Tim, We Need Unreal Engine 7, Not 6

Hey Tim, I think you guys should start working on Unreal Engine 7.

LLMs have hit the video game industry like an atomic bomb. Few industries are experiencing such a dramatic transformation in how work gets done, especially across so many specialized disciplines.

After spending several months working with Rider, Codex, and now MCP, Unreal Engine feels completely outdated. It’s a strange feeling to be working with incredibly advanced AI technology while using tools and workflows that suddenly feel obsolete.

Unreal was designed for teams of specialized human developers, where all those boxes, nodes, and wires made perfect sense. But that entire approach needs to be reconsidered.

Unreal Engine 6 might be a step forward, but I wouldn’t invest too heavily in it. What we really need is a complete rethink of how the engine works and how games are developed.

For a game designer like me, LLMs haven’t replaced my role. In fact, the way I work hasn’t fundamentally changed at all.

I used to write a game design document, and programmers, UX designers, and art directors would turn those ideas into reality.
The technical work—writing perfect syntax, configuring parameters buried in nested menus, resolving compilation dependencies—was always a means to an end, never the end itself.

Talking to Codex through Rider’s chat—whether it’s to request an additional feature, report a bug, or point out something it got wrong—feels almost exactly like talking to a programmer on Slack at work.

The tools are changing dramatically. The creative process feels remarkably familiar.

1 Like

I think it would be important for pretty much everything to have a solid C++ foundation.

Clearly, the worst offenders are StateTree and Animation Blueprints. They burn through tokens like crazy, and I can practically see my poor Codex suffering as it tries to understand and modify them.

And when it has no choice but to switch to computer control mode and physically click its way through the Unreal Editor, it feels more like an endurance test than AI-assisted development.

I’m becoming increasingly convinced that the more logic we keep in C++, and the less we rely on messing around with nodes in the editor, the better Unreal and AI agents will get along.

As a funny anecdote, take a look at what Codex wrote in Rider during one of its adventures with Unreal:

I need to wait; there’s no point in polling if the tools aren’t executing yet. It’s a bit frustrating, but I see that I can call for a log check after 15 seconds. I know I could use clock.sleep(15), but if there’s no progress with the tool, that doesn’t help. So, I guess I just have to be patient and wait for the right moment to check in on things.

Poor thing.