OpenAI has released ChatGPT Work, a product designed to bring its autonomous coding agents to non-engineers, and the company is positioning it as a tool for all white-collar professions. The product, available on the company鈥檚 lowest subscription tier at $20 per month, connects large language models to email, Slack, and productivity apps like Notion and Figma. The goal, according to OpenAI鈥檚 marketing, is to move from answering questions to helping users complete multistep projects with minimal input.
The tool is a modified version of OpenAI鈥檚 Codex coding software, reworked for people who do not use command-line interfaces. Andrew Ambrosino, the lead engineer for OpenAI鈥檚 desktop app, said his team had to adapt the software after internal non-engineering staff found it hostile, with technical readouts meant for code changes. He said the company is building the product to handle the "messy world" of outdated websites and diverse workplace tools, a necessary step to reach a mass audience.
OpenAI鈥檚 internal testing has involved giving the app control over the personal digital accounts of its own engineers. Ambrosino acknowledged a risk that the model could pull private information from direct messages when drafting documents, but said he has not yet experienced that problem. The company sees the product as an autonomous assistant that can parse information across systems like Salesforce and take action, addressing what Akshay Nathan, who leads product engineering, called a "deluge of information" for the average worker.
The commercial stakes are significant. Agents that run for longer periods consume more tokens, making them more profitable for OpenAI per user. The company also faces competition from vertical-specific startups like Harvey for legal work and Clay for sales, which use model-agnostic approaches. Industry analyst Christian Catalini wrote on the a16z blog that if AI labs cannot quickly secure the complementary assets needed to scale, value will accrue elsewhere.
OpenAI is leaning on interface design to ease adoption, with Ambrosino comparing the product鈥檚 simple interface to skeuomorphism, where digital tools mimic physical objects to help users transition. He said the company deliberately added buttons for discoverability, arguing they are still necessary even if users can prompt the model directly. He expects the buttons to eventually disappear as users become more familiar with the technology.
The joint app encompassing Work and Codex has about 20 million users, compared to more than a billion people who prompt ChatGPT online. OpenAI would not say how many use Work specifically. For now, the company is pitching Work for routine, data-intensive tasks, such as generating weekly metrics reports or turning spreadsheets into planning tools.
In practice, the system has shown promise. A TechCrunch reporter used ChatGPT Work to extract a preschool calendar from an email and add it to Google Calendar, a task that would otherwise require repetitive data entry. The reporter declined to grant the app access to a bank account or sensitive materials, but acknowledged that greater trust would likely yield more useful results. An OpenAI engineer described using the tool to analyze a Slack conversation about an engineering problem and receiving a series of charts in response.
More AI news from TechManNews.
