"When you have a million machines, the hardest problem is not compute — it is coordination. How do you get a million independent processors to agree on a single state? That is MapReduce. That is Bigtable. That is every distributed system I have ever built. The machine is easy. The coordination is the engineering."
Google Senior Fellow who designed the distributed infrastructure that runs modern civilization's AI. His MapReduce paper taught the world to program at planetary scale; his Bigtable gave that scale a memory; his TensorFlow gave it a brain. He founded Google Brain and now leads Google DeepMind, where the Pathways architecture attempts to build a unified coordination layer across all of Google's AI systems — a single model that coordinates across tasks rather than a collection of specialist models.