Google DeepMind has made the Interactions API the default interface for its Gemini models and agents, completing a shift away from the older `generateContent` interface that has anchored Gemini development since the models first shipped. The move, announced on Google's developer blog on June 26, elevates an API that has been in beta since December 2025 into the primary way developers are expected to build on Gemini.

The older `generateContent` interface still functions, Google stressed, so existing applications will not break overnight. But the company is now drawing a firm line: new agent capabilities will only ship through the Interactions API going forward. For more context on this story, see our ongoing more AI stories.

> "Interactions sets the stage for the new era of Agents."
>
> — Logan Kilpatrick, Google developer relations lead

Google has published a migration guide to help developers move from the previous interface, and the new API is now the default inside Google AI Studio and across all official Gemini documentation.

What the Interactions API Changes

The most significant change is structural. The previous interface organized conversations around roles — labels like `user` and `model` — mirroring the familiar chat format that most large language model APIs adopted. The Interactions API replaces that role-based structure with typed steps, where every action, from a user's input to a function call to a model response, is its own explicitly defined step.

The result, according to Google, is a schema better suited to complex agent workflows rather than simple back-and-forth chat. Agents routinely need to chain together tool calls, run in the background, and branch based on intermediate results — patterns that the old role-based conversation model handled awkwardly.

Built-In Agent Primitives

Since entering beta in late 2025, the Interactions API has accumulated a set of features that Google is positioning as first-class agent building blocks:

  • Managed Agents that run inside their own Linux sandbox, giving developers an isolated environment for executing agent-generated code and commands.
  • Background execution for long-running tasks, allowing agents to continue work asynchronously rather than blocking on a single request.
  • Tool chaining with Google Search and Google Maps built in, so agents can pull real-time information and location data without developers wiring up those integrations manually.
  • Media generation spanning images, music, and speech, letting a single agent interface produce multimodal output.

Google has also simplified parts of the schema since the beta, removing friction that early adopters had flagged.

Flex and Priority Modes

Developers building on the Interactions API can choose between two execution modes. Flex mode cuts costs by roughly 50 percent, trading some latency for cheaper inference — useful for high-volume or background work. Priority mode instead optimizes for speed, routing requests for the fastest response.

The cost lever matters for Google competitively. Enterprise AI spending has become a central concern across the industry, with companies increasingly scrutinizing token costs as model usage scales. Offering a 50 percent cost reduction mode gives Gemini developers a concrete way to contain bills without leaving the platform.

Why the Shift Matters

The transition reflects a broader industry pivot from chat interfaces toward agent interfaces. Where early LLM products were framed as conversational assistants, the current wave of developer tooling is built around autonomous and semi-autonomous agents that complete multi-step tasks across tools and services.

By making the Interactions API the default — and reserving new agent features exclusively for it — Google is nudging its entire developer base onto infrastructure designed for that agentic future. Developers who stay on `generateContent` will retain compatibility but will progressively miss new capabilities.

The change arrives as competition over the agent layer intensifies. Anthropic has leaned into agent infrastructure with its Claude Code and Claude Tag products, OpenAI continues to expand its agent and computer-use capabilities, and open-source frameworks are multiplying. Google's bet is that owning the developer interface — not just the underlying model — will be decisive as agentic applications move into production.

Google has not announced an end-of-life date for the `generateContent` interface, but the clear direction is that the Interactions API is now the path forward for anyone building on Gemini.

---

Stay Ahead of AI

Get the latest AI news, analysis, and breakthroughs — all in one place.

Read more AI news →