Video summary
How to Make Games in Unreal Engine 5 with AI
Main summary
Key takeaways
Summary
The video demonstrates an AI-assisted workflow for building a game in Unreal Engine 5, using the tower-defense game Blob Siege as an example. The presenter’s central point is that AI can generate and revise complex gameplay code, but the developer still provides the vision, makes design decisions, tests results, and guides each iteration. The demo games shown at the beginning were created through many prompts and rounds of testing—not with a single prompt.
Main Ideas and Lessons
- Treat AI as a capable collaborator, not an automatic game designer. Give it detailed context, then review and test what it creates. Building feature by feature keeps the developer in control.
- Use Unreal’s existing strengths. The editor, tools, and marketplace assets let you adjust AI-generated work and reuse models, animation systems, and other assets rather than generating everything from scratch.
- AI is powerful but imperfect. The presenter finds it effective at generating gameplay code, but notes that it can be slow through Unreal MCP and may struggle with animation, locomotion, and game feel. Results are also unpredictable: identical prompts can produce different outcomes.
- Iteration is a major benefit. AI can shorten the cycle of trying a feature, evaluating it, and changing or discarding it. The presenter recommends testing frequently rather than waiting until many features have accumulated.
- Learning programming still has value. The presenter recommends learning programming fundamentals to understand AI’s output and make better decisions, while arguing that AI can handle much of the code-writing work.
- The presenter distinguishes AI-generated code from AI-generated art. They describe Steam’s policy, as they understood it in October 2026, as requiring disclosure for AI-generated player-facing content—such as art, audio, narrative, or 3D models—but not AI-generated code. This is the presenter’s account of the policy.
Suggested Workflow and Instructions
Connect an AI Desktop Tool to the Unreal Project
- Use a desktop application such as Claude Code or Codex and point it at the project folder.
- The presenter recommends current, capable models, while noting that model rankings change over time.
- Provide screenshots or images when visual positioning or appearance matters.
Choose How the AI Will Interact with Unreal
- Unreal MCP can inspect or edit editor content, such as levels and Blueprints, while Unreal is open. The presenter finds it useful for targeted editor tasks but slower for substantial gameplay programming.
- The tutorial mainly uses AI-generated C++, along with scripts and headless editor runs to inspect or test the project.
- If you prefer a script-based approach, explicitly tell the AI not to use MCP.
Set Up Unreal MCP (Optional)
- Enable the MCP plugin and the relevant tool-set plugin.
- Enable automatic server startup in Unreal’s editor preferences.
- Generate the appropriate client configuration for the AI application.
- The AI can also be asked to perform these setup steps.
Prepare the C++ Build Environment
- Install Visual Studio Community, not Visual Studio Code.
- Select the required .NET and C++ development workloads, including Game Development with C++, as well as the Unreal-related tools and SDK components.
- Unreal commonly needs to be closed for C++ changes to compile. The AI can run the editor headlessly to compile and test, then reopen Unreal when finished.
Give the AI Project-Level Rules and Organize Its Context
- Save persistent instructions in a project memory file such as
CLAUDE.mdorAGENTS.md; separate chat threads may not share context. - Example rules from the tutorial include avoiding MCP unless requested, placing exposed settings near the top of the Details panel, providing useful descriptions, and reopening Unreal after compilation.
- Use separate threads for distinct features when helpful. If a long thread contains important context, ask the AI to create a detailed handoff for a new thread.
Prompt with Context, Then Steer the Result
- Describe the feature, desired behavior, relevant asset names, edge cases, and adjustable settings in natural language.
- A long, informal prompt is acceptable; the presenter favors providing plenty of context over following a special prompt formula.
- Ask for properties to be exposed in the Details panel so gameplay values can be tuned without repeatedly editing code.
- If the result is visually off, provide a screenshot or reference image.
- Test the result, report failures clearly, and ask the AI to investigate and fix them. Correct its direction if it misunderstands the task.
Building and Refining Blob Siege
The tutorial develops the game in stages:
- Create turrets that scan for targets, lock on, and shoot projectiles. Add adjustable properties and editor visualizers.
- Create enemy blobs with movement, health, hit reactions, and a death effect. Then change the turrets to target the blobs rather than the player.
- Add wave-based gameplay with spawn and attack volumes, a central settings actor, escalating waves, player and tower health, and purchases for different defenses.
- Add friendly blobs, an alpha blob that targets the player, and a spline path for enemies to follow.
- Shape the level around the gameplay, test enemy paths and collision, and adjust settings for balance.
- Add pause, settings, and start menus, then refine their visual style and interactions.
Sources and Speakers
- Speaker: The video’s narrator and instructor, Unreal Sensei. No other audible speaker is identified.
- Tools and platforms discussed: Unreal Engine, Unreal MCP and Epic Games’ Unreal skills, Claude Code, Codex/ChatGPT, Visual Studio Community, Fab, Blender, and Steam.
- Other creators, products, and sources mentioned: Matt Botack’s “Grill Me” skill; John Carmack and Notch; Caleb Longbuyer’s Advanced Locomotion System V4; Epic Games’ Game Animation Sample; and the 3D-generation tools Hunyuan, Tripo, and Meshy.
- Promoted source: Unreal Sensei’s Unreal Masterclass, including its AI course.
Rate this summary
Your feedback will help improve summaries.
Improve this summary
Reprocess with a stronger model when the summary feels incomplete or inaccurate.
Translate summary in another language
Ask questions to this video
Chat for follow-up questions, clarifications, and source-backed answers.