Graphic Engine is technically harder, specially when you try to achieve both eye candy effects and performance. However most of time both relies on well known algorithms/concepts that at most get some customization, so nothing really new. Besides, most of time it is harder and more expensive to expand or implement new feature/engine on old software than create from scratch.

What differs AI from graphics is that once you have done the graphic engine, and supposing no bugs and good performance, you are done, you load your objects and see them on the screen. AI instead requires a new phase of work which is the design of the agent behavior and how to use all tools that your system implemented (eg. behavior tree, state machine, influence map, etc...).

Again, let's not mystify DC specially with falcon 4 as standard because the way their implemented it, is not even close to a modern and better approach, although they did a very good job!