You’re Asking the Wrong Question About Agentic AI
The wrong question is: “How do we use AI agents to automate our processes?” The right one is: “What operating model are we building — and are we laying its foundation today?”
These lead to entirely different investment decisions, entirely different architectures, and entirely different competitive positions three years from now.
By Geraldine McBride, CEO, MyWave.ai | Former President & CEO, SAP Asia Pacific & Japan and SAP North America
The Automation Trap
Almost every C-suite conversation I’m having at the moment finds its way to the same question: how do we use AI agents to automate our processes? It’s a reasonable place to start. But I believe it’s the wrong destination, and the organisations that treat it as such may find themselves having optimised their way into a strategic cul-de-sac.
Process automation is not the destination; it’s the on ramp. And yet the majority of enterprises—including some of the most sophisticated organisations in the world— are approaching agentic AI as though it were simply a faster, more capable version of robotic process automation (RPA). They’re deploying agents to accelerate workflows that, in many cases, shouldn’t exist in their current form at all.
This distinction matters enormously if you operate in a complex, highly regulated industry: financial services, healthcare, energy, defence, or any sector where compliance is non-negotiable. The stakes of getting your AI strategy wrong aren’t merely competitive, they can be existential.
Regulatory exposure, audit failures, governance breakdowns: these risks are very real when probabilistic AI is layered onto deterministic compliance requirements without a clear architectural philosophy.
The more generative question is not how you automate your existing processes, but whether you are building the agentic operating model that will define how your enterprise creates value over the next decade or more.
Those are very different conversations — and they lead to very different investment decisions.
Three Waves. Most Organisations Are Surfing the Wrong One
Enterprise AI is evolving in recognisable waves. IDC’s enterprise AI adoption research maps this progression across a 15-year horizon, and understanding where we are, where we’re heading, and clarifies what’s genuinely achievable today versus what still belongs to a more distant future.
Wave 1 — AI-Enhanced Applications (2025–2029): AI adds intelligence to existing systems. Copilots, assistants, smart suggestions. Users still drive workflows manually, but AI makes them somewhat faster and less effortful. This is augmentation. It’s genuinely useful, but increasingly table stakes.
Wave 2 — Agent-Led Execution (2027–2033): Agents take ownership of entire functional domains. They don’t assist with procurement. They execute it. They don’t suggest supplier actions: they validate, route, escalate, and resolve. Humans govern outcomes; agents do the work. This is where meaningful competitive differentiation begins.
Wave 3 — Agents as Applications (2031–2037): Traditional interfaces gradually disappear. Users express intent while agents orchestrate execution across every system in the enterprise stack. This work is executed autonomously, within policy, with full auditability. This is the AI-native organisation in its mature form.
The honest assessment is that most enterprise AI deployments today are firmly Wave 1, dressed in Wave 2 language.
The window to establish genuine Wave 2 capability before your competitors do is narrowing more quickly than industry forecasts tend to suggest. The organisations that act now with architectural intent will have a compounding advantage over those that wait for the market to fully mature.
The Architecture Question That Deserves More Attention
There is a compelling but ultimately flawed idea circulating in the market: that AI agents will, in time, replace ERP systems altogether. It’s worth examining carefully, because in regulated industries this line of thinking leads organisations toward architectures that create rather than manage risk.
ERP systems like SAP, Oracle, and Workday exist to enforce data integrity, execute deterministic transactions, and withstand audit scrutiny. They are engineered for certainty. Large language model (LLM)-driven agents are, by their nature, probabilistic. You cannot run financial close, regulatory reporting, or any mission-critical compliance process on a probabilistic execution engine and expect the outcome to satisfy your auditors, your regulators, or your board.
The architecture that genuinely serves complex enterprises is a three-layer agentic operating model that preserves the strengths of each component:
The Intent Layer: Where humans set direction using natural language to define policies, constraints, risk thresholds. This is where the organisation’s values and standards are expressed, and where AI translates them into executable actions.
The Agentic Orchestration Layer: Where governed agents execute within mandatory guardrails. Every decision is logged. Every exception escalates through defined protocols. Compliance is enforced at the point of transaction, not reviewed after the fact.
The ERP Execution Layer: Your Systems of Record remain the authoritative source of financial truth. They become the engine rather than the interface, doing what they were designed to do, reliably and at scale.
This isn’t a workaround or a compromise position. It’s the only architecture that allows you to move with real pace without compromising the governance infrastructure that regulated industries quite rightly require.
Porter Was Right. He Simply Didn’t Know About Agents
Michael Porter’s Value Chain framework—first articulated in Competitive Advantage (1985)—remains as relevant as it ever was. What has changed, fundamentally, is the unit of execution within it.
In the traditional enterprise, people were the unit of work, supported by systems. In the agentic enterprise, agents become the unit of execution and people become the system of governance. It sounds like a subtle shift. In practice, it rewrites the operating logic of every function in the organisation.
I want to be clear that this is not a polite framing for headcount reduction. It is a genuinely different value proposition for your workforce. Your people move up the chain, from doing the work to setting intent, designing policy, handling the exceptions that genuinely require human judgement, and owning outcomes. The cognitive overhead of routine execution—the scheduling, routing, validating, chasing, documenting—that is what agents absorb.
Across every primary activity in Porter’s framework, from inbound logistics, operations and procurement to marketing and service, the pattern is consistent: agents execute deterministically within policy boundaries while humans govern with strategic judgement. Organisations that redesign their operating model around this reality will not simply be more efficient. They will be structurally more capable of competing.
Where You Start Shapes Where You End Up
When organisations ask me where to begin with agentic transformation, my instinct is always to start where the stakes are highest, the exceptions are most frequent, and the compliance requirements are most demanding. That is where agent mastery compounds the fastest, and where the capability you build is most transferable.
For most complex enterprises, that entry point is procurement. It spans multiple systems, crosses organisational boundaries, touches money and risk at every stage, and routinely encounters the kind of exceptions that traditional automation simply cannot handle gracefully.
An agent that can execute governed procurement workflows—validating suppliers, managing contracts, enforcing spend policy, generating audit trails in real time—has demonstrated that it can operate in your most demanding environment. That is meaningful proof of concept.
What the short-term lens tends to miss is that the capabilities you develop in that first domain do not stay there.
The policy models, the validation logic, the exception-handling intelligence become reusable institutional knowledge. The same patterns that govern procurement extend naturally to invoice processing, expense management, vendor risk, HR onboarding, and IT provisioning. Each domain builds on the last, and the organisation’s agentic capability compounds over time.
This is the distinction that separates organisations that chose a strategic platform from those that purchased a point solution. Both may achieve early ROI. Only one builds an enterprise capability that accelerates with use.
The Decision You’re Actually Making
Every enterprise AI investment being made right now is, at its core, one of two choices, even when it doesn’t present itself that way.
You can choose to optimise the present: deploy agents to accelerate existing workflows, reduce manual effort, and demonstrate quick wins. This is a legitimate and valuable path. But in three years, you are likely to find yourself selecting your next point solution, integrating it with the previous one, and wondering why your broader AI strategy feels less coherent than you’d hoped.
Or you can choose to build for the future: establish an agentic operating model—governed, auditable, policy-driven—that starts in one domain and extends across the enterprise as agent maturity continues to accelerate. The early ROI is comparable. The strategic trajectory is entirely different.
In highly regulated industries, the second path carries an additional advantage worth naming. It builds the governance infrastructure you are going to need regardless as regulators develop frameworks for AI accountability, as auditors demand explainability, and as boards begin asking sharper questions about how autonomous decisions are being made and by whom.
The questions worth sitting with is not how to automate your processes. It is what operating model are you building? Are the decisions you are making today genuinely laying its foundation, or overlooking an opportunity to prepare your organization for an agentic-driven future, today?
That is the question that will shape whether your organisation leads the next chapter of enterprise AI, or finds itself working hard to catch up with those who asked it earlier.
Frequently Asked Questions
What is an agentic operating model?
An agentic operating model is an enterprise architecture in which AI agents — rather than people — serve as the primary unit of execution across business workflows. Humans retain governance responsibility: setting intent, designing policy, handling exceptions that require genuine judgement, and owning outcomes.
The model sits above existing ERP systems (SAP, Oracle, Workday) rather than replacing them, using a three-layer structure comprising an intent layer, an agentic orchestration layer, and an ERP execution layer.
How is agentic AI different from RPA (Robotic Process Automation)?
RPA follows fixed, rule-based scripts. It is brittle by design: any exception outside the script requires human intervention or developer involvement.
Agentic AI reasons contextually. It interprets goals and policies, selects appropriate actions dynamically, and handles exceptions autonomously within defined guardrails.
Where RPA automates a specific task, an AI agent can execute an entire end-to-end workflow—such as supplier onboarding, contract validation, or invoice dispute resolution—adapting in real time to variations in data, stakeholder response, and system state.
Is agentic AI safe to deploy in regulated industries?
Yes, when architected correctly. The critical distinction is between probabilistic and deterministic execution. Compliance-grade agentic AI does not make probabilistic decisions on regulated transactions. Instead, it enforces deterministic policy rules at the point of execution, logs every decision with a complete audit trail, and escalates exceptions through predefined governance protocols. The agent reasons flexibly within boundaries; the boundaries themselves are fixed and auditable. This architecture is specifically designed to meet the requirements of SOX, GDPR, financial services regulation, and similar compliance frameworks.
Will agentic AI replace jobs in enterprise organisations?
Agentic AI changes the nature of enterprise work rather than simply eliminating it. Routine execution tasks — scheduling, routing, validating, documenting, chasing approvals — are absorbed by agents. Human roles evolve toward higher-value activities: setting organisational intent, designing the policies agents operate within, handling genuine exceptions that require contextual judgement, and owning outcomes.
In practice, this means the same team can manage significantly greater workflow volume, or redirect capacity toward more strategic work. Organisations should plan for workforce transition thoughtfully, with investment in reskilling toward governance, policy design, and AI oversight roles.
Where should an enterprise start with agentic AI deployment?
Start where the stakes are highest, the exceptions are most frequent, and the compliance requirements are most demanding. For most complex enterprises, this is procurement. It crosses organisational boundaries, involves multiple integrated systems, touches financial and regulatory risk at every step, and is dense with the kind of edge cases that stress-test agentic capabilities effectively. Mastery built in procurement — the policy models, validation logic, and exception-handling intelligence — transfers directly to adjacent domains including invoice processing, vendor risk management, HR onboarding, and IT service management.
Geraldine McBride is CEO of MyWave.ai, a native enterprise agentic AI company with market leading technology in governed and trusted AI execution for regulated industries.
A New Zealand and British citizen with three decades of global enterprise technology leadership with IBM and she served as President of SAP Asia Pacific & Japan and President and CEO of SAP North America. Geraldine has experience across more than 27 different complex industry value chains and how to apply Agentic AI to these to transform your business releasing new ROI.
MyWave.ai works with enterprise organisations in complex and regulated industries to design and deploy agentic operating models that deliver measurable ROI while meeting the governance standards that boards, auditors, and regulators require.