Essay Refinement Loop – Loop Agent¶

Demonstrates a LoopAgent that iterates sub-agents until a maximum iteration count. The scenario: an essay refinement workflow where a critic evaluates the draft and a reviser improves it, repeating up to 3 times until quality standards are met.

Real-world use case: Essay refinement loop where a writer drafts and a critic provides feedback iteratively. Used by content teams to improve quality through structured iteration.

In other frameworks: LangGraph models loops as conditional back-edges in a StateGraph, requiring a routing function to decide continue vs stop (~25 lines). adk-fluent uses * N for fixed iterations or * until() for conditional loops.

Pipeline topology: ( critic >> reviser ) * 3

Tip

What you’ll learn How to compose agents into a sequential pipeline.

Source: 06_loop_agent.py

adk-fluent

from adk_fluent import Agent, Loop

loop_fluent = (
    Loop("essay_refiner")
    .max_iterations(3)
    .step(
        Agent("critic")
        .model("gemini-2.5-flash")
        .instruct(
            "Evaluate the essay for clarity, structure, grammar, "
            "and argument strength. Provide specific, actionable "
            "feedback. If the essay meets a high quality bar, "
            "say APPROVED."
        )
    )
    .step(
        Agent("reviser")
        .model("gemini-2.5-flash")
        .instruct(
            "Revise the essay based on the critic's feedback. "
            "Improve clarity, fix grammatical issues, and strengthen "
            "weak arguments while preserving the author's voice."
        )
    )
    .build()
)

Native ADK

from google.adk.agents.llm_agent import LlmAgent
from google.adk.agents.loop_agent import LoopAgent

loop_native = LoopAgent(
    name="essay_refiner",
    max_iterations=3,
    sub_agents=[
        LlmAgent(
            name="critic",
            model="gemini-2.5-flash",
            instruction=(
                "Evaluate the essay for clarity, structure, grammar, "
                "and argument strength. Provide specific, actionable "
                "feedback. If the essay meets a high quality bar, "
                "say APPROVED."
            ),
        ),
        LlmAgent(
            name="reviser",
            model="gemini-2.5-flash",
            instruction=(
                "Revise the essay based on the critic's feedback. "
                "Improve clarity, fix grammatical issues, and strengthen "
                "weak arguments while preserving the author's voice."
            ),
        ),
    ],
)

Architecture

        graph TD
    n1(("essay_refiner (loop x3)"))
    n2["critic"]
    n3["reviser"]
    n1 --> n2
    n1 --> n3

Equivalence¶

assert type(loop_native) == type(loop_fluent)
assert loop_fluent.max_iterations == 3
assert len(loop_fluent.sub_agents) == 2
assert loop_fluent.sub_agents[0].name == "critic"
assert loop_fluent.sub_agents[1].name == "reviser"