OpenAI expands ChatGPT distribution and monetization

A data agent, not just a data answer box

OpenAI’s latest ChatGPT Work update is less about adding another sidebar assistant than about putting ChatGPT inside the operational layer of a company. On September 10, 2026, OpenAI introduced the Data agent in ChatGPT Work, describing it as a way for employees to turn company data into answers, interactive dashboards and follow-up actions by asking in natural language [1]. The immediate promise is familiar: fewer queues for reports, less waiting on data teams, and more self-service analysis. The more important shift is architectural. ChatGPT is being positioned as a work execution environment that connects to enterprise systems, interprets business context and produces artifacts that other people can use.

The Data agent connects to approved data sources such as Amazon Redshift, Datadog, Google BigQuery, ClickHouse, Databricks, MongoDB and Snowflake, and it can also bring files and documents from Google Drive and SharePoint into an analysis [1]. OpenAI says the agent can use business terms, metric definitions, custom calculations and data relationships drawn from semantic layers and trusted sources such as Databricks Genie Ontology, dbt, GitHub, Snowflake Horizon and BI dashboards [1]. That framing matters because most corporate data failures are not caused by an inability to draw a chart; they come from mismatched definitions, permissions and context. OpenAI is trying to make ChatGPT useful where enterprise analytics actually breaks: between the warehouse, the dashboard, the spreadsheet and the meeting.

Distribution through plugins and familiar tools

The product is also a distribution play. OpenAI is not asking companies to abandon their data stack. Instead, it is placing ChatGPT Work across the stack through plugins and existing permissions. The Data agent can build and interact with dashboards in Omni, Oracle BI, Power BI, Sigma, Tableau and ThoughtSpot, and it can share findings through Slack or email after a user approves the action [1]. In the ChatGPT Business release notes, OpenAI describes the Data plugin as available in ChatGPT Work and Codex, with administrators able to manage installation policy, enable the data-source plugins and apps the team needs, and apply connected-account permissions and workspace access controls [2].

That is a significant change in the way ChatGPT reaches business users. Instead of distribution depending only on people visiting a chat window, the product is moving into workflows where employees already inspect revenue, support tickets, product usage, spending and forecasts. OpenAI’s own data-team page says ChatGPT Work is meant to help teams answer complex business questions, build interactive dashboards and act on what they find without specialized skills [1]. If it works as described, the spreadsheet is no longer the final destination for analysis. It becomes one of several work surfaces the agent can read, transform and update.

OpenAI’s September 10 release notes also make clear that the Data plugin sits inside a broader Work and Codex environment, not as a one-off analytics feature [2]. Deep Research in Work and Codex can use web, files and supported connected apps that users are authorized to access, while ChatGPT Voice can now use GPT-5.6 or GPT-6 Astra when it needs to search or reason through harder questions [2]. The result is a product suite where data analysis, research, voice coordination, coding and deliverable creation are converging under the same ChatGPT interface.

Monetization is increasingly tied to work done

The monetization signal is just as important as the feature list. OpenAI’s current Business, Enterprise and Edu rate card says ChatGPT Work and Codex usage is priced by token consumption, calculated in credits per 1 million input tokens, cached input tokens and output tokens [3]. The same document says the rate card applies to supported ChatGPT Work and Codex activity including local tasks, cloud tasks, automations, code review, auto review and delegated workers [3]. In other words, OpenAI is building a revenue model that maps more closely to work performed than to simple chat messages.

The rate card lists GPT-6 Astra at 250 credits per 1 million input tokens, 25 credits per 1 million cached input tokens and 1,250 credits per 1 million output tokens for Work and Codex activity [3]. It also says Codex, ChatGPT Work, ChatGPT for Excel and Workspace Agents draw from the same agentic usage and credit pool when those features are available on a plan [3]. For businesses, this creates a new budgeting problem: the more successful the agent becomes at taking on complex, output-heavy tasks, the more carefully teams will have to govern consumption. For OpenAI, that same dynamic is the point. A task that used to be a free-form conversation can become a metered workflow with files, apps, dashboards, automations and approvals.

OpenAI is reinforcing that model with incentives. The September 10 Business release notes say eligible ChatGPT Business users can earn a 50-dollar workspace credit by trying participating partner plugins such as Stripe, Intuit QuickBooks, HubSpot, Gusto, Dropbox and Canva, with a two-redemption, 100-dollar workspace cap and credits expiring after 30 days [2]. That looks like classic platform seeding: encourage administrators and users to connect more apps, prove the agent can complete useful workflows, then let usage expand into the shared credit pool.

Astra raises the ceiling, and the stakes

The Data agent arrives alongside OpenAI’s broader push around GPT-6 Astra, which the company describes as its most capable model for work and says is now available in ChatGPT Work, Codex and the API [4]. OpenAI says Astra is designed for computer use, browsing, professional work, software engineering, cybersecurity and science, and that it can work through everyday applications even when those applications do not have an API [4]. That is central to the ChatGPT Work thesis: the agent is valuable not merely because it can write a SQL query, but because it can move through the messy interfaces and documents that surround the query.

OpenAI also published an explicit business-model argument this week. In a September 8 post, the company said its products reach more than 1 billion weekly active users and 2.5 million businesses, and argued that consumer familiarity, enterprise deployments and developer applications reinforce one another [5]. The same post says OpenAI’s model lets it earn revenue as use grows, with free access supported by advertising and subscriptions and usage-based offerings allowing customers to spend more as they find more value [5]. That is the commercial logic behind the Data agent: make ChatGPT a default entry point for work, then monetize the volume and complexity of the work it performs.

The governance question enterprises cannot skip

The harder question is how much authority companies should give agents that can touch data, files and communications. OpenAI says enterprise administrators choose which data connections are available and which roles can use them, and that queries enforce the connected account’s existing permissions, including table, row and column restrictions [1]. The company also says Astra includes new enterprise admin controls to restrict access to approved websites and desktop applications, manage uploads and downloads, and require confirmation before consequential actions [4].

Those controls will be tested. On September 8, The Register reported on Check Point Research findings that a secret channel through ChatGPT’s internal JFrog Artifactory instance allowed one account to send hidden tasks, such as retrieving Gmail data, to a ChatGPT session under another account; the report said OpenAI decommissioned the specific internal instance involved [6]. The incident is not the Data agent launch, but it is directly relevant to the same enterprise promise: agents become more useful as they gain access to connected tools, and more risky for the same reason.

The bottom line

OpenAI’s Data agent is not simply a convenience feature for analysts. It is part of a larger effort to distribute ChatGPT through plugins, desktop and web workflows, business apps, semantic data layers and shared credit pools. The launch gives OpenAI a sharper workplace narrative: ChatGPT can answer the question, build the dashboard, explain the evidence, identify stakeholders and push the next step through approved tools [1]. But the company has not yet provided a public, Data-agent-specific price, independent enterprise adoption count or measured time-savings benchmark for the new product. For customers, the next phase is practical testing: how often the agent is right, how clearly it shows its work, how safely it handles permissions, and whether the bill tracks value rather than curiosity. For spreadsheets, the outlook is less certain. They may soon request hazard pay.

Sources from the last 72 hours

1. [1]ChatGPT Business - Release NotesSep 10, 2026, 3:33 PM UTC
2. [2]The Work Now Within ReachSep 8, 2026, 12:00 AM UTC
3. [3]Now everyone can put data to workSep 10, 2026, 12:00 AM UTC
4. [4]ChatGPT Rate Card (Business, Enterprise/Edu credit-based pricing)Sep 9, 2026, 10:33 PM UTC
5. [5]GPT-6 Astra: The next generation in intelligence for workSep 9, 2026, 12:00 AM UTC
6. [6]OpenAI's Artifactory opened covert data-stealing channel alongside Hugging Face attackSep 8, 2026, 10:12 PM UTC

AI-generated article based on recent web research, then preserved as a dated editorial snapshot.