I like how this gives an older answer to a problem for the new ai issues we are facing. I have seen this more times than I can count, and until AI's Compression and packet context is rejiggered and rethought about, it will only get worse as we cram more into a space that doesnt have all of the tools to completely understand.
As always, code can be written, but you must be able to verify before you post to a production atmosphere. Maybe the play is to do all of this, and post it to a test layer of production and smoke and play test it all before pushing to build on production as final.
Exactly — the test layer you're describing is what happens naturally when the units are small enough. Each canton can be smoke tested in isolation because its boundaries are fully declared. The composition gets tested as a graph of known units, not a blob of generated code. The 'test layer of production' becomes trivial when 90% of what you're deploying is already proven.
I like how this gives an older answer to a problem for the new ai issues we are facing. I have seen this more times than I can count, and until AI's Compression and packet context is rejiggered and rethought about, it will only get worse as we cram more into a space that doesnt have all of the tools to completely understand.
As always, code can be written, but you must be able to verify before you post to a production atmosphere. Maybe the play is to do all of this, and post it to a test layer of production and smoke and play test it all before pushing to build on production as final.
Exactly — the test layer you're describing is what happens naturally when the units are small enough. Each canton can be smoke tested in isolation because its boundaries are fully declared. The composition gets tested as a graph of known units, not a blob of generated code. The 'test layer of production' becomes trivial when 90% of what you're deploying is already proven.