Add next phase of design notes

Author: linas

Design-F.md

@@ -100,7 +100,7 @@ But this is terrible, because the URL is ugly.
 
 Overall: the OO API seems pointless and useless.
 
-### Stateless Executation
+### Stateless Execution
 A meta-problem might be that we needs something in between `cog-execute!`
 and `cog-set-value!`. The problem with `cog-execute!` is it takes just
 one argument: the Atom to execute. Thus, a Link must be created to wrap

Design-G.md

@@ -0,0 +1,25 @@
+Design Notes G
+==============
+September 2025
+
+Notes about the mechanics of running AST's on GPU's.
+
+The base implementation of the GPU interfaces is now sufficiently
+advanced that the next steps open up: the ability to take a given
+Atomese AST and run it on a GPU. That is, to create a GPU-cenetric
+library of arithmetic functions, and then take a specific abstract
+syntax tree (AST) written as e.g. `(AccumLink (TimesLink A B))` and
+run it on the GPU. But what is the API to this, and what is the
+mechanics of doing this?
+
+One possibility is to have PlusGPULink and TimesGPULink, and then have
+RuleLinks to transpose the AST from its "abstract" form to these
+concrete Atoms. This is problematic because RuleLinks remain difficult
+to work with and apply. They seem like a great idea but are tough.
+There are also little details that threaten coherency: stuff that might
+need to run locally before getting pushed off to the GPU, and the rules
+for these could be very arcane and detailed and prone to bugs.
+
+Another possibility is to have an `(ExecOnGPULink AST)`, similar to
+`(PureExec AST)`
+

Design.md

@@ -19,6 +19,7 @@ actual design.
  unimplemented. Next step should be `PlusLink` and `TimesLink` running
  on the GPU.
 * [Design-F](./Design-F.md) ponders RNN/LTSM embeddings of the AtomSpace.
+* [Design-G](./Design-G.md) ponders mechanics of running AST's on GPU's.
 
 Structuralism Overview
 ----------------------