Project Governance Agentic Project Governance The Execution-to-Mandate Gap in AI-Enabled Project Work Project governanceAgentic AIGeneration-time controlProcess evidence AI4 2026 poster available Key proposition When AI becomes the executor, the path must be governed. A polished output can still be procedurally defective. This paper examines how agentic AI changes project governance when execution shifts from human task performance to machine-mediated work. It argues that a polished output can still be procedurally defective, and develops a framework for keeping the operating mandate influential while work is carried out.
Governed Execution Research
When AI is the executor, its path must be governed.
Agentic AI changes more than workflows. It changes how work is authorized, how attention is allocated, where execution occurs, who remains accountable, and how firms create advantage.
A polished output can still be procedurally defective.
The management question
What changes when AI moves from aiding to doing work?
Once AI systems search, compare, classify, call tools, route tasks, prepare recommendations, and carry out actions, they become participants in executing work. Management changes because the unit of control changes.
Governed Execution at AI4 2026
Five papers.
Five managerial problems.
One research program.
Prepared for the Applied AI Research Conference at AI4 2026, the poster series translates each paper into a managerial question, a visual framework, and a practical implication.
Delegation & Organizational Control Delegation Without Preferences Organizational Control When the Executor Cannot Care DelegationOrganizational controlMachine executionResponsibility AI4 2026 poster available Key proposition AI can optimize without preferences. Humans still bear the consequences. Optimization can be delegated. Responsibility cannot. This paper asks how delegation changes when the executor has no preferences, incentives, role morality, or capacity to accept responsibility. It shows why managerial control must shift from aligning the agent’s interests to specifying the mandate, preserving process fidelity, and retaining human interpretive responsibility.
Work & Organization Design The Workflow Is Not the Work Reconstructing Work Under Delegated Machine Execution Work designWorkflowProcess fidelityEvidence of how AI4 2026 poster available Key proposition The workflow is not the work. Redesign the work for the executor. Preserve the safeguards. Redesign the path. This paper distinguishes the formal workflow from the work actually performed. As AI systems execute tasks across tools and intermediate decisions, workflow completion no longer establishes process fidelity. The paper explains why governance requires evidence of how the work developed, not only confirmation that the workflow ended.
Theory of the Firm Delegated Machine Execution and the Theory of the Firm Process Evidence, Coordination, and Organizational Boundaries Theory of the firmCoordinationFirm boundariesEvidence specificity AI4 2026 poster available Key proposition AI lowers the cost of doing work. The cost of process proof rises. The work may be cheaper to produce, but harder to verify. This paper revisits the theory of the firm under delegated machine execution. It argues that process fidelity and evidence specificity become economically consequential when decision authority, task execution, and accountability can be separated across organizations, models, and providers.
Strategy & Competitive Advantage Modular Execution, Sticky Accountability Firm Scope and Competitive Advantage Under Agentic AI Competitive advantageFirm scopeModularityAccountability AI4 2026 poster available Key proposition Execution can move. Accountability sticks. You can outsource the work, but not the responsibility. This paper studies how agentic AI makes execution more modular while leaving accountability comparatively sticky. It argues that advantage depends not merely on adopting capable models, but on controlling the interfaces that specify, govern, and evidence delegated work.
AI4 2026 Poster Collection
Five perspectives on what AI execution changes.
The conference series presents the central managerial problem, conceptual framework, key terms, and practical implications of each paper.
Shared framework
Move control closer to the work as it develops.
The framework connects organizational intent to generation-time control, evidence, and focused managerial review.
- 01Define
Operating mandate
Objectives, constraints, evidence requirements, and escalation rules.
- 02Encode
Semantic contract
A machine-readable representation of the mandate.
- 03Control
Generation-time control
Evaluate alternatives while the execution path can still change.
- 04Evidence
Evidence of how
A structured record of the path, constraints, and control decisions.
- 05Review
Targeted review
Direct managerial attention to exceptions, risk, and unresolved judgment.
The management shift: review moves from checking every output to examining exceptions and evidence from the execution path.
Managerial implications
What changes for managers.
Project governance
Control moves from periodic review toward continuous evidence of execution-to-mandate fidelity.
Delegation and organizational control
Authority can be delegated to systems that cannot accept incentives, preferences, or responsibility.
Work and organization design
The formal workflow becomes an incomplete representation of the work actually performed.
Firm boundaries and accountability
Execution becomes mobile across providers and models while accountability remains organizationally sticky.
Strategy and competitive advantage
Reliable execution and the ability to scale managerial attention become sources of competitive advantage.
Governed Execution
Explore what AI execution changes.
Request abstracts, conference materials, or a conversation about the managerial research program.