From Asking to Doing: What Happens When AI Starts Acting on Your Behalf
For most people, AI still means a chatbot: you type a question, it types back an answer, and you decide what happens next. A new wave of agentic AI tools, however, works differently. They don’t just answer questions, they complete entire tasks on their own. Prasanna Tambe, faculty co-director of Wharton Human-AI Research, co-authored a new study that looks at what happens when this agentic AI enters the workplace. Using data from OpenAI’s coding platform Codex, the research gives one of the first large-scale looks at how people actually work once AI stops just talking and starts doing.
Key Takeaways
- Adoption is real, but it’s wildly uneven.
Fewer than 1% of everyday consumers use agentic AI tools like Codex. Inside companies, that number jumps to 17%. Inside OpenAI itself, it’s nearly everyone. The gap shows that adoption depends less on the technology and more on how much an organization is set up to support it. - People are handing off bigger jobs, not just quick questions.
Since the start of 2026, the share of users assigning AI a task that would take a person a full workday has grown nearly tenfold. That’s a meaningful shift from “help me draft this” to “just do this.” - The work has moved well past coding.
Codex was built for developers, but its heaviest users now rely on it for drafting documents, analyzing data, and coordinating projects, proof that agentic AI’s reach extends far beyond software teams. - Some employees now manage a small team of AI agents.
A growing group of power users run three or more AI agents at once, shifting their own role from doing the work to reviewing and directing it. - This isn’t just a junior-employee tool.
Adoption climbs at every seniority level, from new hires to executives, so agentic AI isn’t simply automating entry-level tasks, it’s reshaping how work gets done across the board.
Real-World Application
Short on time? Here’s the takeaway:
The organizations getting the most out of agentic AI aren’t just the ones that bought the tool, they’re the ones that rebuilt how work gets assigned and reviewed around it. The researchers tracked usage across three groups — individual consumers, employees at outside organizations, and OpenAI’s own staff — and that last group shows what adoption looks like once the usual barriers, like cost and unfamiliarity, are removed: agentic tools now account for nearly all of OpenAI employees’ AI-driven output, compared to about 63% at other companies and just 17% among individual users. The clearest sign of this shift is what the researchers call “systematization,” where employees build reusable instructions, or “skills,” that let a workflow run automatically instead of being set up from scratch each time. In just a few months, the share of active users relying on at least one of these reusable skills rose from about 5% to nearly 27%, with the heaviest use among employees most embedded in company workflows.
This content was created with the assistance of generative AI. All AI-generated materials are reviewed and edited by the Wharton AI & Analytics Initiative to ensure accuracy, clarity, and alignment with our standards.
About Wharton AI & Analytics Insights
Wharton AI & Analytics Insights is a thought leadership series from the Wharton AI & Analytics Initiative. Featuring short-form videos and curated digital content, the series highlights cutting-edge faculty research and real-world business applications in artificial intelligence and analytics. Designed for corporate partners, alumni, and industry professionals, the series brings Wharton expertise to the forefront of today’s most dynamic technologies.