This is the third blog in the series Unlocking AI ROI. The previous articles discussed goals for agentic ROI and Model Context Protocol focus as foundational for realizing value and establishing relevance for your agentic strategies. This blog will explore how those translate to agentic application delivery patterns in practice.
The organizations we speak with have been performing a lot of AI experimentation, but are struggling to get out of proof of concept (POC) with their AI applications. To achieve the topline and bottom line ROI metrics discussed in the first blog of this series, organizations need to move beyond proof of concept for AI applications and start trying out advanced agentic application patterns. However, many organizations have already learned, the road to production is often riddled with complexity—especially when it comes to deploying autonomous applications. Agentic applications are AI-powered systems that don’t just respond; they reason, plan, and act in dynamic ways. With that power comes the need for a new level of operational rigor paired with very fast and frequent iteration.
Why Observability is Critical for Driving Agentic ROI
One of the biggest barriers to achieving ROI from AI applications is the need for constant observability and adaptability. AI-enabled applications are not “set it and forget it” applications—they must be continuously monitored for response quality, performance degradation and alignment with evolving business and user contexts. For agentic applications that incorporate autonomous reasoning behaviors, monitoring is even more critical. These systems are inherently more complex than GenAI applications, relying heavily on context, prompt chaining, and interactions with multiple services. A lapse in quality or a mismatch in behavior will quickly lead to a poor user experience—or worse, costly errors.
This means organizations must capture telemetry not just from the model, including user interactions, decision-making flows, and downstream actions triggered by agents, but also from the full stack of application services and the application environment. This expanded monitoring scope represents a fundamental shift from traditional software maintenance practices. Further, while conventional applications might function well with monthly or quarterly update cycles, agentic systems operate in a hyper-dynamic reality. These AI applications often require multiple redeployments per day to maintain optimal performance and security. Organizations that recognize and adapt to changing application requirements are better positioned to realize maximum value from their agentic software investments.
To sustain high-quality agentic software, enterprises need detailed and actionable monitoring, model assessment and auditing, and a continuous update culture. This operational overhead can quickly become a bottleneck—unless teams have the right platform in place. This is where application platforms like Tanzu stand out. By abstracting away the environmental complexity, these pre-engineered platforms enable teams to focus on what truly matters: building intelligent, differentiated, agentic experiences.
Platform as a Service (PaaS) solutions give development teams a competitive edge when delivering agentic applications, by automating the management of the applications’ lifecycle and the environment they run on, as well as supporting services. This means updating and deploying agentic software more quickly, consistently and safely without constant manual intervention unless necessary. Ultimately this increases developer velocity and your time to ROI.
Maximizing Agentic ROI Through Platform-as-a-Service
Realizing ROI from agentic applications means you will need to embrace a new set of operational practices—and adopt an automated set of tools rather than creating unique environments from scratch for each application iteration. A platform-centric approach can dramatically reduce friction between innovation and deployment, helping teams transition from experimentation to enterprise-grade delivery faster. As organizations strive to turn AI ambition into business value, the ability to move from POC to payoff—quickly and repeatedly—will define who leads and who lags in the age of agentic applications.
Despite the many ROI benefits, organizations face significant hurdles when preparing to deploy agentic applications in production. Any enterprise deploying AI models in an application requires best in class observability so that response quality can be monitored. Constant AI model monitoring, assessment, and application updates are critical to maintaining accurate and optimized outputs. When adding “thinking” patterns with agentic applications, platform engineering teams should prepare for even more frequent updates and redeployments because of the highly contextual nature of agentic applications. An application platform like Tanzu can facilitate agentic innovation by removing complexity from continuous updates and redeployments so development teams can focus on agentic innovation.
Achieving Continuous Application Iteration with Tanzu Platform
VMware Tanzu Platform is an AI-ready, private PaaS that helps remove complexity so organizations can develop, operate, and optimize intelligent applications as easily as any other application. Tanzu Platform includes fine-grained observability tools for model audit, version control, and feedback loop management, making it easier to identify optimization opportunities and iterate on agentic applications in production.In this recent video, Adib Saikali, Distinguished Engineer, Tanzu, walks through how Tanzu Platform can help organizations master the “AI iteration dance”.
Also be sure to check out Cloud Foundry Weekly vlog series which has multiple shows dedicated to AI application delivery.
If you are looking to build new AI skills, don’t miss Tanzu’s Platform Engineering Skills for GenAI and Agentic Training Workshop on May 13th, co-located with the 2025 North America Cloud Foundry Day in Palo Alto, California.
