Architecting for Agentic AI – Moving from Simple Automation to Autonomous Decision-Making

Today, we see AI features like grammar checks, code completion, and summarization tools embedded across applications, much like spellcheck. These tools are easy to enable and use without much thought. But they only scratch the surface of what AI can do. Beyond these conveniences lies Agentic AI—a step beyond automation toward autonomous systems that act independently across applications and workflows. This shift requires a new level of enterprise architecture, as organizations need to build for AI-driven workflows at the platform level.

What is Agentic AI?

Agentic AI goes beyond the familiar integrations found in most enterprise software today. While typical automation helps with structured tasks within one framework (like a spreadsheet assistant helping with formulas), agentic AI takes contextual action across multiple systems without constant user prompts.

Take an AI-driven customer service platform as an example. A standard chatbot can answer predefined questions, but an agentic AI service platform takes it further. It can detect trends, flag potential issues, and initiate actions on its own. If it identifies repeated customer complaints about a product feature, it could open a support ticket, analyze historical data for solutions, and notify the product team—all without human intervention. It’s not just about following rules; it’s about adapting to real-world business situations.

Imagine a virtual sales assistant that doesn’t just sit waiting for input but actively monitors site activity, detects when a customer is likely to abandon a cart, and initiates a discount offer. This type of AI agent requires a different approach to system architecture—one that connects data across departments, enabling agents to interact with each other and respond autonomously to current business needs.

Architecting for Agentic AI

Traditional enterprise architectures designed for straightforward integrations can’t handle the demands of agentic workflows. These agents need an underlying platform that supports autonomous decisions across different business domains, allowing them to tap into real-time data and communicate seamlessly with each other.

This might look like a CRM agent that analyzes engagement data, cross-references it with support logs, and triggers outreach when engagement metrics fall below a certain threshold. To support these interactions, the architecture needs to shift from task-oriented automation toward a connected, intelligent framework that enables these agents to work together across data systems.

Platform Control vs. Platform Lock-In: Where Do You Build Your Agents?

Every major software and cloud provider wants to be your agentic AI provider. For instance, Salesforce’s ecosystem makes it easy to create agentic workflows within the CRM platform, reducing friction for the end user. But if you need integration across platforms—say between a Salesforce CRM agent and an SAP Logistics agent—challenges arise.

This leaves enterprises with a key decision: build workflows within a single ecosystem, accepting potential integration limitations, or create a more independent platform to support agentic workflows across systems. Infrastructure providers are betting that businesses will favor controlling their own AI platforms, enabling flexibility to design agentic workflows that aren’t constrained by a specific vendor’s ecosystem.

For example, VMware’s Cross-Cloud initiative allows enterprises to leverage a multi-cloud model that works across environments but lacks tight integration across independent systems​. On the other hand, intelligent ERP systems like those offered by SAP provide cross-functional workflows but may be difficult to integrate with systems outside their ecosystem​. Each approach has its trade-offs, making it essential for organizations to align their agentic AI strategy with their broader technology goals.

The CTO Advisor’s Take

AI is moving faster than most lines of business can fully adopt it. For once, this rapid pace has aligned the interests of IT and business units, creating an opportunity to work together on agentic AI. My advice? Let your software vendors handle the simpler generative AI features and integrations. Instead, focus on building the agentic workflows that have a direct impact on your critical business processes. This control is essential for avoiding vendor lock-in and maintaining flexibility for future needs.

By taking advantage of this alignment between IT and business, we can establish sustainable, agentic AI workflows that are tailored to business priorities and built for adaptability. This is our moment to architect solutions that won’t just meet today’s needs but will allow us to leverage AI as a strategic advantage in the years ahead.