New research from Perplexity offers one of the clearest pictures yet of how AI agents are being used inside companies. After months of speculation about whether “agentic AI” would move beyond demos and promotional talk, the data shows these tools are already taking on complex tasks for some of the highest-value workers in the enterprise.
Perplexity analysed hundreds of millions of interactions through its Comet browser and assistant. The findings show that AI agents are no longer just conversational helpers. They are being used to carry out multi-step actions, gather information, and support decision-making with limited human oversight.
Who’s using AI agents — and why it matters
The study highlights a wide gap in adoption. Workers in countries with higher income levels and stronger education systems are far more likely to rely on these tools. Inside companies, the sharpest uptake comes from digital-focused roles. Software engineers, academics, financial analysts, marketers, and founders represent more than 70 percent of all agent users.
These early adopters aren’t just experimenting. Perplexity’s “power users” submit nine times more agent requests than the average user, suggesting that once AI agents fit into someone’s workflow, they tend to stay there.
Cognitive work, not admin chores
A common assumption is that agents will handle low-value tasks like scheduling or email sorting. Perplexity’s data points in a different direction: 57 percent of activity centers on cognitive work.
The largest category, “Productivity and Workflow,” makes up 36 percent of all queries. Another 21 percent falls under “Learning and Research.” Examples in the report include procurement teams using agents to scan case studies before vendor calls, and finance workers delegating the sorting and filtering of stock data. In each case, the agent handles the groundwork so humans can focus on decisions.
Perplexity describes its agents as systems that move through “thinking, acting, and observing” cycles — a structure that makes them better partners for deep work rather than simple administrative support.
How usage evolves over time
One of the clearest insights is how stickiness builds. New users often start with light tasks — movie suggestions or simple questions. But as they gain confidence, they shift toward heavier work like debugging code or reviewing investment reports. Once that shift happens, they rarely go back.
Productivity and workflow tasks show the strongest retention. For companies planning pilot programs, this means employees may need time to move from exploration to true delegation.
Where agents are working
Perplexity also examined the platforms where agents operate. Google Docs is a major hub for editing and document handling, while LinkedIn dominates professional-networking tasks. Learning and research activities spread across sites like Coursera and academic repositories.
Because agents interact directly with these environments via browser control or APIs, the security stakes are higher. When an employee asks an agent to summarise customer material, the agent isn’t just reading text — it’s acting inside systems that may contain sensitive data.
High-traffic platforms also point to where companies can make targeted improvements. For example, LinkedIn accounts for 96 percent of professional-networking agent queries, suggesting that a dedicated governance or API strategy for that platform could have an immediate impact.
Ryan Daws
What businesses should do next
Perplexity’s findings suggest the era of speculative talk about agentic AI is over. These tools are already performing multi-step actions and supporting some of the most skilled workers in the organisation. With the market expected to grow from $8 billion in 2025 to nearly $200 billion by 2034, leaders face three clear priorities:
- Audit high-value workflows. Engineers, analysts, and strategists are already using agents. Standardising those workflows could deliver faster gains.
- Prepare for human-AI collaboration. Users tend to break tasks into smaller pieces and delegate subtasks. Training employees to work effectively alongside agents will matter more than ever.
- Strengthen security and governance. As agents interact with cloud tools, code repositories, and internal documents, companies need updated controls to manage new risks.
