Jakob Freund is the CEO of Camunda, a software company innovating end-to-end process orchestration and automation with agentic AI.
Every week, another enterprise announces an AI deployment. And yet, the results at scale tell a different story. According to my company’s survey of 1,150 senior IT decision-makers, business decision-makers and enterprise software architects responsible for process automation, 71% of organizations now use AI agents, but only 11% of use cases reached production last year. A recent survey from Splunk shows that 44% of enterprises in the Global 2000 use agentic AI, but 68% worry that these agents will behave unpredictably. The gap between what enterprises deploy and what they actually deliver has never been wider.
The reason is not the agents themselves. The problem is what happens around them—the coordination, the handoffs, the governance and the question of who holds the process together when an agent’s decision cascades across a dozen interconnected systems. What most enterprises are missing is orchestration. The CIOs and CTOs who close that gap first will be the ones who move AI from pilots to production this year.
The False AI Automation Binary
Most boardroom conversations treat AI automation as a binary decision: You either hand a business process to an agent or you don’t. That framing misses the more important strategic question entirely. The real decisions are centered around when to use agents, and in what combination with structured, rules-based automation.
Think of automation as a spectrum. At one end, deterministic processing handles high-volume, predictable work, making it faster, cheaper and fully auditable. At the other end, agentic behavior handles genuine complexity—such as ambiguous inputs, novel situations and decisions that require contextual judgment. The best results use both, deliberately, within the same process.
For example, standard purchase orders should run deterministically. Routing them through a reasoning agent is expensive overkill. But when an exception occurs (like an inventory discrepancy, an unusual payment term), you want an agent that handles it without stalling the operation. Using one process with two modes means you’re applying the right approach at the right moment.
The Insight That Separates Leaders From Laggards
Once you understand this false binary, there’s one strategic insight that changes how you should plan your AI road map. The right balance between agentic and deterministic shifts over time, and the best organizations design for that shift from day one.
Consider a pricing approval process. In phase one, an agent analyzes requests and generates recommendations, but humans decide every case. In phase two, the agent auto-approves discounts below 10%, because experience has shown it makes the right call. Phase three reflects higher confidence: The agent applies risk scoring and handles a wider range of scenarios automatically, routing only genuinely ambiguous cases to humans.
Then phase four arrives. As patterns become clear and predictable, the organization hardens them into deterministic rules. If the scenario matches known patterns, the system applies standard logic instantly, at near-zero cost, with complete auditability. The agent moves on to the edge cases that still require genuine reasoning. What was once agentic becomes deterministic as patterns mature.
Transferring once agentic processes to deterministic is not a failure of the agent. It is the highest form of success—and a fundamentally different way to measure AI ROI.
Three Things CIOs Should Do Now
The organizations best at blending deterministic and agentic processes share three common practices.
1. They Map Their Existing Agent Deployments Before Adding More
Most enterprises already have agents scattered across marketing, finance, operations and support—each built by a different team, using different frameworks and with no shared governance. As a CIO, you need a current-state picture before you can make smart decisions about where orchestration adds the most value.
2. They Define Platform Standards Rather Than Tool Mandates
Requiring mandatory logging, versioning, cost tracking and access controls across all agent deployments does not slow teams down—it gives them the foundation to move faster safely. The goal is consistency, not perfection and those standards become the baseline for the orchestration layer that connects everything.
3. They Select One Or Two High-Impact Business Processes And Orchestrate Them From End To End
Some examples could include loan origination, a claims process or an employee onboarding sequence. Any process that involves multiple agents, human approvals, long-running state and an audit trail is a good candidate. Demonstrating measurable outcomes in a real operational context builds the organizational confidence to scale.
What Governance Actually Requires
The EU AI Act and emerging global frameworks require auditability, explainability and demonstrated human oversight for AI in production. That should be a forcing function for building AI the right way. An orchestration layer that captures every agent decision, every human handoff and every system interaction automatically gives CIOs exactly what regulators ask for.
More importantly, it gives operations teams what they need to intervene when something goes wrong and to improve continuously as production data accumulates. Agents are genuinely capable. The enterprises that give them real authority over operations that matter will be the ones that built the coordination layer first—the layer that makes it safe to let AI run the work.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?