Classic data architectures are insufficient for the deployment of AI agents at the enterprise level, according to a study by Google Cloud and MIT Technology Review Insights. Companies want to scale up quickly but are encountering outdated systems, limited data access, and latency issues.
The transition from small-scale AI agent pilots to large-scale applications in enterprises is not without its challenges. Classic data stacks are primarily designed for retrospective analytics, but AI agents require real-time access to current data and business context. The report surveyed three hundred international data and technology leaders about their experiences and expectations regarding AI agents.
AI agents at scale
According to the research, only ten percent of organizations currently use AI agents at scale. Nevertheless, nearly 70 percent expect to be there within two years. Companies recognize that they must adapt to a new way of working, where systems not only provide intelligence but can also actually take action. Recent research from Cloudera showed a similar trend, revealing that 95 percent of organizations postpone AI projects due to infrastructure limitations.
Trust depends on data
Only half of the respondents (51%) trust that their current data architecture enables AI agents to make accurate and relevant decisions. More than half (55%) indicate that legacy data systems form an obstacle to large-scale rollout. For 52 percent, latency in particular is the biggest bottleneck for rapid decision-making.
The report shows that AI has access to an average of only 45 percent of all enterprise data. Organizations that give their AI applications access to more than 70 percent of the data see a clear difference: 36 percent find that their agents’ decisions are consistently accurate, while 64 percent speak of largely accurate results. Sufficient and up-to-date data are therefore crucial for reliability.
Legacy systems
Companies are increasingly moving away from systems that merely analyze and are focusing on so-called ‘systems of action’ that can reason and act in real time. However, existing systems often prove unprepared for this shift. The research makes it clear that modernization of the data architecture is necessary to utilize the full potential of AI agents.
