Deluxe manages 50 AI agents as operational hurdles mount

The primary challenge of enterprise artificial intelligence is shifting away from simply accessing powerful models and toward the daily realities of managing them. Instead of relying on a single provider, businesses are increasingly deploying diverse ecosystems of models to balance cost, speed, and risk. For example, payments and data firm Deluxe now runs more than 50 AI agents, utilizing a centralized gateway to automatically route user queries to the most appropriate model based on specific performance and cost metrics.

However, scaling these multi-model architectures is exposing severe underlying weaknesses in corporate data infrastructure. According to a recent survey by data intelligence company Collibra, 72 percent of AI decision-makers reported that a poor data foundation was the primary reason their enterprise AI initiatives failed to meet expectations. Without clean, unified data, even the most advanced frontier models cannot deliver reliable business value.

Governance and oversight present another major operational bottleneck, particularly as companies adopt autonomous agents. A report released by EY reveals that nearly six in 10 respondents at organizations using agentic AI admit that no single department or group oversees these agents once they are deployed. Furthermore, nearly half of the surveyed organizations have not updated their governance frameworks to address the unique risks of agentic systems, and four in 10 lack complete visibility into the active AI tools running on their corporate networks.

For IT leaders and system architects, this shift means that success no longer depends on chasing the latest model release. Instead, practitioners must focus on building robust middleware, establishing clear lines of ownership, and modernizing data pipelines. The future of enterprise AI integration lies in creating sustainable operational frameworks that can monitor, secure, and orchestrate dozens of different models simultaneously.