800K (80 Ten-Thousand) Work Prompts Reveal Cross-Functional Task Crossover
OpenAI's Economics and Social Science Research Team released a study titled "Work at the Frontier: How AI is Expanding What People Do at Work." The research analyzed a random sample of over 800K (80 ten-thousand) real work prompts from U.S. ChatGPT Business, Team, and Enterprise accounts to quantify the impact of generative AI on workplace roles and organizational boundaries. Using Large Language Models (LLMs) to automatically categorize user prompts matched against the Occupational Information Network (O*NET) database, researchers excluded routine administrative tasks like meeting scheduling and general email writing, focusing on occupation-specific prompts with specialized attributes.
Prompt Data and Crossover Rates
-
800K+ (80 Ten-Thousand+)
Total number of U.S. ChatGPT enterprise user work prompts analyzed in the sample.
-
43.5%
Share of occupation-specific prompts involving cross-functional tasks after excluding routine admin work.
-
16.8%
Overall proportion of cross-functional tasks across the entire prompt dataset including general administrative work.
The study terms this core phenomenon "Task Crossover". Data shows that generative AI assists employees with core responsibilities while encouraging them to bridge traditional job boundaries and handle complex tasks previously assigned to external specialists or adjacent departments. Across the full sample including routine admin work, 21.8% of prompts belonged to core role tasks, 61.5% to general office tasks, and 16.8% to cross-functional tasks.
Cross-Occupational Rankings and Portable Hard Skills
The study categorized crossover behaviors across various occupational groups. After excluding routine admin tasks, the proportion of cross-functional tasks within occupation-specific prompts revealed distinct variance across departments.
Among all professional capabilities borrowed across boundaries, "Financial Calculations" and "Computer Troubleshooting & Debugging" demonstrated the highest cross-departmental portability. Non-technical and non-finance employees frequently rely on ChatGPT to execute code troubleshooting, data cleaning, and cost analysis that once required IT specialists or dedicated accountants. For instance, marketers use AI to debug frontend web scripts, HR teams write Python scripts for automated data processing, and customer support staff perform financial reconciliations.
Reconstructing Cross-Departmental Collaboration Flows
In traditional corporate structures, cross-departmental tasks rely on slow ticketing and approval processes. Generative AI allows employees to resolve lightweight adjacent tasks independently without burdening specialized teams.
Under traditional ticketing pathways, employees submit tickets, hold cross-departmental meetings, and wait on specialized teams, incurring long turnaround times and high friction.
Under AI-assisted autonomous pathways, employees directly query domain knowledge, debug code, or build models using AI, completing adjacent tasks instantly and cutting delivery delay.
AI-Driven Cross-Functional Workflow
-
Identify Adjacent Pain Points
Employees encounter specific bottlenecks beyond their primary role that obstruct overall progress.
-
Construct Prompts and Context
Input business context and data into ChatGPT, invoking relevant domain knowledge and tools.
-
Execute Reasoning and Verification
Obtain code fixes, legal clause comparisons, or financial calculations and verify the output.
Small Business Dynamics and the Engineering Exception
Task crossover is especially pronounced in small businesses. Lacking dedicated legal, design, or senior finance teams, employees in small businesses rely heavily on AI as a versatile compound tool, handling contract initial reviews, marketing copy visualization, and financial analysis directly to cut cross-departmental coordination overhead.
Engineering departments ranked lowest in crossover frequency at 28%. Engineering prompts centered primarily on deep code writing, system architecture design, and core technical iteration rather than non-technical tasks; conversely, non-technical employees frequently penetrated into engineering domains, leveraging AI for simple script drafting and tool automation.
Organizational Challenges and Job Description Lag
This bottom-up crossover behavior presents new challenges for traditional corporate management. Conventional Job Descriptions (JDs) are static and strictly compartmentalized, whereas employee capabilities augmented by AI are dynamic and cross-functional.
Managers face a dual dilemma: while encouraging task crossover boosts business agility and employee autonomy, non-specialists relying on AI-generated contract drafts or financial models may introduce compliance and risk vulnerabilities if unvetted by domain experts. Redefining job boundaries and approval governance will become a central management imperative.
Tool Usage Boundaries and Study Limitations
The research team emphasized that these findings offer an early perspective on the evolution of workplace task allocation. Although quantitative industry-wide productivity gains cannot be directly inferred, employees are spontaneously reshaping their daily workflows using AI well before formal job descriptions update.