Couchbase expands AI data plane to address agent traffic surges

Couchbase Inc. announced an expansion of its AI Data Plane ahead of this week’s release. The update includes native integration with Amazon Bedrock, semantic caching, a managed Agent Memory server, and vector search capabilities within its analytics engine, now named Operational Insights.

Tim Finley, senior vice president of AI at Couchbase, highlighted that current AI cost models are failing because they focus on model pricing rather than data layer performance. As agents perform real work in production, they trigger dozens of queries and multiple model calls per task. This behavior makes the surrounding infrastructure, not just the model, the primary determinant of cost and speed.

The economics of agentic AI

Finley explained the problem using queuing theory, noting that latency grows faster than traffic as load increases. He cited a Cisco Systems Inc. study showing agents generate 450% more traffic per task than humans doing the same work. Additionally, Cloudflare data indicated machine traffic surpassed human traffic on the web in June, occurring about 18 months earlier than industry forecasts predicted.

“Capacity plans built for human growth curves are likely to break,” Finley said. He described agents as the most demanding users for system performance because they operate continuously and require parallel task execution. Most production databases were sized for human interaction patterns, such as clicking and waiting, which do not account for the relentless, parallel nature of agentic workflows.

Infrastructure bottlenecks and trade-offs

Treat agents as “just more users,” and the data tier becomes the bottleneck, usually right after a successful pilot convinces the business to scale. Finley warned that this is the most underappreciated risk in agentic AI. Agents fan out into parallel tasks and need both current and historical data in the same loop without sleeping or pausing.

Companies are facing difficult trade-offs between model choice and capabilities to manage these rising costs. The pre-briefing comments suggest that IT and business leaders have not fully accounted for the infrastructure impact of scaling agents. When companies downgrade intelligence to pay the bill, they risk compromising the utility of their AI investments while still struggling with data layer constraints.

What should you do?

Organizations deploying AI agents should review their capacity planning assumptions, as human-centric growth curves may no longer apply. Security and IT teams should monitor machine traffic patterns closely, especially given that machine traffic has already surpassed human traffic. Leaders should evaluate the total cost of ownership, including data retrieval and processing, not just model inference costs.

What to do and how to stay safe: Couchbase

• Review database capacity plans to ensure they can handle parallel, continuous agent traffic rather than sequential human interactions.
• Monitor machine-generated traffic volumes to detect early signs of infrastructure bottlenecks or latency spikes.
• Evaluate the total cost of AI tasks, including data retrieval and context management, not just model token prices.
• Test agent workloads in production environments to measure actual traffic generation and latency before full-scale deployment.

Step-by-step guide: Prompt Injection Warning Signs: How to Spot AI Manipulation

General security guidance from the Threat Times newsroom. It is not confirmed advice from the organisations named in this story.

Frequently asked questions

How much more traffic do AI agents generate compared to humans?

According to a Cisco Systems Inc. study cited by Couchbase, agents generate 450% more traffic per task than humans performing the same work.

When did machine traffic surpass human traffic on the web?

Cloudflare data shows that machine traffic surpassed human traffic on the web in June, which was about 18 months earlier than forecasted.

What new features are included in Couchbase's AI Data Plane expansion?

The expansion includes native Amazon Bedrock integration with semantic caching, a fully managed Agent Memory server, and vector search in its analytics engine.