The Sequence AI of the Week #773: The Week Google Turned Gemini Into an Agent Runtime
Inside Gemini Deep Research Agent and Interactions API.
Google’s agentic releases last week weren’t “yet another model drop.” They were Google quietly shipping an agent runtime—and then immediately proving it works by dropping a managed research agent that basically demands that runtime to function.
On December 11, 2025, the Gemini API changelog logged two entries that, together, mark a clean architectural pivot: the Interactions API (Beta) and the Gemini Deep Research Agent (Preview).
If you build agents for a living, you already know the pattern: most of your complexity isn’t “getting the model to answer.” It’s dealing with the messy reality that real agent workflows are stateful, tool-heavy, and often long-running. You build a planner, then a tool router, then a memory store, then a retry system, then an audit log, then a UI that explains what’s happening. By the time you’re done, the LLM call is the smallest line item.