Vibe Coding
Pattern: context-budget tooling turned into a product category with published token benchmarks
Three independent things shipped on 2026-09-20 and 09-21 that all treat the agent's context as a budget to be spent rather than a file to be pasted: tokencut folds tool output 1,562 to 174 tokens with recovery by reference, briefd serves a 1,800-token compiled bundle against 12,869 for a full CLAUDE.md, and Qwen Code v0.24.2 started attributing each context file to its extension and stating what it costs (#12318). What is new is not the idea but the evidence: all three publish a reproducible number instead of claiming 'less context'. The benchmarks are self-reported, but each ships the harness that produced them.
Source
↳ Follow the thread