Human-First AI Leverage
A Smith School Resource for Curriculum and Professional Practice
Core position
Smith’s position is Human-First AI Leverage: we teach students and professionals to know what to amplify, when to withhold, and how to govern the amplifier. AI is a source of leverage, not a replacement for human judgment.
Prepared from the AI Leverage through Literacy framework, the evidence review on AI leverage in business education, and the Smith School’s emerging AI portfolio.
How to use this resource
This is a living document. It sets out Smith’s position on AI in business education. The principles are shared across the School; faculty decide how best to teach them in their disciplines. Faculty can adapt the suggested assignments and assessments to their courses. Existing University and program requirements still apply [1]. Through their area chairs, faculty can share an assignment they tried, what their students learned, and what they would change. Those experiences will guide future revisions.
Resource element | Purpose |
|---|---|
Point of view | State Smith’s human-first position on AI as leverage rather than substitution. |
AI leverage equation | Give faculty and students a shared mental model for value creation and risk. |
Competency layers | Define what learners should know and be able to do at Foundational, Intermediate, and Advanced levels. |
Curriculum map | Translate the framework into undergraduate, MBA, master’s, executive, and non-degree contexts. |
Assessment and governance | Move from tool use to defensible AI-assisted work. |
Smith’s AI Point of View
The Smith School’s AI position is that AI is a source of leverage, not a replacement for human judgment. It raises the returns to business education because it amplifies the base of knowledge, analytical discipline, ethical reasoning, communication skill, and managerial judgment that learners bring to the work. That is why the central curriculum question is not whether students will use AI. They already will. The question is whether they can use AI well, responsibly, and defensibly.
Human-first does not mean AI-averse. It means that people remain responsible for the tasks AI helps perform and for the consequences of AI-assisted decisions. In practice, Smith’s position can be expressed as a simple standard: know what to amplify, know when to withhold, and know how to govern the amplifier.
What makes this position distinctive
- Optimistic but disciplined. AI can expand productivity, creativity, and access, but only when task fit, verification, and governance are built into the workflow.
- Business-first, not tool-first. The starting point is the business problem, stakeholder need, decision context, and risk profile, not the model or platform.
- Human capability as the base. The most valuable AI users are not merely fast prompt writers. They are people who know what good work looks like and can improve, challenge, or reject AI outputs. • Governance as an enabler. Responsible AI is not a compliance appendix. It is the operating system that makes AI leverage scalable and trustworthy.
Smith Proof Points
This position aligns with what Smith is already building. The Smith School teaches generative and contextual AI across undergraduate and graduate programs, including bias mitigation, data provenance, workflow design, and human capabilities such as empathy, creativity, judgment, and leadership. The Center for Artificial Intelligence in Business is focused on intentional AI-enabled design and governance frameworks centered on human judgment and creativity [2].
Smith asset | How it supports Human-First AI Leverage |
|---|---|
AI @ Smith | Publicly frames AI as a source of innovation and productivity while emphasizing strengths, weaknesses, dangers, and opportunities. |
Center for Artificial Intelligence in Business | Provides a research and outreach platform for intentional design, safe innovation, governance, and human-centered AI. |
AI and Career Empowerment certificate | Surpassed 50,000 learners in less than a year; uses 10 self-paced modules across AI fundamentals, organizational transformation, functional applications, responsible AI, entrepreneurship, negotiation, and job search [3]. |
AI in key MS programs | Renamed four graduate programs around Business Analytics & AI and Information Systems & AI while emphasizing human ingenuity, creativity, curiosity, and value creation [4]. |
Agentic AI Challenge | Engaged 330 students across campus and 67 teams working on industry problems with autonomous AI agents, judged on real-world outcomes, reliability, and safety [5]. |
Small-business AI outreach | A $1.031 million federal grant supports tools, training, research, online courses, and workshops to help Maryland small businesses adopt AI [6]. |
Taken together, these activities should be treated as one institutional portfolio rather than isolated initiatives. The resource below gives that portfolio a shared language for curriculum, faculty development, executive education, external partnerships, and student assessment.
The AI Leverage Equation
Leverage is familiar in business. Financial leverage uses borrowed capital to amplify returns on invested equity. Operational leverage uses fixed-cost infrastructure to amplify the effect of revenue growth on operating profit. In each case, leverage is powerful, but conditional. It is applied to a base, it can magnify losses as well as gains, and it carries costs that must be managed.
AI works the same way. It does not create high-quality judgment from nothing. It multiplies the knowledge, context, data, process design, and intent that people bring to a task. A strong human base can produce disproportionate value when paired with the right AI tools and workflows. A weak base can produce polished but shallow, biased, or wrong work at scale. Verification is the interest expense on AI leverage: checking an internal draft may take a minute, while a recommendation that commits the firm’s money warrants closer review.
The AI leverage equation
Conceptually, Professional Value = Human Base x AI Multiplier, constrained by Task Fit, Verification, and Governance. Its purpose is to help students and professionals reason about when AI will create value, when it will create risk, and what human capabilities must be strengthened first.
How to Read the Equation
- Human Base: domain knowledge, analytical reasoning, ethical judgment, communication, leadership, stakeholder awareness, and professional standards.
- AI Multiplier: the models, agents, tools, data systems, automation patterns, and collaboration workflows that can increase speed, breadth, quality, or learning.
- Task Fit: whether the task is bounded, pattern-rich, evaluable, and inside the current capability frontier of the tool.
- Verification: the process of checking facts, assumptions, sources, calculations, reasoning, and business implications.
- Governance: the policies, risk tiers, disclosure expectations, data rules, human review gates, and accountability structures around AI use.
The evidence supports a conditional version of this claim. Generative AI can raise productivity and quality on bounded writing, customer-support, coding, consulting, and innovation tasks, but performance can fall when the task is outside the model’s current frontier or when people over-trust fluent outputs [7–10]. This makes the leverage equation especially useful for teaching: it encourages students to ask not “Can AI do this?” but “Should AI be used here, and under what controls?”
Using AI to Build Foundational Knowledge
AI can also help students build the human base. A novice can use a chatbot grounded in course materials as a coach, asking for a hint when they get stuck or another way to think about a difficult idea. An ordinary chat can be useful here, too. A student might explain how they reached an answer, ask AI for a counterexample, and check it against the course material before deciding whether their reasoning holds.
The risk is that students may follow a clear explanation and assume they could solve a similar problem on their own. Asking them to predict the next step before they get feedback, or to try a related problem with the chat closed, gives them a way to test that assumption. In two introductory physics lessons, for example, an AI tutor designed for the course gave students feedback as they worked through each part of a problem. On the quizzes that followed, students who used the tutor scored higher than classmates who worked through the same material in an active learning class [11].
The Three Competency Layers
The Smith curriculum should organize AI capability into three competency layers: Foundational, Intermediate, and Advanced. These are not rigid course levels. They are learning outcomes that can be adapted for undergraduates, MBA students, specialty master’s students, executives, faculty, and professionals. The layers progress by consequence and judgment: safe use, reliable workflow integration, and organizational transformation.
Layer | Learners should be able to... | Representative assignments and deliverables | Evidence of mastery |
|---|---|---|---|
Foundational | Understand what AI is and is not; recognize hallucinations and bias; protect data; use basic prompting; evaluate whether output is accurate, relevant, and safe. | AI-output critique; hallucination diagnosis; source-verification memo; data-sensitivity exercise; short no-AI concept checks. | Identifies flawed outputs, explains what must be verified, refuses inappropriate use, protects confidential data, and discloses material AI assistance. |
Intermediate | Integrate AI into real business workflows; design prompts and quality-control checks; apply AI within a domain; estimate value; document process. | AI-assisted market analysis with verification; finance model support workflow; campaign variant testing; supply chain scenario analysis; weekly reporting redesign. | Designs a repeatable human-AI workflow, documents prompts and verification, improves output quality, explains business value, and preserves accountability. |
Advanced | Evaluate tools; design governance; understand agents, automation, RAG, and fine-tuning at a managerial level; connect AI to strategy, organization design, and risk. | AI adoption plan; vendor evaluation; governance proposal; agent prototype; executive decision simulation; risk-tiered use-case portfolio. | Defends tool choices, governance controls, adoption logic, risk tradeoffs, and stakeholder implications under scrutiny. |
What Each Layer Should Teach
The Foundational layer gives every learner a baseline for safe and responsible AI use. It includes basic data literacy, model limitations, prompting, hallucination recognition, privacy, intellectual property, bias, and ethics. The Intermediate layer is where literacy becomes leverage: learners design human-AI workflows, use structured prompting, apply quality-control protocols, and articulate business value. The Advanced layer prepares learners to shape AI use beyond their own work through tool selection, governance, strategic transformation, organizational design, and accountable automation.
The progression should also protect learning. Education research suggests that generative AI can improve immediate task performance without creating durable learning when students use it as an answer engine rather than a scaffold for reasoning [12]. For that reason, each layer should assess process, not just output: how the learner framed the task, what the learner asked AI to do, what was verified, what was rejected, and what the learner remains accountable for.
Curriculum Mapping
The same AI leverage framework should appear across Smith programs, but the depth, examples, and assessment should differ by audience. A first-year undergraduate, MBA student, specialty master’s student, executive, and non-degree learner all need AI literacy. They do not need the same version of it.
A practical curriculum architecture should include four cross-program threads: productivity and workflow design; judgment and verification; responsible AI governance; and human comparative advantage. The last thread is essential. Business graduates need to know not only how AI can help, but what humans must still contribute: problem framing, domain critique, context, stakeholder sensitivity, communication, and final accountability.
Audience | AI leverage emphasis | Example course or experience | Evidence of mastery |
|---|---|---|---|
Undergraduate | Build the human base early: business fundamentals, analytical reasoning, communication, ethics, and foundational AI literacy. | Introductory AI literacy module; AI-output critique in business communication; AI assisted case preparation with source verification; no-AI checks on core concepts. | Student can explain AI limits, verify claims, disclose use, and defend reasoning without relying on AI-generated language. |
MBA | Connect AI leverage to leadership, strategy, transformation, organizational design, and risk management. | Leadership simulation with AI-generated recommendations; strategy case using AI for scenario generation; governance plan for a business unit. | Student can evaluate AI advice in ambiguous managerial contexts and defend a decision under stakeholder scrutiny. |
Specialty master’s | Pair functional depth with domain-specific AI workflows, technical fluency, and professional standards. | AI-supported forecasting in finance; marketing segmentation; operations optimization; accounting anomaly detection; IS agent prototype. | Student can design and document a field specific workflow with quality controls, risk classification, and business-value logic. |
Executive education | Role-based, immediately applicable capability for individual productivity, team enablement, governance, and transformation. | Half-day productivity clinic; manager workshop on team workflows; executive session on AI strategy and governance. | Participant leaves with a use-case portfolio, risk tiers, measurement plan, and next-step implementation checklist. |
Non-degree and certificates | Scalable baseline access for professionals navigating workforce change and career transitions. | Self-paced AI fundamentals; functional applications; responsible AI; career empowerment; negotiation and entrepreneurship in the age of AI. | Learner can describe AI’s business impact, identify personal use cases, and apply responsible use principles to career and workplace tasks. |
Concrete Assignment Examples
- AI critique assignment: Students receive a polished AI-generated market analysis containing hallucinated sources, questionable assumptions, and hidden bias. They annotate the output, verify claims, and rewrite the recommendation.
- Workflow redesign assignment: Students redesign a recurring business process, such as weekly sales reporting or supplier-risk monitoring, with AI assistance and human review gates.
- Functional AI lab: Students apply AI to a discipline-specific problem, such as variance explanation in accounting, demand forecasting in supply chain, customer sentiment synthesis in marketing, or earnings-call analysis in finance.
- Executive decision simulation: Students receive conflicting AI-generated recommendations and must decide what to accept, reject, verify, or escalate.
- Capstone defense: Teams present an AI-assisted business recommendation and defend their process, source checks, risk controls, and human accountability in oral questioning.
Role and Function Adaptation
The three layers remain consistent, but they should be adapted to the consequences each role bears. Individual contributors and early-career professionals need hands-on mastery of Foundational and Intermediate competencies because they produce AI-assisted work every day. Managers need enough hands-on understanding to guide teams and evaluate quality, plus stronger capability in workflow design, tool selection, value articulation, and risk assessment. Executives need deep Advanced capability in strategy, governance, organization design, and stakeholder accountability, while maintaining enough Foundational skill to challenge AI claims throughout the organization.
Role | Depth needed | Primary teaching emphasis |
|---|---|---|
Individual contributors / early career professionals | Deep Foundational and Intermediate; lighter Advanced. | Hands-on AI use, output evaluation, documentation, quality control, and function-specific workflows. |
Managers | Solid Foundational; strong Intermediate; meaningful Advanced. | Team enablement, workflow design, use-case selection, risk tiering, quality assurance, and value measurement. |
Executives | Strong Foundational for evaluation; lighter Intermediate; deep Advanced. | AI strategy, governance, investment decisions, organizational transformation, stakeholder impact, and accountability. |
Functional adaptation is equally important. A finance student and a marketing student both need prompt design, data literacy, and critical evaluation, but the examples and professional risks differ. The table below gives faculty and program leaders a starting point for localizing the framework.
Function | AI leverage examples | Common risks to teach |
|---|---|---|
Finance | Financial modeling support; forecasting; risk analysis; earnings call synthesis; fraud detection. | False precision, unverifiable assumptions, confidential data exposure, model risk, regulatory implications. |
Marketing | Campaign ideation; segmentation; customer sentiment analysis; copy variants; market-research synthesis. | Bias in personas, hallucinated customer facts, brand inconsistency, privacy and consent issues. |
Operations | Process optimization; scheduling; quality-control analysis; resource allocation; bottleneck diagnosis. | Over-automation, hidden edge cases, poor data quality, local optimization at system expense. |
Supply chain | Demand forecasting; supplier evaluation; logistics scenarios; inventory planning. | Fragile assumptions, geopolitical blind spots, overreliance on historical data, supplier confidentiality. |
Accounting | Reconciliation support; anomaly detection; audit preparation; compliance research. | Material errors, poor traceability, unverifiable sources, independence and accountability concerns. |
Human resources | Job-description drafting; workforce planning; training personalization; employee feedback synthesis. | Discrimination, opaque screening, privacy, impact on worker trust. |
Information systems | System design; AI-assisted development; cybersecurity monitoring; vendor assessment. | Security vulnerabilities, prompt injection, technical debt, unclear ownership. |
Management | Strategic decision support; organizational communication; change planning; team performance analysis. | Shallow strategic reasoning, persuasive but unsupported recommendations, stakeholder harm. |
Assessment Model
The wrong assessment question is “Did the student use AI?” The better question is “Can the student use AI well and defensibly?” A defensible AI-assisted deliverable is one where the learner can explain the task, justify the AI role, document the process, verify the output, revise intelligently, disclose use, and remain accountable for the final judgment.
This means many assignments should assess both product and process. The final memo, model, dashboard, code, or presentation still matters. But it should often be accompanied by prompt logs, source annotations, revision memos, confidence statements, risk notes, or oral defenses. After students practice with AI, a short individual check without it can show whether they can apply the concept to a new problem and explain their reasoning.
Assessment artifact | What it shows |
|---|---|
Prompt log | How the learner framed the task, iterated, constrained the output, and used AI as a collaborator rather than a shortcut. |
Source annotation | Which facts, data, assumptions, and claims were independently verified. |
Revision memo | What the learner accepted, rejected, corrected, or reframed, and why. |
Confidence statement | What the learner is confident about, what remains uncertain, and how uncertainty affects the recommendation. |
Risk note | The task’s risk level, which rules apply, and which safeguards the student chose. |
Oral defense | Whether the learner understands the work beyond the polished output. |
No-AI concept check | Whether the learner has the base knowledge needed to supervise AI assistance. |
Sample Performance Tasks
Layer | Performance task | Rubric focus |
|---|---|---|
Foundational | Identify errors and risks in an AI-generated market analysis. | Accuracy, hallucination detection, source checking, data sensitivity, disclosure, ethics. |
Intermediate | Redesign a weekly reporting workflow with AI assistance and human review. | Task fit, process design, verification, measurable value, documentation, accountability. |
Advanced | Create an AI adoption plan for a business unit. | Strategic fit, governance, risk tiering, organizational design, stakeholder impact, feasibility. |
Sample disclosure language
I used generative AI to brainstorm alternatives, draft an initial structure, and critique clarity. I independently verified factual claims using the cited sources, revised the analysis, and am responsible for the final conclusions.
Governance and Risk
Responsible AI should be taught as an operating system for AI leverage. Without clear rules, students may hide AI use, faculty may create inconsistent expectations, and the institution may reward polished outputs over durable capability. With clear rules, students can experiment responsibly, faculty can design assignments intentionally, and programs can build evidence about what works.
Governance also mirrors professional practice. The NIST AI Risk Management Framework’s Generative AI Profile emphasizes risks around provenance, testing, transparency, documentation, incident handling, and accountability [13]. In business education, those principles should appear in policy, pedagogy, assessment, and institutional use. University rules on data use, academic integrity, and approved tools still apply [1]. The table below gives examples of safeguards instructors can adapt to the risks of each task.
What Students Should Learn to Govern
- Privacy and data security: what information should never be entered into external tools, including confidential, personal, proprietary, or regulated data.
- Intellectual property and provenance: how to track sources, data lineage, ownership, generated content, and citation responsibilities.
- Bias and fairness: how biased data, prompts, models, or evaluation criteria can affect people, markets, and decisions.
- Hallucination and misinformation: why fluent outputs require verification, especially when claims affect people, money, compliance, or strategy.
- Model limitations: why current model capability is uneven, context-dependent, and subject to change. • Human accountability: why people remain responsible for AI-assisted work and decisions.
Risk Tiering for Educational and Business Use
Risk tier | Examples | Examples of safeguards |
|---|---|---|
Low | Brainstorming, drafting outlines, revising style, summarizing public material. | Disclosure when material; basic verification; no sensitive data. |
Moderate | Analysis informing recommendations, customer-facing drafts, internal process design, use of nonpublic but nonregulated data. | Source verification; human review; prompt and output documentation; data-use check; quality-control rubric. |
High | Decisions affecting people, money, legal compliance, admissions, hiring, grading, regulated data, confidential client work, or safety. | Prior approval; strict data controls; human decision authority; audit trail; bias and impact review; escalation pathway. |
Safeguards for Programs to Consider
- Acceptable-use policy for students, faculty, staff, and program contexts.
- Assignment-level AI labels: no AI, AI-optional, AI-encouraged, and AI-required.
- Disclosure standards for material AI assistance.
- Data-use rules for confidential, proprietary, personal, and regulated information.
- Risk tiers with controls based on consequence and sensitivity.
- Tool access and equity principles, including licensed tools where appropriate.
- Review cadence to update policies as tools, law, and evidence change.
- Incident response for misuse, data exposure, bias concerns, or misinformation.
Implementation Roadmap
The roadmap should be modular rather than tied to a rigid timeline. Different programs can start in different places, but the same institutional logic should hold: establish common language, build the base, embed AI in real business work, assess defensible use, and govern the risk.
Roadmap Elements
Element | What to build | Concrete outputs |
|---|---|---|
Coursework | Embed AI literacy and AI-mediated judgment into core and functional courses rather than isolating AI in a single elective. | Syllabus language, AI-labeled assignments, function-specific labs, no-AI concept checks, AI-required capstone tasks. |
Workshops | Create concentrated learning experiences for students, faculty, managers, executives, and partners. | Prompting and workflow clinics, governance workshops, role-based executive modules, faculty course-redesign sessions. |
Experiential projects | Use real problems where AI assistance must be paired with stakeholder needs, evidence, and measurable outcomes. | Client projects, agentic AI challenges, small-business pilots, analytics capstones, oral defenses. |
Communities of practice | Turn individual experimentation into shared institutional knowledge. | Faculty working groups, use-case libraries, prompt and rubric banks, showcase sessions, peer reviews. |
Certificates and non degree learning | Scale foundational access and career transition support beyond degree programs. | Self-paced modules, applied exercises, responsible AI standards, functional application pathways. |
Faculty development | Help faculty redesign courses without imposing a single pedagogical model. | Assignment templates, disclosure examples, assessment rubrics, AI policy menu, examples of strong and weak AI use. |
Assessment and governance | Measure AI literacy and responsible use at key points in the learner journey. | Entry/exit diagnostics, rubric-aligned artifacts, risk-tier checklists, program-level dashboards. |
Faculty Support Model
Faculty development is the hinge. AI will enter courses whether or not the syllabus names it. Faculty therefore need practical supports: examples of no-AI, AI-optional, and AI-required assignments; standard disclosure language; prompt-log templates; grading rubrics; discipline-specific use cases; and a low-friction channel for sharing what works. The goal is not uniformity. The goal is coherent principles plus local pedagogical judgment.
What success looks like
- Students can use AI to improve speed and quality while explaining what they verified, what they rejected, and what they remain accountable for.
- Faculty can design assignments that reward reasoning, source checking, and revision rather than polished AI generated output alone.
- Programs can describe the AI competencies their graduates develop and the evidence used to assess those competencies.
- Executives and professionals can leave Smith offerings with usable workflows, risk controls, and governance habits.
- Smith can connect classroom learning, executive education, research, and community outreach under a single human-first AI leverage position.
The Smith standard
Build the base. Choose the right task. Design the workflow. Verify the output. Govern the risk. Stay accountable for the result.
Selected Sources and Evidence Anchors
This resource draws on the AI Leverage through Literacy framework, the evidence review on AI leverage in business education, and public Smith School materials. Bracketed citations in the text refer to the sources below.
[1] University of Maryland. Guidelines for Use of Generative AI
[2] Robert H. Smith School of Business. (2026). AI @ Smith.
[3] Robert H. Smith School of Business. (2026). Artificial Intelligence and Career Empowerment enrollment surpasses 50K. [4] Robert H. Smith School of Business. (2026). Smith School formally anchors AI in key MS programs.
[5] Robert H. Smith School of Business. (2026). 330 Students From Across Campus Sign up for Smith’s Agentic AI Challenge.
[6] Robert H. Smith School of Business. (2026). Smith’s Center for AI in Business Awarded $1M+ Federal Grant to Help Small Businesses Adopt AI.
[7] Noy, S., and Zhang, W. (2023). Experimental evidence on the productivity effects of generative artificial intelligence. Science. [8] Brynjolfsson, E., Li, D., and Raymond, L. (2025). Generative AI at work. Quarterly Journal of Economics. [9] Dell’Acqua, F., et al. (2026). Navigating the jagged technological frontier: Field experimental evidence of the effects of AI on knowledge worker productivity and quality. Organization Science.
[10] Becker, J., Rush, N., Barnes, B., and Rein, D. (2025). Measuring the impact of early-2025 AI on experienced open-source developer productivity. METR.
[11] Kestin, G., Miller, K., Klales, A., Milbourne, T., and Ponti, G. (2025). AI tutoring outperforms in-class active learning: an RCT introducing a novel research-based design in an authentic educational setting. Scientific Reports, 15, 17458. doi:10.1038/s41598-025- 97652-6
[12] OECD. (2026). Digital Education Outlook 2026.
[13] National Institute of Standards and Technology. (2024). Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile.
Further Reading
Padmanabhan, B., and McCarthy, D. (2026). AI Leverage through Literacy: A Framework for Business and Business Education. Internal framework memo, April 29, 2026.
AACSB. (2025). GenAI adoption in business schools: Deans and faculty respond.
Stanford Institute for Human-Centered Artificial Intelligence. (2025). AI Index Report 2025, Economy chapter. Robert H. Smith School of Business. (2026). 2026 AI Symposium on Design and Governance.