What you'll learn

Learn the fundamentals of autonomous web agents, what they are, how they work, their limitations, and the decision-making strategies taken to optimize their performance.
Build autonomous web agents that can perform tasks such as finding, scraping, and summarizing a webpage, filling out forms, and signing up for newsletters.
Explore the AgentQ framework, which uses a combination of Monte Carlo Tree Search (MCTS), self-critique mechanism, and Direct Preference Optimization (DPO) to teach agents to self-correct.

Course recap

PRO

This course, built in partnership with AGI Inc., teaches how to build AI agents that autonomously navigate and interact with websites. It covers the full spectrum from simple web scraping agents to autonomous browser agents capable of filling forms, clicking through pages, and placing orders. The course introduces AgentQ, a state-of-the-art framework that combines Monte Carlo Tree Search (MCTS), self-critique mechanisms, and Direct Preference Optimization (DPO) to achieve 95.4% success rates on real-world web tasks like booking reservations on OpenTable.

Concept map

Key concepts

AI AgentsAutonomous software programs that perform tasks on behalf of users, consisting of UI, control, knowledge, communication, and data processing modules
Monte carlo tree searchSearch algorithm for finding optimal solutions in complex decision spaces, used in AgentQ for structured exploration of web actions
Reinforcement learningLearning paradigm where agents improve through feedback; DPO is used as a streamlined alternative to RLHF
Prompt EngineeringCrafting instructions for LLMs to guide agent behavior and structured output generation
Exploration vs. exploitationTrade-off between trying new paths (breadth-first) and deepening known good paths (depth-first) in web navigation

Lesson highlights

1.Introduction — Overview of AI browser agents and the AgentQ approach
2.Intro to Web Agents — What web agents are, key components, and current limitations
3.Building a Simple Web Agent — Web scraping agent with structured output using OpenAI
4.Building an Autonomous Web Agent — Multi-step autonomous agent using MultiOn for navigation, form filling, and newsletter signup
5.Agent Q — Framework combining MCTS, self-critique, and DPO for self-correcting web agents
6.Deep Dive into AgentQ and MCTS — Implementing MCTS on grid worlds and visualizing browser agent search trees
7.Future of AI Agents — Current challenges, multi-agent systems, and standardization needs
8.Conclusion — Course recap

About this course

Learn how to build AI agents that interact with websites in Building AI Browser Agents, taught by Div Garg and Naman Garg, Co-founders of AGI Inc, and built in partnership with AGI Inc.

AI browser agents can log into websites, fill out forms, click through web pages, or even place an online order for you. They use both visual information, like screenshots, and structural data, like the HTML or Document Object Model (DOM) of a web page, to reason and take actions.

With the complexity of web pages and many possible actions at each step, it can be challenging for an AI browser agent to complete an assigned task. A single error—like clicking the wrong button or misreading a field—can compound into unexpected outcomes.

In this course, you’ll understand how autonomous web agents work, their current limitations, and how AgentQ enables them to improve through self-correction.

In detail, you’ll:

Learn what web agents are, how they automate tasks online, their architecture, key components, limitations, and an overview of their decision-making strategies.
Build a web agent that can scrape the Deeplearning.AI website and return course recommendations in structured output format based on natural language instructions.
Build an autonomous web agent that can execute multiple tasks, such as finding and summarizing webpages, filling out a form, and signing up for a newsletter.
Explore AgentQ, a framework that enables agents to self-correct through a combination of Monte Carlo Tree Search (MCTS), Self-critique mechanism, and Direct Preference Optimization (DPO), a reinforcement learning algorithm.
Deep dive into MCTS, learn how it finds the most optimal path with an example of Gridworld animation, and use AgentQ to complete web tasks.
Understand the current state and future directions of AI agents—including key factors shaping their evolution, such as hardware, algorithms, and data availability.

By the end of this course, you will have hands-on experience building browser agents and a deeper understanding of how to make them more robust and reliable.