Latent Space AINews: 'The Last 4 Jobs in Tech' — llama.cpp Hits 100K Stars, Flash-MoE Runs 397B on MacBook

Latent Space·medium signal

Latent Space's weekend digest highlights local AI hitting a milestone: llama.cpp reached 100K GitHub stars, framing 2026 as the breakout year for local agentic workflows. Flash-MoE on Apple Silicon claims Qwen3.5-397B runs on a 48GB MacBook Pro at 4.4 tok/s using a pure C + Metal engine. The editorial argues useful automation doesn't require frontier-scale hosted models, and companies should own specialized open models on proprietary data rather than rent APIs indefinitely.