Using Git with coding agents. Simon Willison shows Git as the authoritative context, audit trail, and control plane for coding agents—seed sessions, manage branches, and undo mistakes. Outcome engineers gain a concrete pattern for reproducibility, rollback, and human-in-the-loop control when agents edit code (Principles 03, 06, 13).

Tinybox — offline AI device, 120B parameters. Tinybox ships affordable, on‑prem machines that run and train large models locally, promising offline, private model hosting. This changes deployment trade-offs for outcome teams: you can own data, latency, and verification without cloud dependence (Principles 07, 04).

Flash-MoE: Running a 397B Parameter Model on a MacBook Pro with 48GB RAM. The repo demonstrates streaming weights from SSD and hand-tuned Metal kernels to run a 397B MoE locally and achieve production-quality tool-calling. It surfaces practical engineering techniques (weight streaming, hardware kernels) that make local inference and agent tool-chains feasible for product builds (Principles 07, 11).

Hands-on: Gemini task automation on mobile — impressive but slow and error-prone. The Verge tests Gemini’s mobile automation, which performs full tasks like ordering and booking but suffers latency and reliability issues. Treat mobile agent automation as an end-to-end reliability and UX engineering problem: instrument verifications, design graceful degradation, and plan clear human handoffs (Principles 09, 15).

Meet the CFO who turned Adobe’s finance department into an AI lab. Fortune profiles Adobe’s CFO deploying agentic assistants across forecasting, contract review, and inbox automation inside finance. This is a real-world blueprint for scaling agentic workflows inside a business—outcome engineers must build artifact pipelines, audit trails, and governance to move from pilots to production (Principles 09, 06, 10).