But the general use case is not the most efficient for a greenfield to-be fully managed by an agentic system code-base. It is built to be good around the scaffold(programming like humans) and not the actual problem space.
replyAnthropic's target should be a codebase designed for agentic comprehension from the first commit. Here the codebase adapts to the agent. You can enforce conventions, structured metadata, semantic indexing, explicit dependency graphs. Whatever makes the agent's job trivial rather than heroic.
The large majority of coding is maintenance work, not greenfield development. Even if you are doing greenfield development, it won't be long before it is maintenance.
reply[dead]
reply> So it is optimized for the general case, using robust tooling that works everywhere
replyWhere "robust tooling" is "grep with various regexes while completely missing the big picture even in small codebases"
Nothing wrong with "grep with various regexes".
replyAnd in my experience it hardly ever "misses the big picture". When it misses is small stuff I'd miss myself quite often anyway.