It all kicked off with a shiny new launch. Google had just unveiled Gemini’s multimodal features–a technology that looked promising enough to potentially solve human-browser interaction problems. The engineering team figured it’d be a quick, low-risk experiment. One team tried it. Then another. Then a third. But while the concept was exciting, the actual tool wasn’t quite there. It was early-stage, not ready for production, and definitely not tailored to the team’s use case. “It was a...