Introducing the Hanzo Agent Framework

Automation in commerce has been rule-based: if this, then that. Today we are releasing the Hanzo Agent Framework, enabling intelligent automation that reasons about goals, not just conditions.

Beyond Rule-Based Automation

Traditional automation:

IF cart_abandoned AND time > 1_hour
THEN send_email(abandoned_cart_template)

This works for simple cases. It fails when:

• Multiple conditions interact
• Optimal action depends on context
• Goals conflict (retention vs. margin)
• Situations are novel

Agent-Based Automation

Agents operate differently:

GOAL: Maximize customer lifetime value
CONSTRAINTS: Maintain 20% margin, respect communication preferences
OBSERVE: Customer abandoned cart, has purchased twice before, high-value segment
REASON: Previous purchases indicate price sensitivity, but also brand loyalty
ACT: Offer 10% discount valid 48 hours, single reminder email

Agents reason about goals, not just execute rules.

The Framework

Hanzo Agent Framework provides:

Agent Definition

from hanzo.agents import Agent, Goal, Constraint

class RetentionAgent(Agent):
    goals = [
        Goal("maximize_ltv", weight=0.6),
        Goal("minimize_churn", weight=0.4)
    ]

    constraints = [
        Constraint("margin", min=0.15),
        Constraint("contact_frequency", max=3, period="week")
    ]

    def observe(self, customer):
        return {
            "purchase_history": self.get_purchases(customer),
            "engagement": self.get_engagement(customer),
            "segment": customer.segment
        }

    def act(self, observation, reasoning):
        # Agent decides action based on goals and constraints
        pass

Reasoning Engine

Agents use a reasoning engine that:

• Evaluates possible actions against goals
• Respects hard constraints
• Balances competing objectives
• Explains decisions

Action Space

Pre-built actions for common commerce operations:

• Send communications (email, SMS, push)
• Apply discounts
• Modify recommendations
• Trigger workflows
• Update customer data

Learning

Agents improve through:

• Outcome tracking (did the action achieve the goal?)
• A/B testing of strategies
• Reinforcement learning from feedback

Example: Churn Prevention Agent

churn_agent = ChurnPreventionAgent(
    triggers=[
        "no_purchase_30_days",
        "decreased_engagement",
        "support_complaint"
    ],
    actions=[
        DiscountAction(max_percent=20),
        PersonalEmailAction(),
        LoyaltyPointsAction(max_points=500),
        SurveyAction()
    ]
)

# Agent observes, reasons, and acts autonomously
churn_agent.run(schedule="daily")

Guardrails

Autonomous systems need oversight:

• Approval workflows: High-impact actions require human approval
• Budget limits: Maximum spend per agent per period
• Audit logs: Every decision recorded with reasoning
• Kill switches: Instantly disable any agent

Results

Beta customers using Agent Framework:

• 34% reduction in churn
• 23% improvement in reactivation
• 2.1x increase in repeat purchases
• 15 hours/week saved on manual interventions

Open Source Core

The Agent Framework core is open source:

pip install hanzo-agents

Enterprise features (learning, advanced reasoning) require Hanzo Cloud.

What's Next

• Multi-agent coordination
• Cross-channel orchestration
• Predictive agent activation
• Natural language agent configuration

Agents are the future of commerce automation. Start building today.

---

Zach Kelling is the founder of Hanzo Industries.