# Gemini Interactions API exposes agent steps individually

By Steven Van · 2026-05-08

Google AI Studio will add mid-flight steering and async tool calls on 8 June 2026, alongside breaking changes for existing integrations.

Google AI Studio's Gemini **Interactions API** now breaks agentic workflows into steps, with each thinking pass or tool call recorded as its own step rather than bundled into a single response. Developers building multi-step agents can inspect and act on that structure directly.

Mid-flight steering and async tool calls are coming **8 June 2026**, according to the [announcement](<https://ai.google.dev/gemini-api/docs/interactions-breaking-changes-may-2026>), which also flags breaking changes for existing integrations. Developers using [Google AI Studio](<https://creatorstoolbox.com/tools/google-ai-studio>) should check the linked docs before that date.

[![](<https://ai.google.dev/static/site-assets/images/share-ais-03.png>)](<https://ai.google.dev/gemini-api/docs/interactions-breaking-changes-may-2026>)

![Google AI Studio](<https://www.google.com/s2/favicons?domain=aistudio.google.com&sz=128>)

Google AI Studio

Build, test, and deploy AI apps with Gemini — Google's free web-based IDE for generative AI prototyping.

[View Google AI Studio →](<https://creatorstoolbox.com/tools/google-ai-studio>)

Sources

[Original source](<https://ai.google.dev/gemini-api/docs/interactions-breaking-changes-may-2026>)

[Read Gemini Interactions API exposes agent steps individually on Creators Toolbox](<https://creatorstoolbox.com/blog/google-ai-studio-multi-step-interactions-api>)

---
Canonical source: https://creatorstoolbox.com/blog/google-ai-studio-multi-step-interactions-api