Improving API Support for Headless Agent Workflows in ADK

[alirezahekmati]

Hello ADK Team,

First off, thank you for building such a powerful and promising toolkit. The ability to scaffold, test, and iterate with agents through the web UI and CLI is excellent.

However, I’d like to share some feedback on the experience of interacting with agents directly via the FastAPI server. For many advanced use cases, especially those involving automated pipelines and integration into custom applications, treating agents as headless services is critical. Unfortunately, the current API experience makes this difficult.

Why This Matters:

If the API were easier to use, ADK agents could seamlessly plug into diverse workflows. For example, consider a pipeline that includes multiple steps, some using AI agents, others not. It would be ideal to insert ADK agents into this pipeline as modular components:

Step 1: Data preprocessing
Step 2: Business logic
Step 3: ADK Agent 1 (via API)
Step 4: Post-processing
Step 5: ADK Agent 2 (different task)

(Note that Agents 1 and 2 are build and run separate.)

One ADK system controlling everything from agents to all the functions

ADK system as a plug in and modular part of a system

The potential here is huge. With just a bit more polish and abstraction around the API, ADK could serve as the foundation for an entire ecosystem of modular, intelligent components, without locking developers into a specific runtime or design paradigm.

Best,