Summary
A very thin evening slot with one signal post and one opaque link. Gary Marcus reposted (via bayesiansapien) a victory-lap take on Claude Code: 53 symbolic tools, ~500k lines of symbolic code wrapped around an LLM, which he frames as vindication for neurosymbolic AI rather than pure-LLM scaling. The pointer is to ccunpacked.dev, a source-level walkthrough of Claude Code's agent loop, tool system, and multi-agent orchestration. Everything else in the slot is one undecodable x.com article link.
Posts
- Gary Marcus: Claude Code is the most neurosymbolic system he has ever seen (@GaryMarcus · ccunpacked.dev). Marcus reads Claude Code's 53 tools plus ~500k lines of orchestration code around a frontier LLM as proof that progress is coming from borrowing classical-AI and CS scaffolding, not from scaling LLMs alone. The linked site is a source-level dissection of the agent loop, tool registry, and unreleased features. Worth a click for the architecture explorer alone. See Claude Code architecture (04-17) and the 04-19 follow-up for prior wiki notes on the same codebase.
- Opaque article repost (@ns123abc). Bare
x.com/i/article/link with no extracted content. Click through to read.