Write once, serve everywhere: how Optiver built the context layer for agents

Your data platform has a new customer. It reads at machine speed, pays for every token it loads, and can't ask a colleague which table to trust. At Optiver, a growing share of data consumption now comes from agents. In this session, an AI agent shows how Christophe Godefroy's team used DataHub to build a context layer that ingests, infers, curates, and activates context for those agents. On real researcher questions, governed context raised agent accuracy from 42% to 74%.