Corporate AI 'Agents' Are Actually Just Chatbots

While companies rush to adopt AI agents, the reality is that most cases have not reached a level that can truly be called an 'agent,' according to a survey conducted by VentureBeat in June 2026. The survey, targeting 101 companies with more than 100 employees, conducted a cross-sectional analysis of which platforms enterprises operate agents on and what they prioritize.

At the core of the survey is the gap between 'what they want to do' and 'what they can actually do.' An AI agent refers to a system that can not only answer questions but also autonomously execute multiple steps. However, 71% of respondent companies acknowledged that more than three-quarters of the systems they call 'agents' are actually nothing more than chatbots that only respond to single commands. Only 10% of companies have true agents—workflows that autonomously handle multi-step processing—that exceed half of their total operations.

Regarding platform selection, Anthropic's Claude is the primary foundation for 40% of companies, significantly outpacing second-place Microsoft (18%) and OpenAI (13%). The most commonly cited reason for selection was 'native affinity with cutting-edge base models' (21%), while the top success indicators were 'task completion stability' (32%) and 'multi-step workflow management' (28%). In other words, what companies demand from agents is 'reliable operation,' and the model's performance itself has become the central axis of platform selection.

Regarding control mechanisms, it is expected that by the end of 2026, 51% of companies will adopt a 'hybrid-type' management infrastructure combining provider-offered features with external orchestration tools. Meanwhile, only 6% of companies chose the 'managed service type,' which delegates control entirely to providers. Behind this is vigilance against vendor lock-in—the risk of becoming too dependent on specific companies, making it difficult to switch to others—with 35% of companies citing this as the greatest risk.

As for investment priorities, agent workflow construction tools rank first at 34%, followed by security and access control management at 25%. Meanwhile, serious challenges have emerged in cost management. AI agents incur costs in units called 'tokens' each time they process, but 27% of companies lack the ability to stop processing in real-time if an agent runs amok. It remains common for companies to discover abnormalities only after receiving a billing invoice.

What this survey highlights is the structural problem that agent 'names' and 'capabilities' are misaligned. While orchestration infrastructure—the mechanism for commanding and coordinating multiple agents—is developing rapidly, there is a paradoxical situation where true agents that should operate on top of it barely exist. Demand for hybrid control platforms and real-time cost monitoring tools is expected to grow, and 'how to build an operational system that can actually be run' is emerging as an essential issue for companies beyond platform selection.

The survey was conducted as part of VentureBeat's ongoing Pulse Research series. The target was 101 companies with more than 100 employees, with company sizes of 100-499, 2,500-9,999, and 50,000+ employees each representing 21% of the distribution. Respondents' positions were centered on decision-making layers, including Product/Program Managers (15%), CIO/CTO/CISO (13%), and Consultants/Advisors (13%).