# OpenAI's Internal Experiment Is a Map for the Rest of Us
> OpenAI's dramatic internal adoption of its own agentic AI tools shows that true productivity gains demand a fundamental redesign of human workflows and roles.
**Source:** The Aftermath (https://theaftermath.news/posts/openai-s-internal-experiment-is-a-map-for-the-rest-of-us/)  
**Published:** 2026-10-22
**Original source:** https://ide.mit.edu/news-articles/research-digest/inside-openais-shift-to-ai-agents/
**Topics:** AI in the Workplace, Research, Agentic AI, OpenAI

---

OpenAI's rapid internal adoption of its own agentic AI tools reveals a difficult truth: realizing the technology's promise requires not just new software, but a complete rethinking of how we structure our work and define our roles. Imagine a recruiter at the company's San Francisco headquarters. For years, their job was a series of discrete tasks: scour profiles, write introduction emails, manage calendars. Today, their screen is more likely to show a simple text prompt where they instruct an AI agent to perform all those steps in a single, coordinated effort based on a high-level goal. The job is no longer about doing the work, but about directing it.

This fundamental change from human-led tasks to human-directed automation is the focus of <a href="https://ide.mit.edu/news-articles/research-digest/inside-openais-shift-to-ai-agents/" rel="nofollow noopener">a new working paper from MIT</a>. The research, co-authored by MIT IDE researcher David Holtz and several OpenAI employees, chronicles a dramatic internal transformation. The use of agentic tools like OpenAI Codex surged among employees from composing less than 10% of their AI-assisted output in August 2025 to nearly 100% by June 2026. The adoption wasn't limited to software engineers; it spread virally through departments like legal, finance, and recruiting, demonstrating the tool's general-purpose utility.

## What does it mean to redesign a job?

For most companies, adopting new software means fitting it into existing processes. The MIT study on OpenAI suggests this approach is a trap. The productivity gains observed within OpenAI didn't come from simply giving everyone a faster typewriter; they came from dismantling and rebuilding entire workflows around the AI's capabilities. A lawyer doesn't just use an agent to find a clause; they use it to analyze and summarize an entire contract based on a set of legal principles. A financial analyst might ask an agent to build and test several forecasting models at once, rather than manually constructing one. OpenAI had a cultural advantage in this transition, as its entire mission is oriented around exploring AI's limits. Legacy companies, with more rigid professional identities and established processes, will find this organizational redesign a far greater challenge than the technical implementation.

## How does this change the nature of expertise?

The shift toward agentic AI reframes the value of human expertise. The study implies that an employee's worth is less about their ability to perform repetitive, skilled tasks and more about their capacity for high-level reasoning, judgment, and goal-setting. The expert becomes a director, an editor, and a final arbiter of quality for work executed by the AI. This suggests that the most critical skill in a world saturated with capable AI agents is not the mastery of a specific craft, but the ability to translate nuanced human intent into a clear directive a machine can act upon. The map OpenAI has drawn is not a simple one. It asks every organization, and every worker within it, to reconsider the routes they have always taken.

## FAQ

### What is an agentic AI tool?
An agentic AI tool is a system that can take a high-level goal, break it down into steps, and execute those steps to achieve the desired outcome, often using other software tools along the way. This differs from a simple conversational chatbot, which primarily responds to direct commands.

### Why did OpenAI employees adopt Codex so quickly?
The study suggests the rapid adoption was driven by the tool's significant utility in automating complex, time-consuming tasks across a wide variety of roles, not just coding. OpenAI's experimental, AI-focused culture also created an environment where employees were encouraged to integrate these systems into their daily work.

### What does this study mean for other companies?
The research indicates that for other companies to see the same level of productivity gains from agentic AI, they must do more than just purchase the software. They will need to fundamentally redesign workflows and even redefine job roles to shift from human execution to human direction of AI agents.

<p class="related-inline">Related: <a href="/posts/the-ai-that-wears-a-mask/">The AI That Wears a Mask</a></p>