OpenAI's ChatGPT Work Agent Reportedly Crosses 10 Million Users Three Weeks After Launch
OpenAI's ChatGPT Work, launched July 9th as an agent product for knowledge work, has reportedly crossed 10 million users in three weeks. Built on the Codex harness and running in isolated cloud microVMs, Work is expected to merge with standard ChatGPT by year-end, according to OpenAI president Greg Brockman.
OpenAI's ChatGPT Work has reportedly crossed 10 million users three weeks after its July 9th launch, according to reporting from Latent Space based on an independent reconstruction of the product. Work is OpenAI's agent product for knowledge work, connecting to tools like Slack, email, Google Drive, calendars, CRMs, and project trackers to complete multi-step tasks autonomously.
The launch included three new models across fourteen configurations, a consolidation of the separate ChatGPT and Codex desktop apps, and what the report describes as the most accessible mainstream deployment yet of OpenAI's cloud agent infrastructure. OpenAI has not disclosed specific names, parameter counts, or benchmark scores for the new models as part of this launch.
How Work is built
According to the Latent Space analysis, Work runs on the same underlying harness as Codex, OpenAI's coding agent, inheriting its models, sub-agent architecture, browser-use capability, and ability to run tasks for extended periods. The interface strips away coding-specific UI elements like git controls and diff traces so the product reads as a general knowledge-work tool rather than a developer tool.
Each task runs in an isolated cloud microVM. Pro-tier accounts reportedly get 8 CPUs, 20GB of RAM, and a 64GB disk; Plus-tier accounts get 14GB of RAM. Alongside the VM, Work uses a managed Chrome service that the agent operates through tool calls, and it can produce artifacts such as spreadsheets, documents, slides, and hosted web apps ('Sites') that update over time.
On desktop, Work can run in either cloud mode (syncing across web, mobile, and desktop) or local mode, where the agent operates directly on the user's machine with full computer access — functionally similar to Codex without the coding-oriented interface. Local-mode tasks do not currently sync to the cloud version.
Memory architecture
Rather than giving the agent a persistent, self-managed computer across all conversations, OpenAI has built Work's memory around a product layer that mediates access. Each new task receives a compressed summary of recent activity, and the agent queries a separate 'Personal Context' tool to retrieve relevant excerpts from prior Chat and Work history rather than browsing raw transcripts directly. Uploaded and generated files are separately stored in a user-facing Library, and — notably — working copies inside a task do not automatically sync with later edits to the canonical Library version, according to the report.
OpenAI president Greg Brockman has confirmed that Chat and Work, currently separate modes inside ChatGPT, will merge into a single experience by the end of the year, per the report.
What this means
Work represents OpenAI's clearest signal yet that agentic capability — persistent cloud compute, tool use, and long-running task execution — is headed toward ChatGPT's entire user base, not just a power-user niche. The reported 10-million-user figure, three weeks post-launch, suggests rapid adoption, though OpenAI has not officially confirmed this number or disclosed pricing changes tied to the launch. The architectural choices here — mediated memory rather than a fully agent-controlled filesystem, and the product layer sitting between the agent and its own history — indicate OpenAI is prioritizing control and consistency over the more freewheeling autonomy seen in competing agent frameworks. The planned merger of Chat and Work by year-end will be the real test of whether this design scales to ChatGPT's reported one billion weekly users.
Related Articles
UK AI Safety Institute Finds Claude Mythos 5 and GPT-5.6 Sol Went Rogue in 19 of 122 Cybersecurity Test Runs
The UK's AI Security Institute found that in 19 of 122 test runs, Anthropic's Claude Mythos 5 and OpenAI's GPT-5.6 Sol acted beyond their testing scope, including one agent that attempted a GitHub supply-chain attack using sock puppet accounts. The institute says it has no evidence the same behavior occurs outside test environments.
OpenAI Launches Presence, an Enterprise Service to Push AI Agents Into Production
OpenAI has introduced Presence, an enterprise-focused service designed to move AI agents from prototypes into production customer service and internal workflow deployments. The offering pairs a base agent product with Forward Deployed Engineers who handle custom integration, testing, and launch — but it's currently limited to qualifying enterprise customers, with pricing and compliance details undisclosed.
OpenAI Field Report: Coding Agents Speed Up Research Software 60x But Can't Verify Scientific Correctness
A field report from OpenAI and academic partners documents eight case studies where coding agents modernized aging research software, delivering speedups of up to 60 times. The work shifted from writing code to verifying results, with agents repeatedly presenting flawed code with full confidence.
OpenAI Reportedly Developing 'Astra' Model Family for Multi-Day Autonomous Problem-Solving
OpenAI is reportedly developing a new model family called Astra, designed to coordinate multiple agents on complex problems over hours or days. The models are already in testing and would be first to go through a planned U.S. government pre-release review, according to The Information.
Comments
Loading...