OpenAI Deploys AI Agents Beyond Coding as It Chases Mass Workplace Adoption

OpenAI is gambling that giving AI models deep access to your digital life—email, calendar, spreadsheets, cloud storage—will unlock a new era of workplace productivity. The company launched ChatGPT Work, a $20/month tier designed to let white-collar workers delegate complex, multi-step tasks to AI agents rather than rely on simple Q&A.

The Internal Test Case#

Andrew Ambrosino, lead engineer on OpenAI's desktop app, has volunteered as the experiment's subject. His AI agent now oversees his inbox, Slack account, Notion workspace, Figma designs, and phone. The personal risk exists—a model could theoretically leak sensitive information it shouldn't share—but Ambrosino accepts the trade-off in pursuit of proof of concept.

"If I'm asking it to write a document, is there a possibility that it's going to pull from a private DM on that subject and not know that it's not supposed to share some info? Yes," he told TechCrunch. "I'll do it for the job."

A Long Road from Code to Spreadsheets#

The shift from narrow (software development) to broad (any office task) is where the real business opportunity lies. Internally, nearly every OpenAI employee uses their coding agent, yet fewer than 1% of individual subscribers adopt it—a gap the company sees as both a problem and a massive market.

Making agents work for non-engineers demands more hand-holding. Codex, the underlying tool, started out hostile to non-technical staff, asking questions about code and displaying technical outputs meaningless to accountants or marketers. The team spent months making it general-purpose and intuitive enough for ordinary workers.

"[An agentic product] is going to be something that plays with the messy world of your life and your tools and websites that were built in 1995 and never updated," Ambrosino explained. Designing around this chaos—legacy systems, sprawling workflows, a billion possible use cases—is central to the challenge.

The Product Puzzle: Buttons Versus Prompts#

OpenAI engineers are caught between two design philosophies. Power users often argue that any feature should be accessible simply by asking the model directly. Beginners, however, need discoverable buttons and visual cues. The team defaults to the latter for now, comparing their approach to skeuomorphism—making digital tools resemble the physical objects they replaced.