Overview
Learn to build working agentic systems and understand how agents will change software, organizations, services, and everyday work.
At a Glance
Class
Thursdays, 10:00 AM–12:00 PM, starting February 4, 2027
Location
MIT Media Lab, E14-633
Format
Weekly lectures, technical assignments, and hands-on project work
Commitment
About 6–10 hours per week outside class
Who Should Apply
Students who want to build useful systems with AI and understand where agentic technology is going.
Strong applicants may bring experience from engineering, product, research, design, business, or other fields.
Everyone is expected to build with modern AI tools and agents, learn through experimentation, and contribute to working projects.
MIT students and eligible cross-registrants are welcome to apply.
What You Will Learn and Build
The course connects each topic to the Agentic Web: an ecosystem in which agents discover tools, exchange information, use services, and act for people and organizations.
Personal Agents
Build agents for research, coding, planning, learning, or repeated work using context, memory, personal data, and tools.
Agent Loops and Autonomous Work
Design agents that plan, act, evaluate results, recover from failure, and know when to ask a person for help.
Multi-Agent Systems and Organizations
Coordinate specialized agents through delegation, task routing, shared memory, communication, and review.
Products and Infrastructure for Agents
Build APIs, tools, services, and environments that agents can discover, use, and combine across the web.
Open and Local Models
Explore model choice, local inference, adaptation, privacy, latency, deployment, and cost-quality trade-offs.
Robotics, Embodied Agents, and World Models
Connect multimodal perception, simulation, spatial reasoning, planning, and action across physical and online environments.
The Agentic Web
Study identity, delegated authority, permissions, trust, payments, shared protocols, security, and agent-to-agent communication.
Applied Topics
Apply agentic systems to emerging forms of economic activity, organizational work, and collective decision-making.
Agentic Commerce
Explore markets where agents discover services, compare offers, negotiate, transact, and build reputation on behalf of people and organizations.
Agentic Organizations and Work
Design teams of agents that operate workflows, share knowledge, allocate tasks, and coordinate decisions with people.
Agentic Societies
Study identity, trust, governance, voting, norms, privacy, and dispute resolution when autonomous agents participate in collective systems.
Course sandbox: NANDA Town. This open Nanda at MIT Media Lab environment supports scenarios where agents talk, trade, vote, and team up; test protocols before deployment; and analyze how agent interactions succeed or fail.
Course format: Weekly classes combine instruction and discussion with regular building and testing, culminating in Demo Day.
Admissions
Application Steps
- Complete Step 1: Background and Course Fit.
- Complete Step 2 when it is released in December.
- Register for the course through MIT WebSIS or Harvard cross-registration.
Application Waves
Applications are reviewed in waves, and earlier applications receive priority. In each wave, applicants may be accepted, rejected, or waitlisted.
- Priority deadline: .
- Second deadline: .
- Final deadline: .
- Late applications: reviewed on a rolling basis after Monday, February 1, only if seats remain.
Both application forms (Steps 1 and 2) must be completed for review. Course registration through MIT WebSIS or Harvard cross-registration is separate and not needed for review.
How Applications Are Reviewed
We assess readiness to build, relevant experience, interest in the course, and expected contribution to a team.
Admission is very selective. Submitting application forms, registering at MIT, or obtaining Canvas access does not guarantee a place without an explicit acceptance email from the course team.
Course Schedule
Class Meetings and Key Dates
- Weekly class: Thursdays, 10:00 AM–12:00 PM (Eastern Time), in E14-633 at the MIT Media Lab.
- First class: Thursday, .
- No class: Thursday, (MIT spring break).
- Demo Day: Thursday, .
Projects and Demo Day
The course builds cumulatively from a single agent to a coordinated team and then to an organization connected through the Agentic Web.
1. Individual Agent
Build an agent with useful context, memory, tools, and workflows for an individual user or repeated task.
2. Multi-Agent Team
Connect individual agents into a small team that communicates, coordinates work, and completes tasks collectively.
3. Agentic Organization
Combine the earlier work into an agentic organization that discovers and uses services and interacts with other teams through NANDA Town.
Final Deliverable
A working technical project, such as an infrastructure component, product, service, environment, or shared standard.
A clear interface for agents, documentation showing how they use it, and evidence that it works in realistic scenarios.
Support and Demo Day
Mentor feedback, technical workshops, and support sessions are part of the course.
Final projects are presented at Demo Day on Thursday, .
View detailed project and Demo Day requirements
- Projects may include infrastructure components, agent-ready products, services, environments for agents, and security or reliability tools.
- Each project should solve a clear problem, explain what an AI agent can do with it, and integrate with relevant agents, infrastructure, or services.
- Possible focus areas include identity, reputation, permissions, payments, memory, monitoring, evaluation, adversarial testing, service marketplaces, and workflow automation.
- For mentor sessions, teams should arrive with specific questions, a working artifact, and clear next steps.
Course Instructors
Course Team
Mentors and Guest Speakers
Spring 2027 mentors and speakers will be announced throughout the semester. Sessions will emphasize working systems, implementation choices, failures, deployment, reliability, and lessons from practice.
Background
The course evolved from Sandy Pentland's Development Ventures and has a long track record of helping students turn class projects into working products and ventures.
View previous course offerings
- Fall'26: AI Studio
- Spring'26: AI Studio
- Fall'25: AI Studio
- Spring'25: AI Venture Studio
- Fall'24: MIT Foundations of AI Ventures
- Spring'24: AI Venture Studio
- Fall'23: AI Venture Studio
- Spring'23: AI + Web3 for Impact: Venture Studio
- Fall'22: AI + Web3 for Impact: Venture Studio
- Spring'22: AI for Impact: Venture Studio
- Fall'21: AI for Impact ~ Building Global Ventures Solving Societal-Scale Problems
- Spring'21: AI for Impact
- Fall'20: Global Ventures ~ Data and AI for Resilience after COVID19
- Spring'20: AI for Impact ~ Towards Solving Societal-Scale Problems via Media Ventures
FAQ
What is the Agentic Web?
The Agentic Web is an ecosystem of connected AI agents, tools, services, and shared standards. Agents can discover services, exchange information, and combine tools to complete tasks across different systems.
Is collaboration allowed?
Yes. Collaborative learning is part of the format. Students should help each other learn while submitting their own work and documenting team contributions clearly.
Do I need a startup idea?
No. The course focuses on learning and building agentic systems. Projects may become products or ventures, but they may also be research tools, open-source infrastructure, personal agents, or public-interest systems.
Does the course count toward the Entrepreneurship and Innovation Minor?
It may. Contact eiminor_admin@mit.edu to confirm eligibility and requirements.
Videos and Previous Demo Days
Explore recent student demos and talks that provide context for the ideas, systems, and communities behind the course.