04
The twelve institutional dimensions
The strategic backbone. The dimensions form a MECE starting framework — mutually exclusive, so each element belongs in only one category, and collectively exhaustive, so together they cover the full space. Any agent initiative can be documented or scored against the relevant dimensions.
01Strategic Alignment & PurposeMission fit, build/buy/partner/abstain, portfolio decisions, return-on-community.
- How agents advance institutional mission, strategic priorities, and student-success goals.
- Differentiation between student-facing, faculty/research, administrative/operational, and cross-functional agents.
- Decision criteria for “build, buy, partner, or abstain.”
- Portfolio management: which processes or decisions suit agentic automation versus those that must remain human-led.
- Alignment with existing institutional frameworks (Campus AI, HUMANS, AI Commons, agent/platform models, return-on-community metrics).
02Governance, Policy & OversightOwnership, risk tiers, escalation, kill switches, accountability.
- Institutional AI governance structures: who owns agent design, deployment, monitoring, and retirement.
- Agent-specific policies: authorization, scope of action, escalation, human-in-the-loop requirements, auditability, and kill-switch mechanisms.
- Risk classification of agents, from low-stakes information retrieval to high-stakes advising, financial, or academic decisions.
- Integration with broader AI policy, data governance, academic integrity, and equity policies.
- Multi-institutional or system-level coordination, especially for multi-campus systems.
- Accountability models when agents act, recommend, or initiate workflows.
03Technical Architecture & InfrastructurePatterns, integration, platforms, observability, security boundaries.
- Agent types and patterns: single-purpose, multi-agent systems, orchestrators, tool-using agents, memory/persistence models.
- Integration with core systems of record (SIS, LMS, ERP, identity, data warehouse, research systems).
- Data access, retrieval-augmented generation (RAG), knowledge bases, and real-time vs. batch operation.
- Platform choices: commercial agent platforms, open-source frameworks, custom development, or hybrid.
- Security, identity, authentication, authorization, and least-privilege design for agents.
- Observability, logging, tracing, versioning, and rollback.
- Scalability, latency, reliability, and cost at institutional volume.
04Data, Privacy, Security & ComplianceClassification, FERPA, consent, audit trails, prompt injection, least privilege.
- Data classification and minimization for agent use (student, employee, research, operational data).
- FERPA, state privacy laws, research data protections, and vendor contractual requirements.
- Consent, transparency, and notice when agents interact with students or staff.
- Model training, fine-tuning, and retention of interaction data.
- Security posture: prompt injection, data exfiltration, privilege escalation, and agent-to-agent trust boundaries.
- Audit trails sufficient for compliance, incident response, and academic or employment investigations.
05Use Cases & Functional DomainsStudent success, teaching/learning, faculty/research, admin/ops, cross-cutting.
- Student success and support: advising, early alerts, financial aid guidance, belonging/community navigation, career exploration.
- Teaching and learning: tutoring, formative feedback, study planning, accessibility support, curriculum assistance — with clear academic-integrity boundaries.
- Faculty and research: literature synthesis, grant writing support, lab or research workflow agents, administrative assistance.
- Administrative and operational: HR, finance, procurement, facilities, enrollment management, IT service management.
- Cross-cutting: knowledge management, policy navigation, scheduling, reporting and analytics agents.
- Explicitly separate high-value, low-risk pilots from high-risk domains that require stronger controls.
06Equity, Inclusion, Access & BiasDifferential impacts, bias testing, accessibility, human flourishing.
- Differential impacts on first-generation, underrepresented, disabled, international, and working students.
- Bias testing across demographic groups and decision domains.
- Accessibility: WCAG alignment, multimodal interaction, language support.
- Risk of agents reinforcing or creating new barriers, such as over-reliance that reduces human support capacity.
- Design principles that center human flourishing and belonging rather than pure efficiency.
- Monitoring for disparate outcomes, with remediation processes.
07Human Roles, Change & WorkforceNew roles, literacy, human-in/on-the-loop, change management.
- New and evolved roles: agent designers, orchestrators, evaluators, ethics reviewers, domain supervisors.
- Skills and literacy for faculty, staff, and students who design, supervise, or interact with agents.
- Change-management approaches (e.g., ADKAR) and communication strategies.
- Human-in-the-loop and human-on-the-loop models appropriate to risk level.
- Impact on existing roles, workload redistribution, and potential job redesign or displacement.
- Faculty and staff readiness, incentives, and support structures.
08Pedagogy, Integrity & Learning OutcomesAcademic boundaries, scaffolding vs. substitution, faculty authority.
- Boundaries between legitimate agent use and academic misconduct.
- Effects on learning, critical thinking, and skill development — scaffolding vs. substitution.
- Faculty authority and control over agent use in courses.
- Assessment redesign implications.
- Student agency and literacy in directing or critically evaluating agent outputs.
09Evaluation, Metrics & Continuous ImprovementOutcomes, technical performance, process metrics, sunset criteria.
- Success metrics beyond cost or efficiency: student outcomes, equity gaps, faculty/staff satisfaction, trust, return on community.
- Technical performance: accuracy, hallucination rate, task completion, latency.
- Process metrics: adoption, escalation rates, human override frequency.
- Methods: pilot design, A/B testing where appropriate, qualitative feedback, longitudinal outcome tracking.
- Feedback loops for prompt, tool, and policy refinement.
- Sunset criteria for underperforming or high-risk agents.
10Financial, Vendor & SustainabilityTCO, lock-in, funding, environmental cost, shared services.
- Total cost of ownership: platform, integration, data, monitoring, human oversight, training.
- Build vs. buy vs. consortium models.
- Vendor lock-in, data portability, and exit strategies.
- Funding sources and multi-year sustainability.
- Environmental and compute-cost considerations at scale.
- Shared services or system-level purchasing power.
11Legal, Ethical & LiabilityIP, liability, transparency, ethical review, regulatory alignment.
- Intellectual property in agent outputs and training data.
- Liability when agents give incorrect or harmful advice or take actions.
- Transparency and explainability appropriate to context.
- Ethical review processes, especially for student-facing or high-stakes agents.
- Alignment with emerging regulations, standards, and higher-ed-specific guidance.
12Foresight, Scalability & EcosystemMulti-agent evolution, interoperability, AI Commons, scenario planning.
- Evolution from current generative tools to more autonomous multi-agent systems.
- Interoperability across institutions or within a system.
- Open standards, shared agent registries, or AI Commons approaches.
- Scenario planning for capability jumps, regulatory shifts, or major vendor changes.
- A research and innovation agenda the institution itself contributes to.