What is an AI agent?
An AI agent is software that uses an AI model to pursue a goal on someone’s behalf. It interprets a request, plans steps, draws on data and tools, and takes actions — with some degree of autonomy — while people remain accountable for what it does.1,2,4
This guide’s working definition, synthesized from the sources under Definitions in the field.
What an agent is made of
| Goal | A task or outcome it is working toward, set by a person or another system. |
|---|---|
| Model | A large language or reasoning model that interprets requests and plans the next step. |
| Instructions | Its role, scope, tone, and rules — what it may and may not do. |
| Knowledge | The documents, data, and systems it can draw on, often through retrieval. |
| Tools & actions | APIs, forms, messages, and workflows it can call or trigger. |
| Memory | What it retains within a task or across conversations. |
| Oversight | Logs, limits, approvals, and the people who can review or stop it. |
AI agents and agentic AI
The two terms are often used as if they mean the same thing. They don’t. An AI agent is a single system built for a defined task, using an AI model and tools7. Agentic AI is the broader approach: systems, often several agents together, that break a goal into steps, keep memory, and coordinate with limited supervision7. Generative AI answers and creates; agentic AI plans and acts6.
Put simply, agents are the things. Agentic AI is what they can do and how they are put together.
| AI agent | Agentic AI | |
|---|---|---|
| What it names | A specific system you can point to | A capability and a way of designing systems |
| Unit | One agent | One or more agents working together |
| Scope | A defined task | A multi-step goal, often across several systems |
| Autonomy | Bounded and mostly task-driven | Plans, splits up work, remembers across sessions |
| Campus example | An advising agent that books appointments | An orchestrator that routes one student request across advising, financial aid, and registration agents |
| Typical failures | Wrong answers, brittle instructions | Errors passed between agents, unexpected combined behavior, hard to explain |
How this guide uses the terms. “AI agent” means a specific system you can inventory, assign an owner, and switch off. “Agentic AI” means the capability. Governance attaches to agents: each one gets a registry entry and an accountable owner, including every agent inside a multi-agent system. See the multi-agent orchestrator scenario.
A spectrum of agency
“Agent” covers a range: AI systems vary in their levels of autonomy3. The more an agent can act without a person in the loop, the stronger the controls it needs — see the six lenses and control tiers.
- 01
Answers
Responds to questions from approved sources. Takes no action.
e.g. Policy or FAQ assistant
- 02
Assists
Drafts, summarizes, or recommends. A person decides and acts.
e.g. Advising note drafter
- 03
Acts with approval
Prepares an action — an email, a ticket, a record change — and waits for sign-off.
e.g. Financial aid outreach queued for staff review
- 04
Acts within bounds
Completes defined tasks on its own, inside limits, with logging and escalation.
e.g. IT password-reset or room-booking agent
- 05
Orchestrates
Coordinates other agents and systems across a multi-step workflow.
e.g. Admissions-to-enrollment handoff orchestrator
What usually isn’t an agent
Where a model follows a fixed, predefined path, it is better described as a workflow than an agent5.
A chatbot that only answers
Useful, but without tools or actions it sits at the bottom of the ladder.
Rules-based automation
Workflows and RPA follow fixed scripts; they do not interpret or plan.
A one-off prompt
Asking a model a question is AI use, not an agent deployment.
Definitions in the field
There is no single agreed definition. These are the ones this guide draws on, from foundational computer science to current policy and higher-education practice.
“An agent is anything that can be viewed as perceiving its environment through sensors and acting upon that environment through actuators.”
“A computer system, situated in some environment, that is capable of flexible autonomous action in order to meet its design objectives.”
“Different AI systems vary in their levels of autonomy and adaptiveness after deployment.”
“AI agent systems consist of at least one generative AI model and scaffolding software that equips the model with tools to take a range of discretionary actions.”
“Agents … are systems where LLMs dynamically direct their own processes and tool usage, maintaining control over how they accomplish tasks.”
“AI systems that can plan and take action.”
“Agentic AI systems … represent a paradigm shift marked by multi-agent collaboration, dynamic task decomposition, persistent memory, and coordinated autonomy.”
How agents arrive on campus
Most campuses already have agents — few planned all of them. Agents enter through ten pathways, and each needs a different first move. An inventory that only tracks what IT builds will miss most of them.
- 01
Switched on inside existing platforms
Vendors add agent features to the SIS, LMS, ERP, CRM, or service desk through routine updates.
- Workday agents, Salesforce Agentforce (Education Cloud), ServiceNow AI Agents, Instructure Canvas AI features, Ellucian AI features
- Often low — can arrive with no procurement decision.
- Review release notes; set a default-off or pilot-first rule for agent features.
- 02
Purchased as a standalone product
A department buys an agent for admissions, advising, tutoring, or support.
- Mainstay, Ocelot, Element451, EAB Navigate AI features
- Medium — visible if purchasing routes through review.
- Add agent questions to procurement and security review.
- 03
Built on an enterprise AI suite
Staff and faculty create no-code agents or custom assistants in institution-licensed AI tools.
- Microsoft 365 Copilot and Copilot Studio agents, Google Gemini Gems and Gemini Enterprise, ChatGPT Edu custom GPTs, Claude for Education Projects
- Low to medium — many small agents, few owners on record.
- Require registration above Tier 1; set sharing and data defaults centrally.
- 04
Built by central IT or a development team
Custom agents built on model APIs and frameworks, integrated with systems of record.
- Azure AI Foundry Agent Service, Amazon Bedrock Agents, Google Vertex AI Agent Builder, OpenAI Agents SDK, LangGraph, Model Context Protocol (MCP) integrations
- High — but only if the team logs them.
- Apply the full design-to-retire playbook and the registry.
- 05
Built by departments, faculty, or researchers
Course assistants, lab workflow agents, and departmental tools built locally.
- A course GPT or Gem for a single class, Copilot Studio departmental helpers, n8n or Zapier AI workflows, lab pipelines on model APIs
- Low to medium.
- Offer a supported platform with guardrails so local builders don’t start from scratch.
- 06
Built by students
Class projects, hackathons, clubs, and startups — sometimes touching campus data or services.
- Hackathon apps on OpenAI, Anthropic, or Google APIs; coding agents such as GitHub Copilot, Cursor, and Claude Code
- Low.
- Provide sandboxes and clear rules on campus data and branding.
- 07
Brought by individuals
Personal agents — meeting notetakers, email and browser agents — used with campus accounts.
- Otter.ai, Fireflies.ai, and Read AI meeting notetakers; ChatGPT agent; Perplexity Comet and other AI browsers
- Very low — shadow AI.
- Publish acceptable-use guidance; control app consent in identity systems.
- 08
Delivered through integrations and marketplaces
LMS apps, browser extensions, video-meeting apps, and plug-ins that add agent behavior.
- Canvas and LMS LTI apps, Zoom AI Companion, Microsoft Teams and Slack apps, AI browser extensions
- Low — approved once, updated silently.
- Re-review integrations when vendors add agent capabilities.
- 09
Shared through a consortium, system, or pilot
System-wide services, grant-funded projects, or free vendor pilots.
- Internet2 NET+ service agreements, system-wide licenses such as the California State University ChatGPT Edu rollout, free vendor pilots
- Medium — ownership can be unclear.
- Name a local owner and exit plan before the pilot starts.
- 10
Arriving from outside
Agents acting for applicants, students, vendors, or others interact with campus portals, forms, and staff.
- Applicants’ AI browsers completing forms, vendor bots contacting staff, third-party agents calling campus APIs or MCP servers
- Very low — the institution doesn’t own them.
- Plan for inbound agents: identity checks, rate limits, and terms of use.
Named products are illustrative, not endorsements. Vendor capabilities and names change often — confirm current features before relying on them.
Record every pathway in one place — see the agent registry fields.
References
- Stuart Russell and Peter Norvig, Artificial Intelligence: A Modern Approach, 4th ed. (Pearson, 2020), ch. 2. aima.cs.berkeley.edu
- Michael Wooldridge and Nicholas R. Jennings, “Intelligent Agents: Theory and Practice,” The Knowledge Engineering Review 10, no. 2 (1995): 115–152. doi.org/10.1017/S0269888900008122
- OECD, Recommendation of the Council on Artificial Intelligence, OECD/LEGAL/0449, definition of “AI system” as revised 8 November 2023. legalinstruments.oecd.org/en/instruments/OECD-LEGAL-0449
- NIST Center for AI Standards and Innovation, “Request for Information Regarding Security Considerations for Artificial Intelligence Agents,” Federal Register, 8 January 2026. www.federalregister.gov/documents/2026/01/08/2026-00206/request-for-information-regarding-security-considerations-for-artificial-intelligence-agents
- Erik Schluntz and Barry Zhang, “Building Effective Agents,” Anthropic Engineering, 19 December 2024. www.anthropic.com/engineering/building-effective-agents
- EDUCAUSE, “4 Ways Agentic AI Will Reshape Higher Ed Work,” EDUCAUSE Exchange podcast, 2026. er.educause.edu/podcasts/educause-exchange/4-ways-agentic-ai-will-reshape-higher-ed-work
- Ranjan Sapkota, Konstantinos I. Roumeliotis, and Manoj Karkee, “AI Agents vs. Agentic AI: A Conceptual Taxonomy, Applications and Challenges,” Information Fusion 126 (2026): 103599. arxiv.org/abs/2505.10468
The ten arrival pathways are this guide’s own synthesis of campus practice. All sources across the guide: References.