New adoption data from Perplexity analyzing hundreds of millions of interactions with its Comet browser and assistant provides the first large-scale field study of general-purpose AI agents, revealing that 57 percent of all agent activity focuses on cognitive work rather than routine administrative tasks Cryptopolitan. As the market for agentic AI is projected to grow from $8 billion in 2025 to $199 billion by 2034, the data indicates agentic AI is already being deployed by high-value knowledge workers to streamline productivity and research tasks Cryptopolitan.

Perplexity's researchers developed a hierarchical agentic taxonomy revealing that Productivity & Workflow accounts for 36 percent of all agentic queries, followed by Learning & Research at 21 percent Cryptopolitan. Personal use constitutes 55% of queries, with professional use at 30% and educational use at 16%, with professional queries dominated by document editing (13.3%) and professional networking (12.5%) OpenAI.

The top operating environments are enterprise stack staples: Google Docs for document editing, LinkedIn for professional networking tasks, and course platforms like Coursera for learning and research Cryptopolitan. User behavior evolves over time, initially testing agents with lower-stakes tasks like travel or media but shifting toward cognitively oriented topics like productivity, learning, and career, with usage proving sticky once workflows are established OpenAI.

While Large Language Models serve as reasoning engines, agents act as the hands capable of executing complex, multi-step workflows with minimal supervision, yet visibility into how these tools are actually being utilized in the wild has been opaque, relying largely on speculative frameworks or limited surveys Cryptopolitan. A common view suggests agents will primarily function as digital concierges for rote administrative chores; however, the data challenges this view with cognitive work dominating usage patterns Cryptopolitan.

For CISOs and compliance officers, this presents a new risk profile as AI agents are not just reading data but actively manipulating it within core enterprise applications Cryptopolitan. The post-General Availability period of Comet accounted for 60% of adopters, signaling mainstream interest is spiking, with winners of the next cycle not just building better LLMs but agents capable of executing specific, high-value tasks within environments where knowledge workers spend their day OpenAI.

A procurement professional used the assistant to scan customer case studies and identify relevant use cases before engaging with a vendor, illustrating practical enterprise value beyond theoretical capabilities Cryptopolitan. Usage analysis shows that once a user begins a session in a productivity workflow, they rarely switch context, suggesting agents are successfully maintaining long-horizon tasks OpenAI.

The data confirms that LinkedIn email management and professional networking are "killer apps" for professional users, with LinkedIn accounting for 9.42% of all agent activity environments OpenAI. The concentration of activity within enterprise stack staples means organizations must prepare for agents actively manipulating data within core applications including Google Docs, LinkedIn, and YouTube, requiring governance frameworks addressing this new risk profile Cryptopolitan.

The research reveals that initial pilot programs should anticipate a learning curve where usage matures from simple information retrieval to complex task delegation over time.

The challenge for enterprises is to harness this momentum without losing control of the governance required to scale safely, with agents not just reading data but actively manipulating it within core enterprise applications Cryptopolitan. For startups, the message is clear: the productivity category is the battleground, with document and form editing comprising 21.5% of productivity queries, presenting massive opportunity for specialized agents that integrate deeply with tools like Google Docs and Notion OpenAI.

Organizations must establish frameworks determining which tasks can be delegated to agents, which require human oversight, and which must remain entirely human-controlled. The finding that usage is "sticky" once established in productivity workflows suggests early adopters gain compounding advantages as efficiency gains reinforce continued usage patterns, creating potential competitive gaps between organizations successfully implementing agentic workflows and those remaining in experimental phases.

The transition to enterprise workflows led by AI agents is underway, driven by the most digitally capable segments of the workforce Cryptopolitan. The most encouraging finding for the industry is how user behavior evolves over time, initially testing with lower-stakes tasks but shifting toward more cognitively oriented work, with usage remaining sticky once workflows are established OpenAI. Decision-makers should monitor whether the evolution from simple queries to complex task delegation follows predictable adoption curves, as understanding maturation timelines will inform realistic expectations for enterprise-wide deployment. The dominance of cognitive work over administrative tasks suggests organizations must prepare infrastructure and governance for agents performing high-value knowledge work rather than routine automation.

Source & Date

Source: Perplexity AI Research, Artificial Intelligence News, Harvard University, Euronews, PYMNTS
Date: December 10, 2025