DEDUPED Hot AI repos. Real use cases. What people build. Get the Monday 5

How to chain agents into a deterministic workflow with Adk Python

Use ADK's Workflow runtime to pass work between specialised agents in a fixed graph.

Short answer: Install with pip install google-adk; then Define each step as an Agent with its own instruction; create a Workflow with edges such as START to the first agent to the second, as in the README example. All steps.

Who it is forA builder whose single agent keeps skipping steps in a multi-stage process.

The problemLeaving every step to one model's judgment makes multi-stage tasks unpredictable.

The outcomeA Workflow whose edges run one agent after another, with routing, loops and retries available.

What you need first

  • Python 3.11 or later
  • pip
  • A model to call, such as Gemini; ADK is model-agnostic
  • For Cloud Run deploys: a Google Cloud project; for local containers: Docker

Steps

  1. Install with pip install google-adk
  2. Define each step as an Agent with its own instruction
  3. Create a Workflow with edges such as START to the first agent to the second, as in the README example
  4. Add routing, fan-out and fan-in, loops, retry or human-in-the-loop steps as the Workflow Runtime supports
  5. Run it with adk run or adk web and check each node's output

Prompt for your agent

With google/adk-python, build a Workflow where <agent A> does <step 1> and passes its output to <agent B> for <step 2>. Add a retry on <step> and a human-in-the-loop check before <final step>, then run it in adk web.

Caveat. Only the simplest Workflow is shown in the README; routing and loops are in the guides under docs/guides.

Use this in your AI agent

Ask your agent

Using the Deduped MCP, get the setup steps for google/adk-python and "Chain agents into a deterministic workflow"

Add Deduped to Claude Code once

claude mcp add --transport http deduped https://dedupedai.com/mcp

Cursor, ChatGPT, others

Share: X LinkedIn Hacker News Email Copy link

Third-party repos we don't own. AI-written use cases can be wrong. Disclaimer