Most large organizations built an automation Center of Excellence (CoE) during the robotic process automation boom. The CoE set standards, managed bot licenses, ran a pipeline of automation ideas and reported hours saved. That model served its purpose. But it was designed for a world of deterministic bots that follow scripts, not for AI agents that interpret, decide and act, often embedded in business applications the CoE never touched.
As agents multiply across the enterprise, automation leaders need a new operating model, one that governs a mixed workforce of people, bots and agents across platforms and business units. This article presents a five-level maturity model for hyperautomation in the agentic era, describes the operating model capabilities at each level, and outlines a practical roadmap toward 2030.
Why the Classic CoE Model Is Under Strain
Several forces are pushing automation operating models beyond their original design. The first is scale. Gartner has predicted that by 2028 at least 15% of day-to-day work decisions will be made autonomously through agentic AI, up from virtually none in 2024, and that a third of enterprise software applications will include agentic AI, compared with less than 1% in 2024. Coverage of Gartner’s 2026 business orchestration research also reported a projection that an average Fortune 500 company will run more than 150,000 AI agents by 2028. Whatever the precise figure turns out to be, the direction is clear: the number of automated actors will far exceed what a central team can build or review one by one.
The second force is distribution. Agents arrive from many directions: embedded in CRM, ERP and service management platforms, built by business teams with low-code tools, and developed by engineering teams on managed agent platforms. A CoE that only governs what it builds will see a shrinking fraction of the automated work.
The third force is risk. Agents can take actions with financial, legal and customer consequences, and their behavior is less predictable than a script. The same coverage of Gartner research reported that only 13% of organizations consider their AI agent governance adequate. Gartner has also warned that more than 40% of agentic AI projects could be canceled by the end of 2027 because of cost, unclear value or inadequate risk controls.
The fourth force is platform consolidation. Gartner expects, according to that 2026 coverage, that about 70% of enterprises will use a consolidated automation platform by 2030, up from roughly 10% today. Process orchestration, RPA, integration, document processing, decision management and agent runtimes are converging, changing the tools that operating models are built around.
The Five-Level Hyperautomation Maturity Model
The model below describes how organizations typically progress. Levels are cumulative, and different business units may sit at different levels. The goal is not to reach level 5 everywhere, but to match maturity to the scale and risk of automation in each area.
Level 1: Opportunistic automation
Automation happens in pockets. Individual teams build bots, scripts or early AI assistants with little coordination. There is no central inventory, governance is informal, and benefits are anecdotal. This level is common at the start of any new technology wave, including the current agent wave, where enthusiastic teams experiment with agent frameworks and copilots.
Level 2: Centralized CoE
A central team owns automation standards, tools and a delivery pipeline. It builds most automations itself, manages licenses and reports hours saved. Quality improves and duplication falls, but the CoE becomes a bottleneck as demand grows, and its focus tends to remain on task-level RPA rather than end-to-end processes.
Level 3: Federated automation with process focus
The central team shifts from building to enabling. Business units have their own automation developers working on a shared platform with common standards. Process mining and process intelligence guide prioritization toward end-to-end outcomes. Orchestration platforms coordinate bots, APIs and human tasks. AI is used for document processing and classification within governed processes. Benefits are measured in cycle time, quality and cost per case, not just hours saved.
Level 4: Agentic orchestration with central governance
AI agents are embedded in orchestrated processes, with deterministic steps, agentic steps and human steps clearly defined. A central agent register covers every agent regardless of platform, recording owner, purpose, risk tier, data access and tools. Policies for identity, permissions, evaluation and monitoring are enforced through shared platform services rather than manual review. The central team operates as a platform and governance function; business units own their processes and agents.
Level 5: Agent operations at enterprise scale
Automation is managed like a workforce. Agents, bots and people are planned, deployed, monitored and retired with consistent lifecycle practices. Agent operations, sometimes called AgentOps, includes continuous evaluation, cost and performance management, incident response for agent failures, and regular reviews of autonomy levels. Process intelligence feeds improvement suggestions automatically. Leadership receives a single view of automated work, its value and its risk.
The question is no longer how many bots the CoE has built. It is whether the organization can see, govern and improve every automated actor, human-supervised or autonomous, across every platform.
Operating Model Capabilities by Dimension
Maturity shows up across six dimensions. Assessing each separately helps identify where to invest next.
- Strategy and value: from counting hours saved to managing a portfolio of process outcomes with clear owners and value tracking.
- Governance and risk: from informal approvals to risk-tiered policies, an enterprise agent register, autonomy levels per use case and audit-ready evidence.
- Platforms and architecture: from point tools to a consolidated orchestration platform with open protocols such as MCP and A2A, shared identity and observability.
- Delivery model: from a central build team to federated delivery with reusable components, templates and guardrails.
- People and skills: from RPA developers to process designers, agent engineers, evaluators, and business staff skilled at supervising agents.
- Operations: from bot monitoring to agent operations covering evaluation, cost, incidents and continuous improvement.
Key Roles in an Agentic Operating Model
New responsibilities emerge as organizations move toward levels 4 and 5. A process owner in the business is accountable for each end-to-end process, including the agents and bots within it. Agent owners are responsible for specific agents: their purpose, configuration, evaluation results and incidents. A platform team runs the orchestration, agent runtime, gateway and observability services. An automation governance function, often evolved from the CoE, maintains standards, the agent register and risk policies. Evaluators and quality engineers build and maintain test sets that measure agent behavior. And risk, compliance and security teams set boundaries for autonomy and review high-risk use cases.
Equally important is the role of frontline staff. As agents take on more work, employees increasingly review, approve, correct and coach agent outputs. Training people for this supervisory role, and designing work queues and interfaces that make it efficient, is a core operating model task rather than an afterthought.
Defining Autonomy Levels
A practical tool for governance is an explicit autonomy scale applied to each agent or agentic step. For example: the agent suggests and a human decides; the agent acts but a human approves before effect; the agent acts and a human reviews afterward by sampling; and the agent acts autonomously within limits, with exception-based oversight. Each use case is assigned a level based on risk, evidence of accuracy and regulatory context. Autonomy can be raised as evaluation results and production experience build confidence, and lowered quickly if incidents occur. Embedding these levels in process models, using gateways and approval steps, makes them enforceable rather than aspirational.
Metrics That Matter at Each Stage
Metrics should evolve with maturity. Early on, counting automations deployed and hours saved is reasonable. At level 3, shift to process metrics such as end-to-end cycle time, straight-through processing rate, cost per case and error rates. At levels 4 and 5, add agent-specific measures: task success rate against evaluation sets, human override and correction rates, cost per completed task, incidents per thousand actions, and the share of agents with current evaluations and owners. Governance health metrics, such as the proportion of agents in the register and the proportion with defined autonomy levels, signal whether control is keeping pace with growth.
Running a Maturity Assessment
A maturity assessment is most useful when it is quick, evidence-based and repeated. Run it per business unit or major process area rather than for the enterprise as a whole, because averages hide the gaps that matter. For each of the six dimensions, ask a small set of questions and require evidence rather than opinion: an exported inventory instead of an estimate of how many agents exist, a sample of evaluation reports instead of an assurance that agents are tested, a process dashboard instead of a description of how performance is tracked.
Score each dimension from one to five using the level descriptions above, and plot the results so leaders can see where one dimension lags the others. A common pattern is strong platform maturity paired with weak governance, which signals that technology has outpaced control. Another is good governance on paper with little operational evidence, which signals policies that are not enforced. Use the assessment to select two or three improvement priorities for the next two quarters, assign owners, and repeat the exercise every six months to track progress.
Hypothetical Example: A Global Manufacturer’s Journey
Consider a hypothetical global manufacturer whose RPA CoE built around 300 bots over six years, mostly in finance and procurement. In 2026, business units begin deploying agents embedded in their CRM and service platforms, and engineering teams build agents on a managed cloud platform. The CoE has visibility of none of these.
The company assesses itself at level 2 for RPA and level 1 for agents. It sets a two-year goal of reaching level 4 in finance, procurement and customer service. The CoE is renamed the automation and agent office and reorganized into a platform team and a governance team. It launches an enterprise agent register and requires every production agent, from any platform, to be registered with an owner and risk tier. It adopts a consolidated orchestration platform for end-to-end processes, using MCP and A2A connectors to coordinate agents from different vendors. Process mining identifies order-to-cash and procure-to-pay as priority processes for agentic redesign. Autonomy levels are defined for each agentic step, starting with human approval for any action affecting payments. In this illustration, the CoE’s value shifts from building bots to enabling safe scale.
A Roadmap Toward 2030
- Next six months: assess maturity by business unit; create an enterprise inventory of bots and agents; define risk tiers and autonomy levels; pick priority end-to-end processes.
- Six to eighteen months: consolidate onto an orchestration platform; implement an agent register and shared identity, gateway and observability services; federate delivery with standards and templates; redesign two or three priority processes with agentic orchestration.
- Eighteen to thirty-six months: establish agent operations practices, including continuous evaluation, cost management and incident response; scale agentic processes across business units; train staff for agent supervision; report automated work, value and risk to leadership regularly.
- Toward 2030: manage people, bots and agents as one workforce, with process intelligence driving continuous redesign and governance embedded in every platform.
Pitfalls to Avoid
- Keeping the CoE as a build factory. It will become a bottleneck and lose visibility as agents proliferate elsewhere.
- Governing only what you build. Embedded and business-built agents need the same register and policies.
- Measuring the wrong things. Hours saved does not capture process outcomes, risk or agent quality.
- Granting autonomy without evidence. Raise autonomy levels only when evaluation and production data support it.
- Neglecting people. Supervising agents is a new skill; without training and good tools, human oversight becomes a rubber stamp.
Conclusion
Hyperautomation is entering a new phase. The combination of AI agents, consolidated orchestration platforms and open protocols makes it possible to automate far more of the enterprise, but also demands an operating model that can govern automated work at a scale and pace the classic CoE never anticipated. Use the five-level maturity model to locate each business unit, invest in the dimensions that lag, define autonomy explicitly, and evolve the CoE from a build team into a platform and governance function. Organizations that make this shift will be ready to manage a workforce of people and agents with confidence by 2030.
How Hanoi Technology Can Help
Hanoi Technology helps organizations assess automation maturity and evolve their CoE into an agent-ready operating model, through consulting, dedicated teams or managed services.