By Chris Vessey, VPof Innovation & Customer Value
There is a lot of noise right now about AI agents.
Autonomous this. Self-learning that. “AI-native foundations.” The market is full of hype, sweeping claims, and overly simplistic promises. But after seeing real deployments (the ones that truly work, and the ones that quietly fail) I’m increasingly convinced of something simple:
Agents succeed or fail based on leadership, governance, and architectural intent, not based on how creative and easy the solution sounds.
The difference between success and failure is rarely the model. It’s the operating discipline behind it.
Below are six observations that consistently show up across the organizations seeing meaningful outcomes from agentic systems.
The organizations seeing success with agentic approaches are not the ones who started with technology.
They started with problem and process mastery.
Agents are not shortcuts that allow you to avoid complexity. In practice, they amplify whatever complexity already exists. That is why early agentic breakthroughs happened in Engineering: the people building the first agents understood their problems with clarity and precision.
If your process is unclear, politically sensitive, fragmented across systems, or culturally misaligned, your agent will simply operationalize that dysfunction faster.
The strongest agents we see are designed by leaders and experts who have lived the process end-to-end. These are people who understand risk thresholds, escalation nuance, and the real-world constraints that shape decision-making. They also lead teams with enough creativity to reimagine outcomes instead of just automating tasks.
Because this is not automation. It is structured decision-making at scale, using the full spectrum of technologies available.
And yes, this needs to be said out loud: not everything is an agent. But you can only determine the reality if the right insight is provided by the right people, supported by the right experienced partner. Otherwise, every problem becomes positioned as having an “agentic solution,” regardless of whether that is true.
Yes, you can preconfigure frameworks. Yes, you can run proof of concepts (we do both here at ORO). But agents are not culturally neutral.
“Out of the box” implies plug-and-play, and that is naïve in any real organization that has even one custom field, workflow exception, risk policy or an environment that changes regularly.
That does not mean building agents is difficult. But my experience is that every company’s risk appetite differs. Commercial posture differs as well. As a result, control and execution frameworks vary significantly across customers. An agent operating in a global bank must behave differently than one in a high-growth technology firm.
One of the best ways to think about agents is as employees rather than technology.
Like any employee, an agent has inputs, intelligence, escalation paths, outputs, and actions. The difference is that agents behave more consistently than humans, with a higher degree of speed and accuracy. In many cases, people expect agents to have a lower degree of creativity and judgment, which would be mostly true, but we are seeing even that shift in recent projects.
Also, they don’t need sleep.
However, perhaps most importantly, they can’t be held accountable. So while there are many common alignments of humans and agents, this is a differentiator we need to factor in.
Ultimately, though, in the same way you don’t hire someone “out of the box,” but rather you recruit, onboard, shape, and align them to your operating model, the same is true for agents. They need to be hyper-personalized to your tech stack and ultimately to your culture. But they also need a decision-owning framework.
If you treat agents as static automation, you will either over-constrain them into irrelevance or under-govern them into risk. Neither approach scales, and neither delivers lasting value.
This is where many conversations about agents become overly simplistic. Too often, organizations think of agents as fixed task executors. The real value, however, is found in the way an Agent can adapt to an enterprise workflow which is an intent-driven journeys, evolving dynamically as context shifts rather than linear sequences of tasks.
A conversation might begin with, “I need to onboard a supplier.” Within minutes, that request can expand into a risk assessment discussion, contract review, budget confirmation, policy clarification, or even negotiation support. What appears to be a single workflow quickly becomes a series of shifting intents that require contextual awareness and structured decision-making.
At ORO, we are focused on humanizing procurement, and at the intersection of humanized procurement and agentic workflows sits a critical capability: intent management. We are enhancing and innovating around intent-aware orchestration so users can move seamlessly between intents within a single AI conversation, rather than being forced through disconnected workflows.
Supporting this orchestration is a complex and expanding knowledge graph that connects:
This foundation is essential because speed alone is not value. Accelerating process without enriching decision-making simply compresses poor decisions into shorter timeframes. The real opportunity lies in using AI not only to move faster, but to move smarter.
Intent-aware systems enriched by structured knowledge graphs enable:
This is the shift from automation to augmentation, or from executing tasks to improving the quality of enterprise decisions.
As intent-aware agents operate across multiple workflows, their capability increases, and so does the responsibility associated with deploying them. Implementation cannot be treated as a simple “switch-on” event. Sustainable success requires a structured and disciplined rollout model that balances innovation with control.
Effective deployment requires:
Without these elements, even well-designed agents will struggle to scale in complex enterprise environments.
Governance should operate at two distinct levels. The first is agent behavior governance, which defines what an agent can do, where it can act autonomously, and when escalation is required. The second is intent transition governance, which governs when and how an agent is permitted to move between domains or to pass an action or information to another agent as user intent evolves.
When governance is introduced reactively, agent programs often stall under risk concerns or inconsistent outcomes. When agentic governance is embedded directly into architectural design, it becomes an enabler of scale rather than a constraint.
Much of the public narrative around AI agents focuses on “hours saved.” While time efficiency can be a component of value, this framing reflects an automation mindset rather than a transformation mindset.
Agentic systems reshape enterprise performance across multiple dimensions, including:
An autonomous negotiation agent that improves margin by basis points across categories creates commercial impact that extends well beyond time reduction. An opportunity-identification agent surfacing commercial leakage across a portfolio drives measurable financial recovery. A PR review agent reducing policy breaches while accelerating compliant execution strengthens governance and operational resilience.
These are not primarily time stories; they are strategic value stories.
Value should therefore be measured in terms of commercial uplift, risk mitigation, experience improvement, and the redeployment of human capacity into higher-order strategic work, not just time saved. When agent success is reduced to stopwatch metrics, organizations risk underestimating the scale of transformation available to them; it’s too basic for real world value creation.
No organization should “buy agents” the way it buys software. If you do, it will never be scalable and we will end up with the same deployment issues we have today in the S2P Suite world.
The real work is defining where value truly sits, then designing agents around real operating constraints. That means embedding governance frameworks and running structured agentic design workshops. Then, when deployment begins, agents must be hardened before scale.
This is why partnership models, design, build and run approaches matter:
Technology provides the capability. Partnership engagement provides the discipline.
In many respects, 2025 was defined by experimentation. The next phase will be defined by proven value realization.
Agents will not transform enterprises simply because they are intelligent. Transformation occurs when intent is clearly understood, knowledge is structured and connected, governance is embedded into design, and value is measured with ambition rather than narrow efficiency metrics.
Agents are not magic systems operating in isolation. They are managed colleagues functioning within intent-aware architectures. The organizations that recognize this and design accordingly will be the ones that unlock sustainable advantage.