison to Git itself becoming the standard for code history, and Datadog turning observations into cloud billing attachments: once the attribution layer becomes the default extension, model vendors will compete on who can be included in that line of metadata. The industry phase shifts from "agents can modify files" to "changes can be audited." Claude Code is closed-source, while Codex is open-source with multi-model support, making Git AI's cross-agent support position the latter as a more neutral ruler—even if that ruler is now placed in the OpenAI office.
The structural change is a shift in pricing power: from charging by tokens to charging by attributable output. The mechanism is that without line-level attribution, agents are cost centers; with attribution, they can enter budget categories. Whoever controls the attribution format controls the denominator when enterprises compare models.
ABAB News · Cognitive Laws
Agents must be able to be accounted for line by line before entering budgets.
Whoever takes the open-source ruler defines the comparison standards.
Models are compared by generation, procurement by attribution.