Agentic AI Program Overview
Agentic AI moves beyond generating responses to completing work. Unlike generative AI, which responds to prompts by creating text, images, or code, agentic AI systems can plan, reason, use tools, and execute multi-step workflows toward a defined goal.
In this hands-on agentic AI program, business leaders will learn how to:
- build agentic workflows,
- identify where agents can create meaningful value,
- determine where human oversight remains essential, and
- develop a practical approach to responsible AI implementation.
No coding is required in this program. Through collaborative exercises, you’ll apply what you learn to a real-world business challenge, refine your approach alongside peers, and consider the governance, validation, and risk factors required to implement agentic AI responsibly and effectively.
At the end of this program, you will leave with working agentic workflows, practical agentic AI implementation frameworks, and a clear roadmap for turning agentic AI use cases into measurable business value.
Benefits
With nearly 74 percent of organizations expecting to use AI agents by 2027 and U.S. job postings mentioning agentic AI skills nearly quadrupling from 2024 to 2025, it’s clear that organizations are moving quickly to build out their agentic AI capabilities.
This no-code AI agents program equips you to meet this demand by focusing on real-world business challenges. You will gain a vendor-neutral mental model — spanning memory, skills, and tools — that stays relevant as technology evolves, alongside the judgment to determine where agents can act autonomously and where human oversight is essential. Crucially, you will learn to rigorously validate AI outputs to avoid the costly pitfall of uncritical acceptance, backed by a governance framework that helps you deploy agentic AI safely and in line with your organization’s risk and compliance requirements.
After attending this agentic AI program, you will be able to:
- Build an AI agent and run agentic workflows without code, using the memory-skills-tools model and improving agents with the run-read-fix loop.
- Design an agentic workflow for a real problem, deciding what the agent does autonomously and where a human must validate.
- Evaluate agentic AI outputs and risks, and apply a governance framework to deploy agentic AI responsibly in your organization.
Participants will also earn a Certificate of Completion from the Harvard Division of Continuing Education.
Who Should Take an Agentic AI Program
This no-code agentic AI program is designed for business leaders and professionals who are responsible for their team’s workflows and have the authority to redesign how the work gets done.
Mid- to senior-level professionals interested in turning AI strategy into working practice and developing AI agents for business purposes who work in the following areas would benefit from this program:
- Financial services and/or banking
- Professional services and/or consulting
- Healthcare
- Technology
- Public sector
- Operations-heavy business
You do not need a technical background to participate in this program.
Curriculum
During this agentic AI program, you’ll build a working AI assistant to help with your own productivity. You’ll also leave with a concrete plan for bringing agentic AI to your organization.
To accomplish this, you’ll learn about topics such as:
- Agentic AI Foundations: Understand the shift from large language models to autonomous agents.
- Building Reusable AI Skills: Create instructions that an AI agent can run repeatedly to complete a defined task in a consistent way.
- Improving Agent Outputs: Practice an iteration loop to improve reliability and usefulness.
- Reducing Errors with Multi-Agent Review: Use multiple agents to review the same material independently and consolidate feedback for stronger results.
- Designing Agentic Business Workflows: Map a real business task, determine what the agent should handle versus what requires human review, and prototype agentic workflows.
- Governance, Risk, and Human Oversight: Apply a practical agentic AI governance framework for data access, autonomous execution, approvals, and output validation to learn where an agent can act independently and where human review, approval, or judgement is required.
- Adopting Agentic AI in the Organization: Explore how to move from individual workflows to team- and business-unit-level adoption by identifying repetitive, high-volume, multi-step work where an AI can create measurable value.
Technology Requirements: A Mac or Windows PC laptop with a recent operating system and Wi-Fi capability to connect to the on-campus network is required for this program; tablets and phones do not meet this requirement.
Before the program begins, participants must download and install the required agentic AI software (license provided) and confirm it works. Software details and setup instructions will be provided three weeks before the program start date.
Note: Employer-managed laptops may require IT assistance, and organizational restrictions may prevent installation or use. Participants are encouraged to check with their IT team early; a personal laptop that meets these requirements may be used instead.
Considering this program?
March Schedule
Day 1
- The AI Landscape: What Every Business Leader Needs to Know
- Building Agentic Workflows
- Building Your Knowledge Base
Day 2
- Designing Workflows for Your Organization
- Responsible AI – Making Agentic Workflows Safe
July Schedule
Day 1
- The AI Landscape: What Every Business Leader Needs to Know
- Building Agentic Workflows
- Building Your Knowledge Base
Day 2
- Designing Workflows for Your Organization
- Responsible AI – Making Agentic Workflows Safe
Instructor
-
![Hanspeter Pfister]()
Hanspeter Pfister
Hanspeter Pfister is the An Wang Professor of Computer Science at Harvard University and a leading expert in AI, data science, and visual computing. He previously co-chaired the Harvard Business Analytics Program and teaches in executive education programs across Harvard. His research spans AI interpretability, computer vision, and advanced analytics, with more than 400 peer-reviewed publications and 50 U.S. patents.
Instructor
-
![Hanspeter Pfister]()
Hanspeter Pfister
Hanspeter Pfister is the An Wang Professor of Computer Science at Harvard University and a leading expert in AI, data science, and visual computing. He previously co-chaired the Harvard Business Analytics Program and teaches in executive education programs across Harvard. His research spans AI interpretability, computer vision, and advanced analytics, with more than 400 peer-reviewed publications and 50 U.S. patents.
Earn a Certificate of Leadership Excellence
The Certificates of Leadership Excellence (CLE) are designed for leaders with the desire to enhance their business acumen, challenge current thinking, and expand their leadership skills.
This program counts toward the CLE in Strategy and Innovation.
Register today and get started earning your certificate.
FAQs
What is agentic AI?
Agentic AI is an advanced class of artificial intelligence systems designed to act with increasing independence, make decisions, and execute multi-step tasks to achieve specific goals under human direction and review.
Unlike traditional or static AI models that simply respond to individual prompts, agentic AI possesses agency and can break down goals, plan sequences of actions, evaluate its own progress, use external tools (like email, web search, and calendars), and adapt its strategy when encountering obstacles.
How is agentic AI different from generative AI?
Agentic AI is different from generative AI primarily in its scope and primary function:
- Generative AI focuses on content creation: producing text, images, audio, or code in response to explicit, single-step prompts. It acts as a passive creation engine that waits for human direction at each stage.
- Agentic AI focuses on action and goal accomplishment. It often uses generative AI models as its core reasoning engine and adds the ability to plan, use tools, check its own work, and carry out steps on its own
In short: generative AI creates, while agentic AI acts.
What are agentic workflows?
Agentic workflows are iterative design patterns that allow AI to accomplish complex, open-ended tasks through step-by-step reasoning rather than single-shot generation. Core components of agentic workflows include planning, tool usage, reflection and critique, and multi-agent collaboration.
What are examples of agentic AI use cases?
Examples of agentic AI use cases include:
- Autonomous Software Engineering: Coding agents that read issue tickets, locate bugs across a codebase, write fixes, execute unit tests, iterate until tests pass, and open pull requests for developer review.
- Customer Support Escalation & Resolution: Support agents that go beyond static FAQ responses by authenticating customers, checking inventory or billing databases, and preparing refunds and ERP updates for staff approval
- Market Research & Strategic Intelligence: Research agents that design search queries, scour dozens of web sources, cross-verify data points for credibility, synthesize market trends, and produce executive summary briefs.
- Financial Auditing & Invoice Reconciliation: Compliance agents that cross-check supplier invoices against purchase orders and receipts, flag discrepancies, reach out to vendors for missing documentation, and prepare audit-ready logs.
- Supply Chain Optimization: Logistics agents that monitor real-time stock levels, track weather or shipping delays, adjust routing orders, and draft purchase orders for approval when threshold criteria are met.
How do organizations govern AI agents?
Organizations govern AI agents by deciding what an agent may access, what it may do without approval, and how its work is checked. Key mechanisms of AI agent governance include:
- Data Access: What the agent is allowed to see, such as specific folders, mailboxes, or business systems.
- Execution: What the agent may do on its own, and which actions, such as sending a message, changing a record, or making a payment, need human-in-the-loop approval first.
- Output Validation: How results are checked before anyone relies on them, for example, by requiring a source for every claim and having a second agent or a person review the work.
Organizations then add their own policies, access controls, and audit requirements, often guided by frameworks such as the NIST AI Risk Management Framework or the EU AI Act.
