How AI Is Reshaping Job Roles: OpenAI Study Details "Task Crossover" in the Workplace

Understanding "Task Crossover": How Workers Are Crossing Job Boundaries

For years, studies evaluating artificial intelligence in the workplace focused on a narrow question: which existing human tasks can an AI model execute or replace? However, new empirical data published by OpenAI in its report, Work at the Frontier: How AI is Expanding What People Do at Work, indicates that generative AI is fundamentally changing who performs which tasks.

By analyzing more than 800,000 anonymized, work-related ChatGPT user messages from the United States, researchers observed that 16.8% of all work-related interactions—and 43.5% of non-generic, occupation-specific prompts—involve tasks historically assigned to a completely different profession. OpenAI defines this structural shift as task crossover.

Rather than simply speeding up repetitive duties within a fixed job description, workers are leveraging large language models to "borrow" capabilities from neighboring disciplines. A small business owner drafts legal contracts and performs basic accounting; a marketing specialist debugs front-end code; a customer support representative generates technical documentation. The study suggests that AI is quietly reorganizing the daily content of jobs long before official job titles or organizational charts catch up.

Which Professions Borrow Tasks the Most?

To measure task crossover accurately, OpenAI filtered out generic workplace activities—such as drafting routine emails, summarizing text, or scheduling meetings—which are shared broadly across almost all roles. Among the remaining specialized prompts, the proportion of tasks originating outside the user's core field varied dramatically by occupation:

• Customer Experience: 77% of specialized prompts involved tasks from other fields.
• Designers: 75% of specialized prompts crossed boundary lines.
• Human Resources: 69% involved borrowed occupational tasks.
• Legal Workers: 56% drew on external task domains.
• Marketers: 53% involved outside activities.

The report highlights distinct directional flows between disciplines. Some occupations act primarily as importers of external tasks, while others act as key exporters across the enterprise.

Designers represent a clear importer profile: 35.2% of all messages from designers involved tasks usually handled by other roles (such as copywriting or front-end logic), whereas design tasks accounted for only 1.7% of prompts originating from non-designers. Conversely, Engineering functions heavily as an exporter. Only 18.5% of prompts from engineers involved outside fields, but technical and software troubleshooting tasks made up 7.4% of all prompts sent by workers outside engineering.

Marketing stands out by moving heavily in both directions: marketers devoted 24.3% of their prompts to outside fields, while marketing tasks accounted for 8.9% of prompts from non-marketing staff—the highest outward spread in the entire dataset. Across almost all sectors, basic financial calculations, technical troubleshooting, and marketing copy generation emerged as the most widely borrowed task categories.

Overview: Task Crossover Across Occupations and EU Enterprise Factors

The table below summarizes task crossover metrics alongside operational considerations for European companies under current EU regulations.

SMEs vs. Large Enterprises: Where AI Reshapes Roles Faster

The study also demonstrates that organizational size plays a significant role in task crossover. In smaller organizations (workspaces with 2–5 seats), the average share of outside-occupation tasks reached 18.9%, compared to 16.3% in large enterprises with over 100 seats.

In a small business, specialized internal departments rarely exist. When a need arises, the employee closest to the problem uses generative AI as a generalist force multiplier rather than submitting a ticket to an internal team or hiring external consultants. In contrast, large enterprises maintain formal handoff processes, specialized service desks, and established administrative boundaries.

For European SMEs, this dynamic presents a compelling economic argument. A subscription to ChatGPT Team costs approximately $25 to $30 per user per month (roughly €23 to €28 / month depending on billing cycles), while ChatGPT Enterprise is available via custom commercial agreements. By enabling staff to expand their functional scope, smaller companies can achieve operational agility that previously required headcount expansion.

The European Angle: GDPR, Works Councils, and the EU AI Act

While task crossover drives productivity, implementing it within European organizations requires navigating strict legal and institutional frameworks that do not apply in the United States.

First, GDPR and Data Security become major focal points when employees take on tasks outside their traditional core expertise. When a marketing practitioner uses ChatGPT to analyze customer spreadsheets or debug a database query, unvetted prompts can expose personally identifiable information (PII) or confidential source code to model training pipelines. European deployments must ensure that enterprise contracts explicitly include Data Processing Agreements (DPAs), regional EU data storage hosting, and opt-outs from model retraining.

Second, the EU AI Act introduces strict rules regarding AI system deployment in the workplace. While general-purpose productivity tools like ChatGPT, Claude, or local LLMs are regulated primarily at the provider level, their application in HR, performance evaluation, or candidate scoring falls into the high-risk category. If an HR staff member uses an AI tool to summarize candidate portfolios or evaluate employee productivity, the company acts as a deployer subject to human oversight, transparency, and risk management duties under Article 26 of the Act.

Third, European labor practices—particularly in countries like Germany, Austria, and France—involve Works Councils (Betriebsräte) and formal labor agreements. Significant shifts in job definitions, cross-functional workload transfers, and automated performance tracking typically require consultation with worker representatives. Framing AI adoption around task augmentation rather than job substitution is essential for smooth integration in European workplaces.

Editorial Insights: Task Crossover in Our AI Arena Workflow

At ai-jarvis.eu, task crossover is not just a statistical finding from a research paper—it reflects our daily operational reality. In our test lab, where we run the AI Arena benchmark testbed (equipped with local hardware such as Nvidia RTX GPUs alongside Ollama and commercial APIs), cross-functional AI usage is standard practice.

Our editorial team regularly uses local models (such as Mistral NeMo and Llama 3.1) to write Python scripts for data validation or format complex JSON feeds. Conversely, our technical contributors utilize specialized models to refine publication language, draft legal disclaimers, or design promotional graphics. In our internal measurements, deploying open-weights models locally allows non-technical staff to execute technical workflows securely on-premise without exposing sensitive media assets or draft data to external cloud servers.

OpenAI's latest study confirms what many forward-thinking teams are experiencing: generative AI is less about replacing whole jobs and more about redefining the boundaries of human capability across every department.