New deployment data from Perplexity reveals how AI agents are driving greater workflow efficiency by taking over complex enterprise tasks.
For the past year, the technology sector has operated on the assumption that the next evolution of generative AI will move beyond conversation and into action. While large-scale language models (LLMs) act as inference engines, “agents” act as hands that can execute complex multi-step workflows with minimal supervision.
Until now, however, visibility into how these tools are being used in practice has been opaque, relying primarily on speculative frameworks and limited research.
New data released by Perplexity analyzes hundreds of millions of interactions with the Comet browser and assistant, providing the first large-scale field study of a general-purpose AI agent. Data shows that agent AI is already being deployed by high-value knowledge workers to streamline productivity and research tasks.
Understanding who is using these tools is essential to predicting internal demand and identifying potential shadow IT vectors. This study revealed significant heterogeneity in adoption. Users in countries with high GDP per capita and high educational attainment are much more likely to use agent tools.
Even more important for business planning is the breakdown of job categories. Implementations are concentrated in digital and knowledge-intensive areas. The “Digital Technology” cluster has the largest share, accounting for 28 percent of adoptions and 30 percent of queries. This is closely followed by academia, finance, marketing and entrepreneurship.
Together, these clusters represent over 70% of the total adapters. This suggests that the individuals most likely to utilize agent workflows are software engineers, financial analysts, and market strategists, who are the most expensive assets in an organization. These early adopters aren’t dabbling. Data shows that “power users” (those with pre-existing access) make nine times more agent queries than the average user, making this technology essential when integrated into workflows.
AI agents: partners for enterprise tasks, not butlers
To move beyond marketing narratives, companies need to understand the utility these agents provide. The prevailing view suggests that agents act as “digital concierges” that primarily perform mechanical administrative tasks. However, data casts doubt on this view. This means that 57 percent of the agent’s total activity is focused on cognitive tasks.
Perplexity researchers have developed a “hierarchical agent taxonomy” for classifying user intent, making the use of AI agents practical rather than experimental. The primary use case is ‘productivity and workflow’, accounting for 36% of all agent queries. This is followed by “learning and research” at 21%.
A specific anecdote from the study shows how this translates into company value. For example, procurement professionals used the assistant to scan customer case studies and identify relevant use cases before negotiating with vendors. Similarly, treasurers delegated the tasks of filtering stock options and analyzing investment information. In these scenarios, the agent handles information collection and initial synthesis autonomously, allowing humans to focus on the final decision.
This distribution provides clear metrics for operational leaders. The immediate ROI of agentic AI is to extend human capabilities rather than simply automating low-level friction. The study defines these agents as systems that “automatically cycle between three iterative phases: thinking, acting, and observing to achieve an end goal.” This feature allows it to support “deep cognitive work,” where it acts as a thought partner rather than a simple butler.
Stickiness and cognitive transfer
A key insight for IT leaders is the “stickiness” of AI agents to enterprise workflows. The data show that users exhibit strong persistence within topics in the short term. When a user asks an agent for a productivity task, subsequent queries are more likely to remain in that domain.
However, user journeys often evolve. New users often “test the waters” with low-risk queries, such as asking for movie recommendations or general trivia. Over time, metastasis occurs. The study notes that although users may access it through a variety of use cases, the share of queries tends to shift toward cognitively oriented areas such as productivity, learning, and career development.
Once users hire an agent to debug their code or summarize their financial reports, they are unlikely to return to low-value tasks. The Productivity and Workflow categories have the highest retention rates. This behavior means that initial pilot programs should anticipate a learning curve as usage matures from simple information retrieval to complex task delegation.
The “where” of agent AI is just as important as the “what.” Perplexity’s research tracked the environments in which these AI agents operate: specific websites and platforms. The intensity of activity varies by task, but the top-level environment is the primary environment in a modern enterprise stack.
While Google Docs is the primary environment for editing documents and spreadsheets, LinkedIn dominates professional networking tasks. For “learning and research,” activities are divided between course platforms such as Coursera and research repositories.
This presents a new risk profile for CISOs and compliance officers. AI agents do more than just read data. We’re actively working with it within our core enterprise applications. In this study, we specifically define agent queries as queries that involve “browser control” or actions to external applications via APIs. When an employee asks an agent to “summary a customer story,” the agent is working directly with their own data.
The concentration of environments also highlights potential for platform-specific optimizations. For example, the top five environments account for 96% of professional networking queries, primarily on LinkedIn. This high degree of concentration suggests that companies could quickly gain efficiencies by developing specific governance policies and API connectors for these high-traffic platforms.
Agent AI business planning based on Perplexity data
The proliferation of capable AI agents creates new areas of consideration for business planning. Data from Perplexity confirms that we are past the speculative stage. Agents are now used not only to exchange information but also to plan and execute multi-step actions that modify the environment.
Operations leaders should consider three immediate actions:
- Audit productivity and workflow friction point Within high-value teams: The data shows that this is where agents naturally find their footing. If your software engineers and financial analysts already use these tools to edit documents or manage accounts, formalizing these workflows can standardize efficiency.
- Get ready for augmented reality. The researchers point out that while agents have autonomy, users often break down tasks into smaller parts and delegate only subtasks. This suggests that the work of the near future will be collaborative, requiring employees to be upskilled in how to effectively “manage” their AI counterparts.
- Address infrastructure and security layers. The boundaries of data loss prevention expand as agents operate in an “open world web environment” and interact with sites such as GitHub and corporate email. Policies should differentiate between chatbots that provide advice and agents that run code or send messages.
Early evidence from Perplexity serves as a bellwether, as the market for agent AI is projected to grow from $8 billion in 2025 to $199 billion by 2034. The transition to AI agent-driven enterprise workflows is being driven by the most digitally capable segments of the workforce. The challenge for companies is to capitalize on this momentum without losing the governance control they need to scale safely.
See also: Accenture and Anthropic partner to power enterprise AI integration
Want to learn more about AI and big data from industry leaders? Check out the AI & Big Data Expo in Amsterdam, California, and London. This comprehensive event is part of TechEx and co-located with other major technology events. Click here for more information.
AI News is brought to you by TechForge Media. Learn about other upcoming enterprise technology events and webinars.
