inflearn logo

Building an Unreal Agent from Scratch (Liu)

In the era of agents, true skill comes from 'control.' You will perfectly master the operating principles of agents by building the core architectures of Claude Code—Agent, Tool, MCP, Skill, and AgentTeam—from scratch.

(4.8) 13 reviews

241 learners

Level Basic

Course period Unlimited

C#
C#
Unreal Engine
Unreal Engine
Web API
Web API
AI
AI
AI Agent
AI Agent
C#
C#
Unreal Engine
Unreal Engine
Web API
Web API
AI
AI
AI Agent
AI Agent

Reviews from Early Learners

4.8

5.0

이채연

54% enrolled

I've only finished about half of the course, but I was so impressed by the quality that I came here to leave a review first... The course covers everything from the very basics of AI agents to concepts and hands-on practice, which has greatly deepened my understanding of agents like Claude Code. I didn't expect it to be this detailed, and it has been a huge help in my AI studies. I've also gained the confidence that I can customize agents myself to increase efficiency. I started taking this course because I use Claude Code a lot while studying game development, but I think it can be applied to all fields. I'm really looking forward to the Unreal Engine part that I'll be taking next.. Great lecture, highly recommended!

5.0

지모찡

100% enrolled

Unreal was just the starting point; the core of this lecture was about how to utilize LLMs and agents. While listening to the lecture, I realized that I hadn't been utilizing AI properly until now. In particular, understanding how LLMs call tools and the workflow through which agents perform tasks has significantly shifted my perspective on AI utilization. It would have been even better if the lecture had covered how to connect other LLMs, but I believe that if you understand the structures and principles explained in the course, you should be able to expand and integrate them yourself. I think this lecture is sufficiently helpful not only for those interested in Unreal but also for anyone interested in AI agent structures and LLM utilization methods.

5.0

asdsa112

49% enrolled

I didn't even dare to think about implementing AI Agent technology because it felt so far out of reach, but this lecture was incredibly helpful as you taught me everything and walked me through the implementation from the ground up. It was a course that truly helped wash away my fears regarding AI Agents. Additionally, I feel like I learned a lot of very clean code writing patterns as a bonus. From the curriculum to the quality of the instruction, everything was perfect. Thank you.

What you will gain after the course

  • The ability to directly design and implement the core architecture of Claude Code (Thinking, Tool calling, MCP, Permission authorization, Context management, and Agent teams).

  • General-purpose agent development capabilities to design and integrate agents for any domain.

Building Agents from Scratch

Design and implement everything about the agent yourself, except for the LLM.

Are you
'using' agents, or 'understanding' them?

Claude Code, Codex — the performance of LLM-based agents is crossing boundaries every day.
They have already begun to replace human roles in numerous development areas,
and this trend is irreversible.


"It's not quite at that level of performance yet, is it?"

"It didn't work the way I wanted it to?"


On the other hand, team leaders at major corporations around me
are already getting everything done through vibe coding alone.


Same tool, different results.

There is only one difference.


How deeply you understand the agent and

control them.

The developer's answer is always the same

What is the best way to handle agents?

It may seem like a detour, but in the end, there is only one answer.
Building it yourself.


The difference between those who have built the internal structure by hand and those who haven't
will widen overwhelmingly as time goes on. theo thời gian.


In this course, you will design and implement everything about an agent from scratch, excluding the LLM itself.
And it's not just any agent—you will perfectly replicate the core features of Claude Code, which is currently considered the most powerful in the coding market, and go a step further to create an agent that controls the entire Unreal Engine.


Being able to build your own agent means you are no longer someone waiting for a desired feature to appear, but someone who creates it right now. These are the rarest and most highly valued talents in the market today.

See for yourself the agents you will build


"Build me a kingdom"a single command that shows the agent completing everything on its own within Unreal Engine, from asset creation and material production to object placement and lighting setup


Agent-exclusive UI fully integrated into the Unreal Editor

Real-time tracking of AI's thinking flow and tool calling process


Request for permission approval when AI uses tools

Directly design and implement the security layer


Check the context structure, the core of the prompt, at a glance

Completely recreate the architecture of Claude Code, including Claude.md, Skill, and Tool



/ Instantly explore commands and skills with a single character

Implementing the command system of Claude Code exactly as it is



Ask the AI questions directly by mentioning assets within the Unreal Editor



AI analyzes and answers everything from the internal properties and structure to the configuration values of mentioned assets



With just a single screenshot of a Blueprint, AI reads and explains the entire logic



Agent Teams — Divide complex tasks and automatically organize teams

Directly recreating the Agentic Teams, the latest feature of Claude Code



Monitor each member of the established agent team by tab



Systematically manage project information in the same way as Claude Code

I didn't just use MCP in Claude Code.
I built the Agent, MCP, and Tools all from scratch.

From design to implementation, everything was built from scratch.

This course focuses on the process itself of designing and building an agent using C#.

We will build it using the same principles as Claude Code, such as Agents, MCP, Tools, and conversation loop processing. The difference is that we will be building this from scratch ourselves.

As an example of that process, we will be working with Unreal Engine.

This is because it is one of the most difficult domains for an agent to handle.

If you can pull this off, you can attach an agent to any domain.

This is not a game development course. It is an agent development course.

After completing this course, you will be able to create your own agents anytime, anywhere.


And the most powerful characteristic of an agent

Once built, your agent will grow stronger alongside the LLM every time it evolves.

There are many developers who use agents.


There are few developers who build agents.

Let's become a developer with scarcity.

Recommended for
these people

Who is this course right for?

  • Developers who use agents every day but feel frustrated because they can't control them as desired

  • Developers who are curious about the internal architecture of tools like Claude Code and Cursor but don't know where to start digging.

  • Developers who want to build their own agents and integrate them into their own domains (games, industry, internal company tools, etc.)

Need to know before starting?

  • C# Basic Syntax (Covered in the lecture even if you have no experience with ASP.NET or Blazor)

  • Basic usage of Unreal Engine (it is sufficient if you have at least opened the editor before)

Hello
This is Rookiss

84,914

Learners

4,105

Reviews

5,492

Answers

4.9

Rating

55

Courses

Education

- Completed Condorcet Prépa MPSI (France)
- Bachelor of Software Engineering, Tsinghua University (China)
- Master of Information Security, Korea University Graduate School (Korea)

Experience

- Former Krafton (TERA Server Team)
- Former NCSOFT (Blade&Soul2 Content Programming Team)

Areas of Interest

- Reversing and vulnerability analysis
- MMORPG game development (C++, C#)
- Foreign languages (English, French, Chinese, Japanese)

Introduction

I started game development out of pure curiosity to study hacking...
and as time passed, I have somehow become a 10th-year game developer.
"When will I ever finish making a game at this rate?"
Starting from basic syntax that makes you sigh,
to gradually building my own game world.
And the sense of pride when seeing many players connect to that world.
The more I study, the more I find myself unable to escape the charm of MMORPGs.
I hope you can feel the same excitement that I felt.

 

More

Co-instructor

Curriculum

All

78 lectures ∙ (17hr 20min)

Course Materials:

Lecture resources
Published: 
Last updated: 

Reviews

All

13 reviews

4.8

13 reviews

  • kilo1089님의 프로필 이미지
    kilo1089

    Reviews 18

    Average Rating 4.9

    5

    100% enrolled

    Since I'm building it from the very bottom, I don't fully understand everything yet, but I think it will be very helpful once I go through the material again.

    • hkr04164242님의 프로필 이미지
      hkr04164242

      Reviews 1

      Average Rating 5.0

      5

      31% enrolled

      It's good

      • sarah0310343542님의 프로필 이미지
        sarah0310343542

        Reviews 4

        Average Rating 5.0

        5

        54% enrolled

        I've only finished about half of the course, but I was so impressed by the quality that I came here to leave a review first... The course covers everything from the very basics of AI agents to concepts and hands-on practice, which has greatly deepened my understanding of agents like Claude Code. I didn't expect it to be this detailed, and it has been a huge help in my AI studies. I've also gained the confidence that I can customize agents myself to increase efficiency. I started taking this course because I use Claude Code a lot while studying game development, but I think it can be applied to all fields. I'm really looking forward to the Unreal Engine part that I'll be taking next.. Great lecture, highly recommended!

        • asdsa1127309님의 프로필 이미지
          asdsa1127309

          Reviews 7

          Average Rating 5.0

          5

          49% enrolled

          I didn't even dare to think about implementing AI Agent technology because it felt so far out of reach, but this lecture was incredibly helpful as you taught me everything and walked me through the implementation from the ground up. It was a course that truly helped wash away my fears regarding AI Agents. Additionally, I feel like I learned a lot of very clean code writing patterns as a bonus. From the curriculum to the quality of the instruction, everything was perfect. Thank you.

          • jimojjing님의 프로필 이미지
            jimojjing

            Reviews 3

            Average Rating 5.0

            5

            100% enrolled

            Unreal was just the starting point; the core of this lecture was about how to utilize LLMs and agents. While listening to the lecture, I realized that I hadn't been utilizing AI properly until now. In particular, understanding how LLMs call tools and the workflow through which agents perform tasks has significantly shifted my perspective on AI utilization. It would have been even better if the lecture had covered how to connect other LLMs, but I believe that if you understand the structures and principles explained in the course, you should be able to expand and integrate them yourself. I think this lecture is sufficiently helpful not only for those interested in Unreal but also for anyone interested in AI agent structures and LLM utilization methods.

            Rookiss's other courses

            Check out other courses by the instructor!

            Similar courses

            Explore other courses in the same field!

            25% off for new members

            $258.80

            25%

            $338.80