Terrastruct's AutoLayout Algorithm (TALA) is now open-source under MPL-2.0; it's an orthogonal layout engine for software-architecture diagrams that emphasizes aesthetic objectives (symmetry, clustering, flow), supports fixed or partial node positioning, and is bundled in D2 v0.9.0. The algorithm trades deterministic stability and DAG-oriented layouts for more whiteboard-like arrangements, uses randomized seeds, and can scale nonlinearly on large diagrams (benchmarks linked).

Dan Luu ran a systematic evaluation of coding agents instructed to use 26 different testing techniques/skills (implementing Zstd in Rust) and found that agents are generally poor at testing: default prompts often do as well or better, named techniques and popular skills rarely improve correctness, and agents mostly apply techniques superficially. On high effort runs fuzzing and property-based testing showed modest gains while TDD and most formal-methods/skill prompts underperformed.