That's a false choice, and pie menus are the way out of it. The same menu can lead novices by showing them labels, follow intermediate users who know roughly where things are, and get out of the way of experts who flick in the direction without waiting for it to appear. One UI, no "simple mode" and "advanced mode", and the novice turns into the expert just by using it, by self-revealing rehearsal! I went into that and the research behind it upthread.
As I titled my 1998 Xerox PARC talk: "Natural Selection: The Evolution of Pie Menus", reviewed here:
https://baychi.org/calendar/19981013/
> Pie menus appealed to Hopkins because they conform to the rule that the user's attention should not be squandered. A pie menu selection offers direct feedback versus a user selection that brings up intermediate menus. The programmed pie menu ignores users' gestures except the motions to select or cancel. Hopkins uses pie menus because they take advantage of Fitts' Law, which relates fast selection speed and low error rate to the pie menus' large target size and small distance between selections.
The other half is letting users rearrange it themselves, without registry hacks. HyperCard set the precedent in 1987: the user interface was made of the same stuff as the content, so anyone could open it up and change it. Arthur van Hoff's HyperLook took that to NeWS and PostScript and networking: users could build and edit not just their own nested linear and pie menus, but also customize the live user interfaces of running apps, including the SimCity I ported to it, by copying, pasting and configuring scriptable components from "warehouse" stacks. We were taking it all the way to the window frames themselves, with a HyperCard-like window manager where you could build your own frames out of the same parts, a built-in structured PostScript graphics editor you can use in your own stacks, windows with any shape (like my Lava Lamp shaped Cellular Automata Machine), and you can compose your own custom applications inside them.
HyperLook Cellular Automata Machine:
https://github.com/SimHacker/WillWrightShowForFood/blob/main...
HyperLook SimCity:
https://github.com/SimHacker/WillWrightShowForFood/blob/main...
Why we also need a non-editable runtime system (with a removable UI editor): User Interface Vandalism:
https://github.com/SimHacker/WillWrightShowForFood/blob/main...
My comments in the "What HyperCard Might Have Been" discussion:
https://news.ycombinator.com/item?id=22282361
I also discussed HyperLook and HyperCard with Alan Kay, who said "This work is so good -- for any time -- and especially for its time", and also "I wish he had done a real end-user language for this":
https://github.com/SimHacker/WillWrightShowForFood/blob/main...
Simon Schneegans's Kando carries that spirit today. Its menu editor is direct manipulation all the way down: you drag and drop items and submenus right into the pie, see exactly where they'll go, and put the things you use most in the directions that feel right to your hand. That's how a power user should customize a menu, and a novice can do it too, because it's just dragging things around:
I took that idea further in MediaGraph, a music navigation prototype I made in Unity for Will Wright at the Stupid Fun Club in 2010. It has two layers of pie menus.
The first layer is ordinary nested pie menus for editing the content, with live feedback: as you move around the menu that sets a song's size or its "biome" (a cellular automaton growing around it on the map), you see the effect in the world before you commit. Same biomes merge, different biomes compete for territory.
The second layer is the content itself. The songs are laid out on a map and connected by roads, and the map is a geographic, networked pie menu. Touch one song to another -- kissing them together -- to make a road between them, and kiss them again to break it. Flick on a song in a direction and it fires you down the road to the next song that way, like the cannons in Super Mario World. Point at a song and it plays an earcon, click it to play.
Following links, editing links, panning, zooming and flicking are all the same kind of gesture, with continuous zoom and pan and inertia. As I wrote at the time, what I was proudest of was the "Fahrvergnügen" of connecting and disconnecting places by bumping them together, and traveling down the roads by flicking between them, "without creating another mode". The menu editor disappears and dissolves into the content, by replacing hierarchy with geography.
MediaGraph demo:
https://www.youtube.com/watch?v=2KfeHNIXYUc
How it works, with a timeline from DreamScape and iLoci to MediaGraph:
https://github.com/SimHacker/WillWrightShowForFood/blob/main...