Five pieces worth your time this week. Ahmad Alfy makes the case that URLs are state containers, not just addresses. Shrivu Shankar documents every Claude Code feature he actually uses in production. Yusuf Aytas argues AI has made technical interviews functionally broken. Wu Xiaoyun details how he cut TikTok $300k in costs during a single internship. And TOON is a new serialization format built specifically to reduce LLM token consumption.
The URL piece is the anchor. Alfy's framing reframes something developers interact with hundreds of times daily as an underused data layer. The Claude Code breakdown is the most practically dense item: not a review, a working inventory. The TikTok story has a number attached to a name and a timeline, which is rarer than it should be.
TOON is the one to watch longer term. A serialization format targeting token efficiency is a direct response to the cost structure of running LLMs at scale. If it gains traction, it belongs in the same conversation as MessagePack and CBOR. Read the full newsletter at changelog.com/news/168.
[READ ORIGINAL →]