OpenAI has built the world's most-used consumer AI product, with roughly 900 million daily active users on ChatGPT according to Leiphone's analysis of the company's strategy. On June 2, 2026, the company pointed that distribution at a different prize: enterprise knowledge workers who pay for finished work, not chat.
The vehicle is Codex, OpenAI's developer and agent platform, the coding assistant that grew out of the original 2021 Codex model and has since expanded into a general-purpose agent surface. At the company's "Intelligence at Work" livestream, OpenAI unveiled six role-specific plugins: Data Analytics (paired with Snowflake, Databricks Genie, Hex, and Tableau), Public Equity Investing (Moody's, FactSet, S&P), Investment Banking, Product Design, plus two more roles aimed at sales and creative work. Each plugin arrives pre-loaded with 62 business applications and 110 automated skills, and five additional roles (Corporate Finance, Private Equity Investing, Marketing Strategy, Strategy Consulting, Legal) sit on the public roadmap, per the OpenAI "Codex for Knowledge Work" post and Digital Applied's summary of the event.
Underneath sits GPT-5-Codex, a model OpenAI describes as roughly 25% faster than its predecessor and designed to combine coding performance with broader reasoning and professional knowledge. The six plugins are not six chatbots. Each one is wrapped around a workflow: data sources, common tasks, and pre-built skills that turn the model from "answers in a chat box" into "deliverables in a tool."
The commercial weight behind this pivot is already visible in OpenAI's own numbers. According to the company's "The Next Era of Knowledge Work" report, Codex has passed 5 million weekly active users, and knowledge workers now account for about 20% of that base and are growing more than three times faster than the developer segment. On the revenue side, Leiphone reports that OpenAI's chief revenue officer, Denise Dresser, told CNBC that enterprise (B-side) revenue is already roughly 40% of total and is expected to reach parity with consumer revenue by year-end.
Why now? The Chinese analyst commentary Leiphone assembled (from observers 朱颐均, 葛工, 韩松涛, and 吴哲) reads the move as a deliberate correction of the GPT Store's 2023 failure: 300 million GPTs were built, but none became sticky, because the underlying model could not deliver finished work. GPT-5-Codex's agentic planning and tool-use, wrapped in workflow-integrated plugins with named enterprise data sources, is the bet that this time the model can.
The same Leiphone analysis frames this as a strategic divergence with Anthropic, OpenAI's closest frontier-lab rival. Anthropic, per Leiphone, filed a confidential S-1 with the SEC around June 1, just before OpenAI's announcement, and already ships 11 Cowork plugins that the piece describes as broader in scope and more open-source-friendly than OpenAI's six. The reading: OpenAI is positioning as a platform and distribution entry point (an App Store for the office), while Anthropic positions as the per-seat base layer. Both are going after the same enterprise dollar, but with visibly different shapes.
That platform framing also surfaces a real risk for OpenAI's own partners. Codex Sites, the hosting surface previewed alongside the role plugins, keeps apps, data, and users inside OpenAI's rails. Tools like Canvas, Figma, Vercel, Wix, Webflow, and Lovable gain distribution today, but they also face long-term competition from the six official plugins now covering high-value verticals.
For the average ChatGPT user, the open question is whether the consumer product visibly keeps pace as engineering weight shifts toward enterprise. Brand dilution is a concrete risk if the free ChatGPT experience stagnates while the contract business accelerates. Pedaily's framing of the pivot likewise emphasizes the trade between recurring enterprise revenue and consumer mindshare.
The next 12 months will turn on three concrete triggers. First, whether the in-ChatGPT Codex integration OpenAI described on June 2 as "rolling out in the coming weeks" actually ships; until then, Codex and ChatGPT remain separate surfaces. Second, whether enterprise revenue tracks Dresser's stated trajectory toward parity with consumer. Third, whether the six roles expand into the five on the roadmap fast enough to make the App-Store metaphor stick.
The pivot is not a repudiation of OpenAI's 900 million consumer users. It is a bet that the same model, distributed through six office roles, can extract more revenue per user than a free chat ever did. Whether the enterprise sales curve justifies the consumer mindshare trade is the question the next year will answer.