Descriptions:
Thor Schaeff and Philipp Schmid, both from the developer experience team at Google DeepMind, led a hands-on workshop at AI Engineer on building conversational agents with the Gemini API. The session walks attendees through creating API keys in Google AI Studio (accessible at ai.dev), then progresses into building a working coding agent capable of executing shell commands and responding to natural language instructions.
A key focus of the workshop is the Gemini Interactions API, a server-side conversation state management system that stores input and output history linked to a unique interaction ID. On the free tier, state persists for one day; on paid plans, it extends to 55 days. Schaeff and Schmid explain how to use these interaction IDs to branch conversation histories — enabling parallel workflows that all stem from a single base prompt — and confirm that the Interactions API is coming to Vertex AI.
The session also covers practical limits: Gemini models currently support up to one million tokens of context, and exceeding that ceiling returns an error, though the team notes that context compaction techniques are in development. All demonstrations run on the free tier and require only a standard Google account, making this a low-friction starting point for developers looking to build stateful, tool-using agents on top of the Gemini platform.
📺 Source: AI Engineer · Published April 30, 2026
🏷️ Format: Hands On Build
