From Coding Engine to General Knowledge Work Assistant
Codex is an AI-powered agent from OpenAI that began as a coding specialist but is rapidly evolving into a general knowledge work tool for research, analysis, documents, and interactive content. Originally launched as a research preview in May 2025 and tuned for software engineering tasks, Codex could write features, fix bugs, and propose pull requests in isolated sandboxes. Since then it has grown into a core part of OpenAI’s productivity expansion, reaching 5 million weekly active users despite an already crowded market for AI coding assistants. The agent now sits inside desktop, web, and enterprise environments, and OpenAI has reset rate limits to keep up with demand, a sign it is scaling infrastructure to support broader adoption. While Codex still serves developers, its trajectory shows that coding is becoming only one slice of a wider AI productivity story.
Five Million Users and a Shift Toward Enterprise Knowledge Work
Codex’s path from research preview to 5 million weekly users took a little over a year, helped by tight integration into ChatGPT Business and Enterprise tiers. Enterprise teams at companies such as Cisco, Nvidia, Ramp, Rakuten, and Harvey have rolled it out to developer groups, with Cisco reporting code review times cut by 50% and project timelines compressed from weeks to days. Within business and enterprise accounts, Codex users grew 6x between January and April 2026, underlining OpenAI’s enterprise focus. At the same time, the user base is changing: 20% of weekly users are now knowledge workers rather than coders, and this group is adopting Codex more than three times as fast as developers. Rate limit resets and expanding SDK and Slack integrations suggest OpenAI is preparing Codex as a central AI research and analysis tool for organizations, not only as a coding assistant.
New Sites and Annotations Features Turn Outputs Into Live Workspaces
To serve knowledge workers, OpenAI has added Codex knowledge work features that go beyond traditional code completion. The new Sites capability lets users create interactive websites and internal tools from within Codex, then share them via URL with others in their workspace. Teams can spin up custom dashboards, scenario planners linked to financial models, or launch hubs that keep messaging, milestones, and decisions in one place. OpenAI describes this as moving from a single chatbot to a shared workspace where role-specific context and outputs can become live tools. Partnerships with platforms like Wix, Base44, Replit, Lovable, Figma, and Emergent further extend what these Sites can do. Annotations deepen this shift: users can point Codex at a specific part of a document, slideshow, spreadsheet, or site, then have the agent edit, refine, or reuse that exact section, turning static files into interactive, AI-editable artifacts.
Spreadsheets, PDFs, and Parallel Tasks Redefine AI Productivity
Usage data shows Codex spreadsheet capabilities and document tools are changing how people work day to day. Among knowledge workers, data analysis tasks are growing 110% week over week, with research up 37% and other knowledge artifacts up 36%. Within those artifacts, work on PDFs and spreadsheets has climbed more than 50%, signaling strong demand for AI document automation. Each week, 72% of knowledge worker users now produce reports, memos, contracts, images, audio, video, PDFs, or spreadsheets through Codex. About 50% of users keep more than one task running at once, up from less than one third in mid-April, enabling a single person to inspect datasets, draft scripts, assemble reports, and check applications in parallel. Data labeling, drafting messages, understanding contracts and regulations, and hiring workflows are among the fastest-growing categories as the line between coding and other knowledge work continues to thin.
Role-Specific Plugins and the Future of AI Knowledge Work
Role-specific plugins aimed at sales, data analytics, and other verticals are turning Codex into a platform for specialized workflows, not only a generic assistant. Within organizations, product managers build dashboards, researchers script dataset cleaning, designers ship prototypes, and executives assemble internal tools that reconcile files and generate weekly reports. Case studies already span government data platforms, sales teams, teaching, and personal projects, showing how AI research and analysis tools are moving into mainstream office routines. Personal users, now over 5% of the base and growing more than four times as fast as developers, use Codex for hobbies, education, self-learning, personal finance, and entertainment. As OpenAI’s productivity expansion continues, Codex looks less like a niche coding agent and more like a general-purpose orchestration layer for knowledge work, where spreadsheets, reports, and interactive Sites sit alongside code in a single AI-managed workspace.
