The first large-scale study of how people actually use AI agents has arrived—and it challenges everything we thought we knew about artificial intelligence in the workplace. Drawing on hundreds of millions of anonymised interactions, Harvard Business School and Perplexity researchers have uncovered patterns that should fundamentally reshape how business leaders think about AI adoption.
The headline finding? 57% of all AI agent activity focuses on cognitive work—learning, research, and productivity tasks that augment human capability rather than replace routine functions. This isn’t the “digital concierge” vision of AI handling your calendar whilst you focus on strategy. It’s something far more consequential: AI as a genuine thinking partner.
The Real Story: Agents as Cognitive Amplifiers
The research introduces a rigorous taxonomy of AI agent use cases, and the results overturn popular assumptions. Whilst media narratives have positioned AI agents as sophisticated task-handlers—booking hotels, managing appointments, handling administrative chores—the data tells a different story entirely.
Strategic Reality: The “digital butler” use case represents a minority of actual agent usage. Most interactions involve complex cognitive tasks where agents serve as research partners, analysts, and learning facilitators.
The breakdown of actual usage reveals the cognitive dominance:
| Category | Query Share | Primary Function |
|---|---|---|
| Productivity & Workflow | 36% | Document editing, account management, email, spreadsheets |
| Learning & Research | 21% | Course assistance, research summarisation, information analysis |
| Media & Entertainment | 16% | Social media, video, gaming |
| Shopping & Commerce | 10% | Product search, comparison, purchasing |
| Job & Career | 7% | Professional networking, job applications |
| Travel & Leisure | 7% | Flight search, trip planning, accommodation |
The dominance of productivity and learning categories—accounting for 57% combined—represents a fundamental shift in how AI is being deployed in practice.
What’s Really Happening: The Evolution from Novelty to Necessity
One of the study’s most revealing findings concerns how usage patterns evolve over time. New users typically test AI agents with low-stakes queries—travel recommendations, entertainment suggestions, general information requests. But something interesting happens as familiarity grows.
Critical Context: Users who engage with productivity or learning tasks early become significantly more likely to develop into long-term active users. The agent becomes indispensable not through convenience, but through cognitive partnership.
The researchers draw a compelling parallel with personal computers. Early PCs were often marketed for recipes, games, and household management. They became transformative through spreadsheets and word processing—tools that augmented professional capability. AI agents appear to follow the same trajectory.
Transition patterns show clear gravitational pull toward cognitive tasks:
- Productivity and learning categories demonstrate the highest user retention rates
- When users switch between topic categories, they most commonly transition into productivity, learning, or career-related tasks
- Over time, query shares shift measurably from entertainment and travel toward productivity and learning
This isn’t users discovering how to automate their lives. It’s users discovering how to think more effectively.
The Adoption Gap: Who’s Actually Using AI Agents?
The research identifies substantial variation in who adopts AI agents and how intensively they use them. Understanding these patterns helps business leaders anticipate where AI adoption will accelerate—and where targeted investment may be required.
By economic development: Strong positive correlations exist between agent adoption and both GDP per capita (r = 0.85) and average years of education (r = 0.75). Economically developed nations with higher educational attainment demonstrate significantly higher adoption rates.
By occupation: Six core occupational clusters drive 70% of all agent activity:
| Occupation Cluster | Adopter Share | Query Share | Key Characteristic |
|---|---|---|---|
| Digital Technology | 28% | 30% | Highest volume |
| Students | 12% | 16% | Highest learning focus |
| Financial Services | 10% | 9% | Productivity-focused |
| Marketing & Sales | 9% | 11% | Highest “stickiness” |
| Management & Entrepreneurship | 8% | 9% | Cross-functional usage |
| Education | 6% | 6% | Learning-focused |
Hidden Cost: The adoption gap between knowledge-intensive and physical sectors suggests a potential productivity divergence. Organisations in less digitally-native industries may need deliberate intervention to capture AI agent benefits.
The stickiness metric matters enormously. Marketing, sales, and entrepreneurship professionals show the highest ratio of usage intensity to adoption—meaning once they start using agents, they use them heavily. This pattern indicates where genuine behavioural change is occurring versus mere experimentation.
The Human Factor: Context Shapes Everything
Perhaps the most strategically valuable finding concerns how usage context shapes agent deployment. The research categorises interactions into personal, professional, and educational contexts:
- Personal use: 55% of queries
- Professional use: 30% of queries
- Educational use: 16% of queries (note: increasing over time)
But the raw percentages obscure the more interesting pattern: context dramatically reshapes what people ask agents to do.
| Context | Dominant Activities | Key Insight |
|---|---|---|
| Personal | Productivity (34%), Media (28%) | Shopping, social media, entertainment |
| Professional | Productivity (47%), Career (18%) | Document editing, networking, job-related tasks |
| Educational | Learning (89%) | Course assistance, research, exercise completion |
Success Factor: AI agents aren’t one-size-fits-all tools. They become specialised instruments shaped by their deployment context. A “learning partner” in education becomes a “workflow amplifier” in professional settings and an “information concierge” for personal use.
For business leaders, this suggests that AI agent value depends critically on how deployment is framed and supported. Generic “AI assistant” implementations may underperform compared to context-specific positioning.
Strategic Implications: Preparing for the Agentic Shift
The research carries immediate strategic implications for organisations considering AI agent adoption:
For early-stage adopters
Priority actions:
- Start with cognitive tasks, not automation. The highest retention and value comes from research, analysis, and productivity applications—not calendar management
- Target knowledge workers first. Marketing, sales, entrepreneurship, and finance professionals show the strongest adoption-to-intensity ratios
- Frame agents as thinking partners. Position AI as augmentation for complex work, not replacement for simple tasks
For organisations with existing AI investments
Priority actions:
- Audit current usage patterns. Are your teams using AI for cognitive tasks or administrative convenience? The former predicts sustained value
- Bridge the adoption gap. If physical or service-oriented roles show low adoption, consider targeted training and use-case development
- Track evolution metrics. Monitor whether users transition toward productivity and learning tasks over time—this indicates genuine integration
For strategic planners
Priority actions:
- Anticipate workforce implications. The cognitive partnership model suggests AI will transform how knowledge workers operate rather than whether they’re needed
- Plan for task concentration. The top 10 of 90 identified tasks represent 55% of all queries—focus integration efforts on high-frequency applications
- Consider competitive dynamics. Organisations that successfully integrate AI as cognitive amplifiers may develop sustainable productivity advantages
SME Advantage: Smaller organisations can move faster on AI agent adoption than enterprises with complex governance requirements. The research shows that early, intensive adopters gain the greatest benefits—suggesting first-mover advantage exists in this space.
Hidden Challenges: What the Data Doesn’t Show
Whilst the research provides invaluable insights, business leaders should consider several limitations and non-obvious challenges:
1. Early adopter bias The study covers July–October 2025, capturing primarily early adopters of a new product. These users likely skew more tech-savvy and experimentally-minded than the general business population. Adoption patterns may differ as AI agents reach mainstream audiences.
2. Performance versus intent The research classifies what users ask agents to do, not how effectively agents complete those tasks. High query volume in productivity categories doesn’t necessarily indicate successful task completion. Organisations should track outcome metrics, not just usage metrics.
3. The training investment Whilst agents appear to deliver greatest value for cognitive tasks, these applications also require the most contextual understanding. Generic implementations may underperform without investment in customisation, prompt engineering, and workflow integration.
Warning: ⚠️ The “stickiness” of cognitive applications creates dependency risks. As agents become integrated into thinking processes, organisations should ensure they maintain human capability to perform critical analysis independently.
4. Security and governance gaps The research notes that enterprise users were excluded from the analysis. Professional deployments of AI agents raise data protection, intellectual property, and governance considerations that personal usage does not. Organisations need robust AI policies before scaling agent adoption.
The Strategic Takeaway: Cognitive Partnership Is the Future
This research fundamentally reframes the AI agent conversation. The dominant narrative—AI as sophisticated task automation—understates both the opportunity and the transformation underway.
The core insight: AI agents are most valuable not as digital assistants but as cognitive partners. They amplify human capability for complex, knowledge-intensive work. The 57% of usage devoted to productivity and learning represents a new model of human-AI collaboration that extends rather than replaces human intelligence.
Three success factors for business leaders:
- Deploy for thinking, not tasking. Position AI agents as research partners, analysis tools, and learning facilitators—not administrative assistants
- Target high-retention use cases. Focus on productivity, learning, and career applications where users develop sustained engagement
- Invest in context-specific implementation. Generic agent deployment underperforms; tailor positioning, training, and workflows to specific professional contexts
Take Action: 📥 Evaluate your current AI strategy against these findings. Are you positioning AI for cognitive augmentation or task automation? The research suggests the former delivers substantially greater long-term value.
Next steps for your organisation:
- Audit current AI usage patterns against the productivity/learning benchmark
- Identify high-potential cognitive applications within your workflow
- Develop context-specific positioning for professional AI deployment
- Establish metrics for tracking cognitive partnership value (not just automation efficiency)
- Review AI governance frameworks for agent-specific considerations
Source: Yang, J., Yonack, N., Zyskowski, K., Yarats, D., Ho, J., & Ma, J. (2025). The Adoption and Usage of AI Agents: Early Evidence from Perplexity. Harvard Business School & Perplexity. How People Use AI Agents. Analysis and strategic commentary provided by Resultsense.
Preparing your organisation for AI agent adoption?