OpenAI Showcases Company-Wide Codex Use at a 5-Person Startup
On June 2, 2026, OpenAI's official developer account published a case study in which the just-five-person startup "Proaction" uses the company's AI coding agent "Codex" across its operations, from sales to development. It was presented as a strong example of operating an AI agent as an "extension of the team." Source
What Happened
OpenAI Developers (@OpenAIDevs) posted a case study about Proaction, which develops fleet management software. Despite being a five-person team, the company reportedly uses Codex for a wide range of work, including sales demos, support follow-ups, marketing assets, and engineering work. A sense of speed was shown in which "today's customer call becomes tomorrow's new product workflow."
Proaction is a company that provides an OS for complex fleets, integrating maintenance, inspection, vendors, telematics, and reporting, and automating workflows from when an issue arises to its resolution. Proaction official
Background and Significance
Codex is a cloud-based software engineering agent that OpenAI released around May 2025. It has evolved from its initial code-completion tool, now supporting GitHub integration, parallel task processing, and autonomous execution. InfoWorld, Fast Company
It features cloud-isolated environments, Git integration, and long-running autonomous execution (there are even reports of 14-hour continuous tasks), and it is offered mainly to ChatGPT Pro/Enterprise/Team users. MindStudio
OpenAI also makes heavy use of Codex internally for the autonomous operation of its data platform and for creating PRs, and it has positioned Proaction as a strong example of an outside company showing productivity gains for a small team. Forbes Note that specific figures such as the number of tokens used or cost reduction rates were not disclosed this time, and the Proaction official site likewise makes no mention of benchmarks or pricing.
Reactions
This post came immediately after being published, and engagement remains low, with only about 11 to 12 likes. Source
Reactions to Codex in general have been notably positive. Within OpenAI, there have been voices saying "we wrote 80% of PRs with Codex and released Agent Builder in six weeks," post, and there is also a report of completing a frontend task spanning more than 12 hours. On the other hand, as general points, the importance of context management and the fact that human judgment can become a bottleneck are also being discussed. Voices specific to Proaction users cannot be confirmed on X at this time.