Define roles, responsibilities, and decision rights

Clear roles turn agentic transformation from a set of willing volunteers into an accountable operating model. When you name who owns what, decisions move faster, work stops falling through the cracks, and every agent has a person answerable for its value and its risk. This article describes the roles typically involved in agentic transformation, how those roles shift across patterns, the decision rights that separate what the Center of Excellence (CoE) owns from what each domain owns, and how to bring the same clarity to tasks with a sample RACI (Responsible, Accountable, Consulted, and Informed) matrix.

Why role clarity matters

Role clarity is the foundation of an organization that governs, enables, optimizes, and scales agents. When roles are explicit, you gain:

• Accountability. Every agent and every decision has a named owner. Nothing sits in limbo.
• Efficiency. People know who to ask, which removes rework and shortens cycle time.
• Collaboration. Business, platform, and risk teams work from a shared model instead of competing for control.
• Consistency. Standards apply the same way across domains so quality does not depend on who built the agent.
• Governance. Risk reviews, guardrails, and approvals attach to roles rather than to individuals who happen to be available.
• Empowerment. People who understand their role and their guardrails act with confidence instead of waiting for permission.

Scaling agents means balancing the opportunity to move fast against the risk of moving carelessly. That balance rests on three kinds of clarity:

• Goal clarity. Everyone understands what agentic transformation should achieve. Define this first in Define your vision, metrics, and goals and run a short workshop with stakeholders to agree on the goals and the risks to manage.
• Role clarity. Everyone knows their part and who to turn to. Review Roles typically involved in agentic transformation.
• Task clarity. Everyone knows the specific work to do and the standard it must meet. Review Bring clarity to tasks.

Roles typically involved in agentic transformation

Agentic transformation involves more people than the CoE alone. This section describes typical roles, grouped by what they focus on. You won't need all of them on day one, and early on one person often holds several roles. As your portfolio grows and the work justifies dedicated capacity, add roles and split responsibilities. Some roles anchor in the CoE, and others live in the business domains where agents are built and run.

Strategy and leadership

• Executive sponsor: Sets the ambition and guardrails, secures funding, removes organizational blockers, and champions agents visibly across leadership.
• Business owner: Owns the value, adoption, and key performance indicators (KPIs) for a business domain, and decides whether to scale, stop, or iterate on each agent.
• CoE lead or AI program manager: Runs intake, prioritization, and the operating rhythm, and is the single point of accountability for how the organization scales agents.
• Agent product owner: Translates strategy into a prioritized agent portfolio and roadmap, and balances demand against capacity.

Governance, security, risk, and responsible AI

• Security, risk, and compliance lead: Owns risk tiering, agent identity and access standards, and the reviews that higher-risk agents must pass before release.
• Responsible AI lead: Owns responsible AI review gates, bias and transparency checks, and AI disclosure, and connects to the enterprise responsible AI council.
• Data governance and stewardship: Owns data quality and classification, and the permissions and sensitivity labels that keep agent grounding accurate and safe.
• Privacy and compliance: Ensures agents meet regulatory and internal policy requirements, with auditable controls.

Platform, architecture, and operations

• Platform and operations: Own environments, monitoring, support, and incident response so agents stay reliable in production.
• Enterprise and solution architects: Set architecture and integration standards, and decide which agent-building surface to use and when.
• Service and platform admins: Administer the underlying services, such as Microsoft 365, Copilot Studio, and Microsoft Foundry, including licensing, environments, and configuration.

Build and knowledge

• Makers: Business users, from citizen makers to advanced makers, who build and configure agents within the guardrails and golden paths the CoE provides.
• Pro-code developers: Engineers who build custom, pro-code agents for scenarios that exceed low-code, on platforms such as Microsoft Foundry.
• Domain experts: Own the quality, accuracy, and currency of the knowledge that agents rely on, and validate that answers are correct.

Adoption and enablement

• Adoption leads: Drive behavior change, training, and communication so people actually use the agents.
• Champions: Enthusiastic users who advocate for agents, support their peers, and feed insight back to the CoE.

Service owners and end users

• Service owners: Own end-to-end service reliability and service-level agreements for agents in production use, and manage escalations.
• End users: The people who use agents in their daily work. Their adoption and feedback are the ultimate measure of whether an agent delivers.

How roles shift across the transformation patterns

The roles don't change from pattern to pattern. Their weight does. The same title has different job responsibilities depending on the pattern, and each role's level of involvement shifts too. Set these expectations before you build because they determine who owns each agent and how much oversight each agent needs.

The matrix shows how the six core roles change job across the six patterns.

Two things shift as you move from left to right, from assistive patterns toward patterns where agents execute:

• Weight moves to governance and risk. The executive sponsor, security and risk, and platform teams move from light involvement in Employee AI enablement to heavy involvement in Core business process, External engagement, and AI-first capabilities, where an agent's actions carry profit-and-loss, brand, or safety consequences.
• Ownership moves into the business. The business owner goes from not applicable, where individuals help themselves, to a named process or product owner accountable for the outcome. The adoption lead focuses most on Employee AI enablement, where success depends almost entirely on behavior change, and stays active, though lighter, everywhere else.

The direction is consistent: the more an agent executes, the more each role moves from doing to governing, and the more formal and business-owned the accountability becomes. Review Choose your CoE structure and operating model to choose the CoE structure that fits each pattern.

Decision rights

Ambiguity about who decides is where most CoEs slow down. Make the split explicit: the CoE owns the standards and the guardrails, and domains own the choices inside them.

The CoE owns:

• Platform and environment strategy
• Security and compliance policies
• Architecture and integration standards
• Release-readiness gates
• Monitoring and observability standards
• Risk tiers and the criteria that assign them
• Autonomy limits for agents
• Responsible AI guidelines

Domains own:

• Prioritization of agents within their domain
• Agent design within the CoE standards
• Knowledge curation and quality
• Day-to-day operation of lower-risk agents
• User experience for their audience
• Domain-level KPIs and value tracking
• Continuous improvement of their agents

Bring clarity to tasks

Roles say who. Tasks say what. Break your agentic goals into concrete tasks, assign each task to a role, and define the standard it must meet. For agents, pay particular attention to the tasks that manage risk:

• Assigning a risk tier and the reviews it requires
• Conducting security and responsible AI reviews before release
• Setting autonomy limits for what an agent can do without a human
• Building and running agent evaluations to catch drift
• Onboarding, change management, and handover from build to run

The following example groups common agentic transformation tasks by area. Use it as a starting point, not a complete list. Tasks vary by your organization's size and adoption maturity.

Prioritize tasks across horizons

You can't do everything at once. Sort tasks into horizons so you address what's urgent while planning for what comes next:

• Horizon 1, now. Tasks for the next few weeks: Name owners, set minimum guardrails, and ship the first one or two agents.
• Horizon 2, next. Tasks for the coming months: Build golden paths, formalize reviews, and expand to more teams.
• Horizon 3, later. Strategic tasks: Scale across the estate, mature governance, and pursue more advanced patterns.

Make ownership explicit with a RACI matrix

To remove any remaining doubt, map your key tasks to a RACI matrix that marks who is Responsible, Accountable, Consulted, and Informed for each task.

• Responsible (R) does the work.
• Accountable (A) is ultimately answerable for the outcome. Assign exactly one accountable role per task. If two roles are accountable, no one is, and the task stalls the moment they disagree.
• Consulted (C) provides input before the work is done.
• Informed (I) is kept up to date on progress and outcomes.

In your matrix, list tasks in the first column and roles in the top row, then enter a value in each cell. The following sample covers common agent lifecycle tasks. Adapt it to your organization, and revisit it as roles mature so it reflects how the CoE works today, not its launch state.

Next step

After you define roles and decision rights, put governance into practice by classifying agents according to their risk level.