Laks Alagappan, VP & Client Engagement Manager at Genpact, leads enterprise technology delivery for Tier 1 banks across the UK and Europe.
Enterprise IT organizations have developed new capabilities with every technology shift. From cybersecurity and data management to cloud engineering and production support, leaders have had to establish new disciplines with each shift.
Agentic AI may require the next discipline.
As enterprises move from AI experiments to growing portfolios of production agents, a question will emerge: Who will manage this expanding estate?
The answer may lie in a new federated capability: agent management. This discipline focuses on how models and agents are orchestrated, contextualized, governed, monitored and managed throughout their lifecycle. Anchored within the AI-native organization, agent management could involve responsibilities across business, technology, data and control functions.
Agent management could bridge the gap between AI governance and operational execution, translating policies and standards into the day-to-day management of AI agents across the enterprise.
Why Agents Require Dedicated Management
Traditional enterprise IT is organized around people and the technology assets and services they design, build and operate: applications, infrastructure, networks and data platforms.
Agents introduce a different management challenge. Organizations must govern the decisions agents make and the actions they take, alongside the software itself. As agents interpret objectives, formulate plans, retrieve information and invoke applications, they need clear authority limits, traceable decisions and defined points for human intervention.
An incident-management agent, for example, could analyze telemetry, consult previous incidents, identify likely causes and initiate remediation within predefined controls, escalating higher-impact actions for human approval. Many enterprises are delegating decision making and execution authority to software, subject to explicit boundaries and oversight.
At scale, managing that authority cannot remain an informal responsibility spread across architecture, engineering, data, cybersecurity and individual programs. It requires an enterprise capability with clear ownership and accountability.
Six Core Capabilities
Agent management involves six key components:
1. Model And Agent Orchestration: Enterprises need mechanisms to select models, route workloads, coordinate agents and manage interactions with enterprise systems. These mechanisms must handle failed actions, conflicting instructions and escalation to humans, as well as when agents delegate work to one another.
2. Context Management: Agents need business knowledge, which should include policies, procedures, product definitions, historical decisions and operational data. Context management determines what information an agent receives, its provenance, freshness and quality and whether the agent is authorized to use it. Data governance and lineage are operational foundations for agent behavior.
3. Agent Identity And Access: Every production agent should have an identity, an accountable owner, a defined purpose and controlled credentials. Permissions should follow least-privilege principles, with clear boundaries for delegated access, periodic reviews and revocation when an agent is suspended or retired.
4. Governance And Autonomy: An autonomy framework should distinguish between agents that retrieve information, recommend actions and execute decisions. It must define execution limits, human approvals, prohibited actions and escalation requirements. Authority should reflect the consequences of an error and the reversibility of an action.
5. AgentOps And Evaluation: Beyond availability and latency, agents need to be evaluated on task success, accuracy, policy compliance and human-intervention rates. Changes to models, prompts, tools or context should trigger proportionate testing. Teams need visibility into agent behavior to investigate failures, roll back changes and stop unsafe execution.
6. Lifecycle And Portfolio Management: An agent registry should record ownership, purpose, dependencies, risk classification, permissions, performance, costs and lifecycle status. Portfolio reviews can identify duplication, excessive access and underperforming agents, informing decisions to improve, consolidate or retire them.
Turning Capabilities Into Enterprise Workstreams
Agent management can provide foundations and operational oversight, while technology and business leaders can remain accountable for delivery and outcomes. The six capabilities above should underpin six workstreams for enterprise leaders:
1. AI Platform And Agent Enablement: Leaders must provide approved models, agent components, orchestration services and enterprise integrations. They can combine orchestration, context management and lifecycle controls to help teams build on common foundations.
2. Agent Assurance And Control: Leaders should connect agent identity, permissions, security testing, evaluations and autonomy boundaries. They can apply identity, governance and AgentOps capabilities to establish what agents may do and verify that they operate within those limits.
3. AI-Enabled Software Engineering: Leaders should embed AI across requirements, development, testing and delivery. They can use enterprise context, controlled access and evaluations to improve delivery while maintaining engineering quality and accountability.
4. AI-Enabled Technology Operations: Leaders must coordinate AI adoption across production and infrastructure operations, covering incident detection, diagnosis, remediation, provisioning and capacity optimization. From there, apply operational context, orchestration and AgentOps, with approval and rollback controls proportionate to potential impact.
5. Business Workflow Transformation: Leaders should extend agents into processes such as client onboarding, reconciliation and regulatory reporting. Combine business context, orchestration and autonomy controls with defined human approval points and clear accountability across the workflow.
6. AI Economics And Workforce Transformation: Finally, leaders should assess business outcomes against model, token, infrastructure and human-review costs, drawing on AgentOps and portfolio data. Establish baselines before deployment, redesign roles and develop skills as needed and define how people will supervise agents.
A Federated Operating Model
Agent management does not need to become a large central department.
Within a CIO-led organization, the CIO can own shared platforms, orchestration services, the agent registry, evaluation tooling and operational standards. Business and technology domains can then build and operate agents close to the processes they understand.
A CDO could retain accountability for data governance, while cybersecurity could lead security policy and controls, and risk and compliance could provide oversight. Agent management would then embed these requirements into the agent lifecycle, with explicit decision rights and escalation paths.
People, Technology And Agents
As agents take on more execution pathways, humans will remain responsible for orchestration, judgment, exception management and accountability.
Informally distributed agent management responsibilities may work during experimentation, but thousands of agents interacting with applications, data, employees, customers and one another require a more deliberate operating model. CIOs need answers: Which agents are operating? What can they access and execute? How is their performance evaluated? Who can stop them? Who remains accountable?
Previous technology shifts established many new enterprise disciplines. Agent management could become the next: a capability that makes delegated authority visible, controlled and accountable, enabling enterprises to sustain agentic AI at scale.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?