"""
Pattern 105: Hierarchical Planning
Description:
The Hierarchical Planning pattern breaks down complex goals into a hierarchy
of sub-goals and actions at multiple levels of abstraction. This pattern
enables agents to handle long-horizon tasks by decomposing them into manageable
pieces, planning at different levels of granularity, and progressively refining
high-level strategies into concrete actions.
Hierarchical planning is inspired by hierarchical task networks (HTN) and
goal decomposition techniques. It allows agents to reason about problems at
multiple scales: strategic (what to achieve), tactical (how to achieve it),
and operational (specific actions). Each level focuses on appropriate
abstractions, making planning more tractable and interpretable.
This pattern is particularly effective for complex, multi-step tasks where
flat planning would result in intractably large search spaces. By organizing
goals hierarchically, the agent can plan efficiently, adapt at different
levels, and explain its reasoning at appropriate levels of detail.
Key Components:
1. Goal Hierarchy: Tree structure of goals and sub-goals
2. Abstraction Levels: Strategic, tactical, operational layers
3. Decomposition Rules: How to break goals into sub-goals
4. Level Planner: Plans at each abstraction level
5. Refinement Engine: Converts abstract plans to concrete actions
6. Execution Monitor: Tracks progress at all levels
7. Re-planning Trigger: Identifies when to replan at each level
Abstraction Levels:
1. Strategic Level: High-level goals and objectives
- What needs to be achieved
- Overall strategy and approach
- Long-term planning horizon
2. Tactical Level: Sub-goals and procedures
- How to achieve strategic goals
- Intermediate milestones
- Medium-term planning
3. Operational Level: Concrete actions
- Specific actions to execute
- Immediate next steps
- Short-term execution
Planning Phases:
1. Goal Decomposition: Break high-level goal into sub-goals
2. Level Planning: Plan at each abstraction level
3. Plan Refinement: Refine abstract plans into detailed plans
4. Execution: Execute lowest-level actions
5. Monitoring: Track progress at all levels
6. Re-planning: Adapt when needed at appropriate level
Use Cases:
- Complex project planning
- Multi-step task execution
- Robotic mission planning
- Software development workflows
- Travel itinerary planning
- Research project organization
- Business process automation
Advantages:
- Manages complexity through abstraction
- More efficient than flat planning
- Allows reasoning at appropriate granularity
- Easier to explain and understand
- Facilitates partial replanning
- Modular and reusable sub-plans
- Scales to long-horizon tasks
Challenges:
- Defining appropriate abstraction levels
- Goal decomposition complexity
- Ensuring consistency across levels
- Handling dependencies between sub-goals
- Determining when to replan at each level
- Balancing planning depth vs. efficiency
LangChain Implementation:
This implementation uses LangChain for:
- LLM-based goal decomposition
- Multi-level plan generation
- Plan refinement and detailing
- Natural language planning
Production Considerations:
- Define clear abstraction levels for domain
- Implement efficient goal decomposition
- Track dependencies between sub-goals
- Monitor execution at all levels
- Enable replanning at appropriate level
- Cache and reuse common sub-plans
- Provide visualization of plan hierarchy
- Handle failures at correct abstraction level
"""
import os
from typing import List, Dict, Any, Optional, Tuple
from dataclasses import dataclass, field
from datetime import datetime
from enum import Enum
from dotenv import load_dotenv
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.output_parsers import StrOutputParser
from langchain_openai import ChatOpenAI
load_dotenv()
class PlanLevel(Enum):
"""Planning abstraction levels."""
STRATEGIC = "strategic" # High-level goals
TACTICAL = "tactical" # Sub-goals and procedures
OPERATIONAL = "operational" # Concrete actions
class GoalStatus(Enum):
"""Status of a goal."""
PENDING = "pending"
IN_PROGRESS = "in_progress"
COMPLETED = "completed"
FAILED = "failed"
BLOCKED = "blocked"
@dataclass
class Goal:
"""
Represents a goal at any level of the hierarchy.
Attributes:
goal_id: Unique identifier
description: Goal description
level: Abstraction level
parent_id: Parent goal ID (None for root)
children_ids: List of child goal IDs
status: Current status
priority: Priority (1-10)
dependencies: IDs of goals that must complete first
actions: Concrete actions (for operational level)
metadata: Additional information
"""
goal_id: str
description: str
level: PlanLevel
parent_id: Optional[str] = None
children_ids: List[str] = field(default_factory=list)
status: GoalStatus = GoalStatus.PENDING
priority: int = 5
dependencies: List[str] = field(default_factory=list)
actions: List[str] = field(default_factory=list)
metadata: Dict[str, Any] = field(default_factory=dict)
created_at: datetime = field(default_factory=datetime.now)
completed_at: Optional[datetime] = None
class HierarchicalPlanner:
"""
Hierarchical planning system that plans at multiple abstraction levels.
This planner decomposes high-level goals into hierarchies of sub-goals,
plans at each level, and progressively refines abstract plans into
concrete actions.
"""
def __init__(self, temperature: float = 0.3):
"""
Initialize hierarchical planner.
Args:
temperature: LLM temperature for planning
"""
self.llm = ChatOpenAI(temperature=temperature, model="gpt-3.5-turbo")
self.goals: Dict[str, Goal] = {}
self.goal_counter = 0
self.execution_log: List[Dict[str, Any]] = []
def create_goal(
self,
description: str,
level: PlanLevel,
parent_id: Optional[str] = None,
priority: int = 5,
dependencies: Optional[List[str]] = None
) -> Goal:
"""
Create a new goal.
Args:
description: Goal description
level: Abstraction level
parent_id: Parent goal ID
priority: Priority (1-10)
dependencies: Goal dependencies
Returns:
Created goal
"""
self.goal_counter += 1
goal = Goal(
goal_id=f"goal_{self.goal_counter}",
description=description,
level=level,
parent_id=parent_id,
priority=priority,
dependencies=dependencies or []
)
self.goals[goal.goal_id] = goal
# Update parent's children
if parent_id and parent_id in self.goals:
self.goals[parent_id].children_ids.append(goal.goal_id)
return goal
def decompose_goal(
self,
goal: Goal,
target_level: PlanLevel
) -> List[Goal]:
"""
Decompose a goal into sub-goals at lower abstraction level.
Args:
goal: Goal to decompose
target_level: Target abstraction level for sub-goals
Returns:
List of sub-goals
"""
# Use LLM to decompose goal
prompt = ChatPromptTemplate.from_template(
"You are a hierarchical planning assistant.\n\n"
"Goal: {goal_description}\n"
"Current Level: {current_level}\n"
"Target Level: {target_level}\n\n"
"Decompose this goal into 3-5 sub-goals at the {target_level} level.\n"
"List each sub-goal on a new line, numbered.\n"
"Sub-goals should be concrete, achievable, and collectively accomplish the parent goal.\n\n"
"Sub-goals:"
)
chain = prompt | self.llm | StrOutputParser()
result = chain.invoke({
"goal_description": goal.description,
"current_level": goal.level.value,
"target_level": target_level.value
})
# Parse sub-goals from LLM output
lines = [line.strip() for line in result.strip().split('\n') if line.strip()]
sub_goals = []
for line in lines:
# Remove numbering (e.g., "1. ", "1) ", etc.)
description = line
for prefix in ["1.", "2.", "3.", "4.", "5.", "1)", "2)", "3)", "4)", "5)", "-", "*"]:
if description.startswith(prefix):
description = description[len(prefix):].strip()
break
if description:
sub_goal = self.create_goal(
description=description,
level=target_level,
parent_id=goal.goal_id,
priority=goal.priority
)
sub_goals.append(sub_goal)
return sub_goals
def plan_strategic(self, objective: str) -> Goal:
"""
Create strategic-level plan.
Args:
objective: High-level objective
Returns:
Strategic goal
"""
# Create strategic goal
strategic_goal = self.create_goal(
description=objective,
level=PlanLevel.STRATEGIC,
priority=10
)
return strategic_goal
def plan_tactical(self, strategic_goal: Goal) -> List[Goal]:
"""
Create tactical-level plans from strategic goal.
Args:
strategic_goal: Strategic goal to decompose
Returns:
List of tactical goals
"""
tactical_goals = self.decompose_goal(strategic_goal, PlanLevel.TACTICAL)
return tactical_goals
def plan_operational(self, tactical_goal: Goal) -> List[Goal]:
"""
Create operational-level plans from tactical goal.
Args:
tactical_goal: Tactical goal to decompose
Returns:
List of operational goals
"""
operational_goals = self.decompose_goal(tactical_goal, PlanLevel.OPERATIONAL)
# Generate concrete actions for operational goals
for op_goal in operational_goals:
op_goal.actions = self.generate_actions(op_goal)
return operational_goals
def generate_actions(self, operational_goal: Goal) -> List[str]:
"""
Generate concrete actions for an operational goal.
Args:
operational_goal: Operational goal
Returns:
List of concrete actions
"""
prompt = ChatPromptTemplate.from_template(
"Generate 2-4 specific, concrete actions to accomplish this goal:\n\n"
"Goal: {goal}\n\n"
"List each action on a new line, numbered.\n"
"Actions should be specific and executable.\n\n"
"Actions:"
)
chain = prompt | self.llm | StrOutputParser()
result = chain.invoke({"goal": operational_goal.description})
# Parse actions
lines = [line.strip() for line in result.strip().split('\n') if line.strip()]
actions = []
for line in lines:
# Remove numbering
action = line
for prefix in ["1.", "2.", "3.", "4.", "1)", "2)", "3)", "4)", "-", "*"]:
if action.startswith(prefix):
action = action[len(prefix):].strip()
break
if action:
actions.append(action)
return actions
def create_full_hierarchy(self, objective: str) -> Goal:
"""
Create complete hierarchical plan from objective.
Args:
objective: High-level objective
Returns:
Root strategic goal with full hierarchy
"""
# Strategic level
strategic_goal = self.plan_strategic(objective)
# Tactical level
tactical_goals = self.plan_tactical(strategic_goal)
# Operational level
for tactical_goal in tactical_goals:
self.plan_operational(tactical_goal)
return strategic_goal
def get_next_action(self) -> Optional[Tuple[Goal, str]]:
"""
Get next executable action from the plan.
Returns:
Tuple of (goal, action) or None if no actions available
"""
# Find operational goals that are ready to execute
operational_goals = [
g for g in self.goals.values()
if g.level == PlanLevel.OPERATIONAL and g.status == GoalStatus.PENDING
]
# Sort by priority
operational_goals.sort(key=lambda g: g.priority, reverse=True)
for goal in operational_goals:
# Check dependencies
deps_satisfied = all(
self.goals[dep_id].status == GoalStatus.COMPLETED
for dep_id in goal.dependencies
if dep_id in self.goals
)
if deps_satisfied and goal.actions:
return (goal, goal.actions[0])
return None
def execute_action(self, goal: Goal, action: str) -> bool:
"""
Execute an action (simulated).
Args:
goal: Goal being executed
action: Action to execute
Returns:
Success status
"""
# Mark goal as in progress
if goal.status == GoalStatus.PENDING:
goal.status = GoalStatus.IN_PROGRESS
# Log execution
self.execution_log.append({
"timestamp": datetime.now(),
"goal_id": goal.goal_id,
"action": action,
"level": goal.level.value
})
# Remove completed action
if action in goal.actions:
goal.actions.remove(action)
# Mark goal complete if no more actions
if not goal.actions:
goal.status = GoalStatus.COMPLETED
goal.completed_at = datetime.now()
self.propagate_completion(goal)
return True
def propagate_completion(self, goal: Goal):
"""
Propagate goal completion up the hierarchy.
Args:
goal: Completed goal
"""
if not goal.parent_id or goal.parent_id not in self.goals:
return
parent = self.goals[goal.parent_id]
# Check if all children are complete
all_children_complete = all(
self.goals[child_id].status == GoalStatus.COMPLETED
for child_id in parent.children_ids
if child_id in self.goals
)
if all_children_complete:
parent.status = GoalStatus.COMPLETED
parent.completed_at = datetime.now()
self.propagate_completion(parent)
def get_hierarchy_summary(self, goal_id: Optional[str] = None) -> str:
"""
Get textual summary of goal hierarchy.
Args:
goal_id: Root goal ID (None for all roots)
Returns:
Formatted hierarchy string
"""
def format_goal(goal: Goal, indent: int = 0) -> str:
status_symbol = {
GoalStatus.PENDING: "⏳",
GoalStatus.IN_PROGRESS: "🔄",
GoalStatus.COMPLETED: "✅",
GoalStatus.FAILED: "❌",
GoalStatus.BLOCKED: "🚫"
}
symbol = status_symbol.get(goal.status, "❓")
prefix = " " * indent
result = f"{prefix}{symbol} [{goal.level.value}] {goal.description}\n"
# Add actions for operational level
if goal.level == PlanLevel.OPERATIONAL and goal.actions:
for action in goal.actions:
result += f"{prefix} → {action}\n"
# Recursively add children
for child_id in goal.children_ids:
if child_id in self.goals:
result += format_goal(self.goals[child_id], indent + 1)
return result
if goal_id:
if goal_id in self.goals:
return format_goal(self.goals[goal_id])
return "Goal not found"
# Format all root goals
root_goals = [g for g in self.goals.values() if g.parent_id is None]
return "\n".join(format_goal(g) for g in root_goals)
def get_statistics(self) -> Dict[str, Any]:
"""Get planning statistics."""
goals_by_level = {level: 0 for level in PlanLevel}
goals_by_status = {status: 0 for status in GoalStatus}
for goal in self.goals.values():
goals_by_level[goal.level] += 1
goals_by_status[goal.status] += 1
return {
"total_goals": len(self.goals),
"by_level": {level.value: count for level, count in goals_by_level.items()},
"by_status": {status.value: count for status, count in goals_by_status.items()},
"total_actions_executed": len(self.execution_log)
}
def demonstrate_hierarchical_planning():
"""Demonstrate hierarchical planning pattern."""
print("=" * 80)
print("HIERARCHICAL PLANNING PATTERN DEMONSTRATION")
print("=" * 80)
# Example 1: Basic three-level hierarchy
print("\n" + "=" * 80)
print("Example 1: Three-Level Planning Hierarchy")
print("=" * 80)
planner = HierarchicalPlanner()
print("\nCreating hierarchical plan for: 'Build a web application'")
# Strategic level
strategic = planner.plan_strategic("Build a web application")
print(f"\n[STRATEGIC] {strategic.description}")
# Tactical level
tactical = planner.plan_tactical(strategic)
print(f"\n[TACTICAL] Decomposed into {len(tactical)} sub-goals:")
for i, tac_goal in enumerate(tactical, 1):
print(f" {i}. {tac_goal.description}")
# Operational level (for first tactical goal)
if tactical:
operational = planner.plan_operational(tactical[0])
print(f"\n[OPERATIONAL] '{tactical[0].description}' decomposed into {len(operational)} goals:")
for i, op_goal in enumerate(operational, 1):
print(f" {i}. {op_goal.description}")
if op_goal.actions:
for action in op_goal.actions:
print(f" → {action}")
# Example 2: Complete hierarchy creation
print("\n" + "=" * 80)
print("Example 2: Complete Hierarchical Plan")
print("=" * 80)
planner2 = HierarchicalPlanner()
objective = "Organize a team offsite event"
print(f"\nObjective: {objective}\n")
root = planner2.create_full_hierarchy(objective)
print("Complete Hierarchy:")
print(planner2.get_hierarchy_summary())
# Example 3: Plan execution
print("\n" + "=" * 80)
print("Example 3: Executing Hierarchical Plan")
print("=" * 80)
planner3 = HierarchicalPlanner()
# Create simpler plan for execution demo
strategic = planner3.create_goal(
"Complete project report",
PlanLevel.STRATEGIC
)
tactical1 = planner3.create_goal(
"Gather data and research",
PlanLevel.TACTICAL,
parent_id=strategic.goal_id
)
tactical2 = planner3.create_goal(
"Write and format report",
PlanLevel.TACTICAL,
parent_id=strategic.goal_id,
dependencies=[tactical1.goal_id]
)
# Operational goals for tactical1
op1 = planner3.create_goal(
"Collect data from databases",
PlanLevel.OPERATIONAL,
parent_id=tactical1.goal_id
)
op1.actions = ["Connect to database", "Run queries", "Export results"]
op2 = planner3.create_goal(
"Review literature",
PlanLevel.OPERATIONAL,
parent_id=tactical1.goal_id
)
op2.actions = ["Search papers", "Read abstracts"]
print("Initial Plan:")
print(planner3.get_hierarchy_summary())
print("\nExecuting actions...")
# Execute actions
for i in range(5):
next_item = planner3.get_next_action()
if next_item:
goal, action = next_item
print(f" {i+1}. Executing: {action} (for '{goal.description}')")
planner3.execute_action(goal, action)
else:
break
print("\nUpdated Plan:")
print(planner3.get_hierarchy_summary())
# Example 4: Priority-based planning
print("\n" + "=" * 80)
print("Example 4: Priority-Based Goal Selection")
print("=" * 80)
planner4 = HierarchicalPlanner()
# Create goals with different priorities
strategic = planner4.create_goal(
"Launch product",
PlanLevel.STRATEGIC,
priority=10
)
high_priority = planner4.create_goal(
"Fix critical bug",
PlanLevel.OPERATIONAL,
parent_id=strategic.goal_id,
priority=9
)
high_priority.actions = ["Identify bug", "Write fix", "Test fix"]
low_priority = planner4.create_goal(
"Update documentation",
PlanLevel.OPERATIONAL,
parent_id=strategic.goal_id,
priority=3
)
low_priority.actions = ["Review docs", "Make updates"]
print("Goals by priority:")
for goal in sorted(planner4.goals.values(), key=lambda g: g.priority, reverse=True):
if goal.level == PlanLevel.OPERATIONAL:
print(f" Priority {goal.priority}: {goal.description}")
print("\nExecution order (priority-based):")
for i in range(5):
next_item = planner4.get_next_action()
if next_item:
goal, action = next_item
print(f" {i+1}. [P{goal.priority}] {action}")
planner4.execute_action(goal, action)
else:
break
# Example 5: Dependency handling
print("\n" + "=" * 80)
print("Example 5: Goal Dependencies")
print("=" * 80)
planner5 = HierarchicalPlanner()
strategic = planner5.create_goal("Deploy application", PlanLevel.STRATEGIC)
goal_a = planner5.create_goal(
"Run tests",
PlanLevel.OPERATIONAL,
parent_id=strategic.goal_id
)
goal_a.actions = ["Execute test suite"]
goal_b = planner5.create_goal(
"Build for production",
PlanLevel.OPERATIONAL,
parent_id=strategic.goal_id,
dependencies=[goal_a.goal_id]
)
goal_b.actions = ["Compile code"]
goal_c = planner5.create_goal(
"Deploy to server",
PlanLevel.OPERATIONAL,
parent_id=strategic.goal_id,
dependencies=[goal_b.goal_id]
)
goal_c.actions = ["Upload to server"]
print("Dependency chain: Tests → Build → Deploy\n")
print("Initial state:")
for goal in [goal_a, goal_b, goal_c]:
deps = ", ".join(goal.dependencies) if goal.dependencies else "none"
print(f" {goal.description}: dependencies=[{deps}], status={goal.status.value}")
print("\nExecution respects dependencies:")
for i in range(3):
next_item = planner5.get_next_action()
if next_item:
goal, action = next_item
print(f" {i+1}. {action} (from '{goal.description}')")
planner5.execute_action(goal, action)
# Example 6: Progress monitoring
print("\n" + "=" * 80)
print("Example 6: Multi-Level Progress Monitoring")
print("=" * 80)
planner6 = HierarchicalPlanner()
root = planner6.create_full_hierarchy("Write research paper")
print("Initial Statistics:")
stats = planner6.get_statistics()
print(f" Total Goals: {stats['total_goals']}")
print(f" By Level: {stats['by_level']}")
print(f" By Status: {stats['by_status']}")
# Execute some actions
print("\nExecuting 10 actions...")
for _ in range(10):
next_item = planner6.get_next_action()
if next_item:
goal, action = next_item
planner6.execute_action(goal, action)
print("\nUpdated Statistics:")
stats = planner6.get_statistics()
print(f" Total Goals: {stats['total_goals']}")
print(f" By Status: {stats['by_status']}")
print(f" Actions Executed: {stats['total_actions_executed']}")
# Example 7: Plan visualization
print("\n" + "=" * 80)
print("Example 7: Hierarchical Plan Visualization")
print("=" * 80)
planner7 = HierarchicalPlanner()
# Create multi-level plan
strategic = planner7.create_goal("Learn Python", PlanLevel.STRATEGIC)
tac1 = planner7.create_goal("Master basics", PlanLevel.TACTICAL, parent_id=strategic.goal_id)
tac2 = planner7.create_goal("Build projects", PlanLevel.TACTICAL, parent_id=strategic.goal_id)
op1 = planner7.create_goal("Learn syntax", PlanLevel.OPERATIONAL, parent_id=tac1.goal_id)
op1.actions = ["Study variables", "Practice loops"]
op1.status = GoalStatus.COMPLETED # Mark as done
op2 = planner7.create_goal("Learn data structures", PlanLevel.OPERATIONAL, parent_id=tac1.goal_id)
op2.actions = ["Study lists", "Study dictionaries"]
op2.status = GoalStatus.IN_PROGRESS
op3 = planner7.create_goal("Build web app", PlanLevel.OPERATIONAL, parent_id=tac2.goal_id)
op3.actions = ["Choose framework", "Create app"]
print("Hierarchical Plan with Status:")
print(planner7.get_hierarchy_summary())
# Example 8: Replanning scenario
print("\n" + "=" * 80)
print("Example 8: Adaptive Replanning")
print("=" * 80)
planner8 = HierarchicalPlanner()
print("Original Plan:")
strategic = planner8.create_goal("Travel to conference", PlanLevel.STRATEGIC)
tac1 = planner8.create_goal("Book transportation", PlanLevel.TACTICAL, parent_id=strategic.goal_id)
op1 = planner8.create_goal("Book flight", PlanLevel.OPERATIONAL, parent_id=tac1.goal_id)
op1.actions = ["Search flights", "Purchase ticket"]
print(planner8.get_hierarchy_summary())
print("\nSimulating failure (flight cancelled)...")
op1.status = GoalStatus.FAILED
# Create alternative plan
print("\nReplanning at tactical level:")
op2 = planner8.create_goal("Book train", PlanLevel.OPERATIONAL, parent_id=tac1.goal_id)
op2.actions = ["Search trains", "Purchase ticket"]
print(planner8.get_hierarchy_summary())
# Summary
print("\n" + "=" * 80)
print("SUMMARY: Hierarchical Planning Pattern")
print("=" * 80)
summary = """
The Hierarchical Planning pattern demonstrated:
1. THREE-LEVEL HIERARCHY (Example 1):
- Strategic: High-level objectives
- Tactical: Sub-goals and procedures
- Operational: Concrete actions
- Progressive decomposition
- Abstraction management
2. COMPLETE HIERARCHY (Example 2):
- Full plan generation
- Multi-level decomposition
- Parent-child relationships
- Structured goal tree
- Hierarchical visualization
3. PLAN EXECUTION (Example 3):
- Sequential action execution
- Status tracking at all levels
- Progress propagation
- Completion detection
- Hierarchical updates
4. PRIORITY HANDLING (Example 4):
- Priority-based selection
- Urgent task handling
- Importance ordering
- Resource allocation
- Efficient scheduling
5. DEPENDENCY MANAGEMENT (Example 5):
- Goal dependencies
- Execution ordering
- Prerequisite checking
- Constraint satisfaction
- Sequential execution
6. PROGRESS MONITORING (Example 6):
- Multi-level tracking
- Status aggregation
- Statistics collection
- Performance metrics
- Completion monitoring
7. PLAN VISUALIZATION (Example 7):
- Hierarchical display
- Status indicators
- Tree structure
- Action details
- Clear representation
8. ADAPTIVE REPLANNING (Example 8):
- Failure handling
- Alternative generation
- Level-appropriate replanning
- Plan modification
- Resilience
KEY BENEFITS:
✓ Manages complexity through abstraction
✓ Efficient planning for long horizons
✓ Reasoning at appropriate granularity
✓ Easy to understand and explain
✓ Facilitates partial replanning
✓ Modular and reusable
✓ Scales to complex tasks
✓ Clear progress tracking
USE CASES:
• Complex project planning
• Multi-step task execution
• Robotic mission planning
• Software development workflows
• Travel itinerary planning
• Research project organization
• Business process automation
• Strategic decision-making
ABSTRACTION LEVELS:
→ Strategic: What to achieve (objectives)
→ Tactical: How to achieve (procedures)
→ Operational: Specific actions (execution)
BEST PRACTICES:
1. Define clear abstraction levels for domain
2. Decompose goals systematically
3. Track dependencies between goals
4. Monitor execution at all levels
5. Enable replanning at appropriate level
6. Cache and reuse common sub-plans
7. Visualize plan hierarchy
8. Handle failures at correct level
TRADE-OFFS:
• Planning overhead vs. execution efficiency
• Abstraction depth vs. simplicity
• Flexibility vs. structure
• Decomposition detail vs. overview
PRODUCTION CONSIDERATIONS:
→ Implement domain-specific decomposition rules
→ Use caching for common sub-plans
→ Monitor performance at each level
→ Enable interactive plan adjustment
→ Provide clear visualization
→ Track dependencies explicitly
→ Implement efficient replanning
→ Handle partial plan execution
→ Support concurrent goal execution
→ Log decisions at all levels
This pattern enables agents to tackle complex, long-horizon tasks
by organizing planning at multiple levels of abstraction, from
high-level strategies to concrete actions.
"""
print(summary)
if __name__ == "__main__":
demonstrate_hierarchical_planning()
Agentic Design Pattern: Hierarchical Planning
Pattern 105: Hierarchical Planning – Line-by-Line Explanation
Overview
This document provides a detailed line-by-line explanation of the Hierarchical Planning pattern implementation.
Pattern Description
Pattern 105: Hierarchical Planning
Description: The Hierarchical Planning pattern breaks down complex goals into a hierarchy of sub-goals and actions at multiple levels of abstraction. This pattern enables agents to handle long-horizon tasks by decomposing them into manageable pieces, planning at different levels of granularity, and progressively refining high-level strategies into concrete actions.
Hierarchical planning is inspired by hierarchical task networks (HTN) and
goal decomposition techniques. It allows agents to reason about problems at
multiple scales: strategic (what to achieve), tactical (how to achieve it),
and operational (specific actions). Each level focuses on appropriate
abstractions, making planning more tractable and interpretable.
This pattern is particularly effective for complex, multi-step tasks where
flat planning would result in intractably large search spaces. By organizing
goals hierarchically, the agent can plan efficiently, adapt at different
levels, and explain its reasoning at appropriate levels of detail.
Key Components: 1. Goal Hierarchy: Tree structure of goals and sub-goals 2. Abstraction Levels: Strategic, tactical, operational layers 3. Decomposition Rules: How to break goals into sub-goals 4. Level Planner: Plans at each abstraction level 5. Refinement Engine: Converts abstract plans to concrete actions 6. Execution Monitor: Tracks progress at all levels 7. Re-planning Trigger: Identifies when to replan at each level
Abstraction Levels: 1. Strategic Level: High-level goals and objectives – What needs to be achieved – Overall strategy and approach – Long-term planning horizon
2. Tactical Level: Sub-goals and procedures
- How to achieve strategic goals
- Intermediate milestones
- Medium-term planning
3. Operational Level: Concrete actions
- Specific actions to execute
- Immediate next steps
- Short-term execution
Planning Phases: 1. Goal Decomposition: Break high-level goal into sub-goals 2. Level Planning: Plan at each abstraction level 3. Plan Refinement: Refine abstract plans into detailed plans 4. Execution: Execute lowest-level actions 5. Monitoring: Track progress at all levels 6. Re-planning: Adapt when needed at appropriate level
Use Cases: – Complex project planning – Multi-step task execution – Robotic mission planning – Software development workflows – Travel itinerary planning – Research project organization – Business process automation
Advantages: – Manages complexity through abstraction – More efficient than flat planning – Allows reasoning at appropriate granularity – Easier to explain and understand – Facilitates partial replanning – Modular and reusable sub-plans – Scales to long-horizon tasks
Challenges: – Defining appropriate abstraction levels – Goal decomposition complexity – Ensuring consistency across levels – Handling dependencies between sub-goals – Determining when to replan at each level – Balancing planning depth vs. efficiency
LangChain Implementation: This implementation uses LangChain for: – LLM-based goal decomposition – Multi-level plan generation – Plan refinement and detailing – Natural language planning
Production Considerations: – Define clear abstraction levels for domain – Implement efficient goal decomposition – Track dependencies between sub-goals – Monitor execution at all levels – Enable replanning at appropriate level – Cache and reuse common sub-plans – Provide visualization of plan hierarchy – Handle failures at correct abstraction level
File Statistics
- Total Lines: 946
- Classes: 4
- Functions: 1
- Constants: 0
Module Documentation
"""
Pattern 105: Hierarchical Planning
Description:
The Hierarchical Planning pattern breaks down complex goals into a hierarchy
of sub-goals and actions at multiple levels of abstraction. This pattern
enables agents to handle long-horizon tasks by decomposing them into manageable
pieces, planning at different levels of granularity, and progressively refining
high-level strategies into concrete actions.
Hierarchical planning is inspired by hierarchical task networks (HTN) and
goal decomposition techniques. It allows agents to reason about problems at
multiple scales: strategic (what to achieve), tactical (how to achieve it),
and operational (specific actions). Each level focuses on appropriate
abstractions, making planning more tractable and interpretable.
This pattern is particularly effective for complex, multi-step tasks where
flat planning would result in intractably large search spaces. By organizing
goals hierarchically, the agent can plan efficiently, adapt at different
levels, and explain its reasoning at appropriate levels of detail.
Key Components:
1. Goal Hierarchy: Tree structure of goals and sub-goals
2. Abstraction Levels: Strategic, tactical, operational layers
3. Decomposition Rules: How to break goals into sub-goals
4. Level Planner: Plans at each abstraction level
5. Refinement Engine: Converts abstract plans to concrete actions
6. Execution Monitor: Tracks progress at all levels
7. Re-planning Trigger: Identifies when to replan at each level
Abstraction Levels:
1. Strategic Level: High-level goals and objectives
- What needs to be achieved
- Overall strategy and approach
- Long-term planning horizon
2. Tactical Level: Sub-goals and procedures
- How to achieve strategic goals
- Intermediate milestones
- Medium-term planning
3. Operational Level: Concrete actions
- Specific actions to execute
- Immediate next steps
- Short-term execution
Planning Phases:
1. Goal Decomposition: Break high-level goal into sub-goals
2. Level Planning: Plan at each abstraction level
3. Plan Refinement: Refine abstract plans into detailed plans
4. Execution: Execute lowest-level actions
5. Monitoring: Track progress at all levels
6. Re-planning: Adapt when needed at appropriate level
Use Cases:
- Complex project planning
- Multi-step task execution
- Robotic mission planning
- Software development workflows
- Travel itinerary planning
- Research project organization
- Business process automation
Advantages:
- Manages complexity through abstraction
- More efficient than flat planning
- Allows reasoning at appropriate granularity
- Easier to explain and understand
- Facilitates partial replanning
- Modular and reusable sub-plans
- Scales to long-horizon tasks
Challenges:
- Defining appropriate abstraction levels
- Goal decomposition complexity
- Ensuring consistency across levels
- Handling dependencies between sub-goals
- Determining when to replan at each level
- Balancing planning depth vs. efficiency
LangChain Implementation:
This implementation uses LangChain for:
- LLM-based goal decomposition
- Multi-level plan generation
- Plan refinement and detailing
- Natural language planning
Production Considerations:
- Define clear abstraction levels for domain
- Implement efficient goal decomposition
- Track dependencies between sub-goals
- Monitor execution at all levels
- Enable replanning at appropriate level
- Cache and reuse common sub-plans
- Provide visualization of plan hierarchy
- Handle failures at correct abstraction level
"""
Purpose: This module implements the Hierarchical Planning pattern, providing a structured approach to agent design and implementation.
Import Statements
Standard Library Imports
import dataclasses.dataclass
import dataclasses.field
import datetime.datetime
import enum.Enum
import langchain_core.output_parsers.StrOutputParser
import langchain_core.prompts.ChatPromptTemplate
import os
import typing.Any
import typing.Dict
import typing.List
import typing.Optional
import typing.Tuple
Purpose: Core Python functionality:
– dataclasses.dataclass: Simplified class definitions with @dataclass
– dataclasses.field: Simplified class definitions with @dataclass
– datetime.datetime: Date and time handling
– enum.Enum: Enumeration support
– langchain_core.output_parsers.StrOutputParser: Regular expression operations
– langchain_core.prompts.ChatPromptTemplate: Regular expression operations
– os: Environment variable access and system operations
– typing.Any: Type hints for better code documentation (List, Dict, Any, Optional)
– typing.Dict: Type hints for better code documentation (List, Dict, Any, Optional)
– typing.List: Type hints for better code documentation (List, Dict, Any, Optional)
– typing.Optional: Type hints for better code documentation (List, Dict, Any, Optional)
– typing.Tuple: Type hints for better code documentation (List, Dict, Any, Optional)
LangChain Imports
from langchain_openai.ChatOpenAI
Purpose: LangChain framework components: – LLM Interface: Wrapper for OpenAI models – Chains: Composable building blocks for complex workflows
Other Imports
from dotenv.load_dotenv
Purpose: Additional dependencies for specialized functionality.
Classes
Class: PlanLevel
Lines 113-117
Inherits from: Enum
Purpose:
Planning abstraction levels.
Class: GoalStatus
Lines 120-126
Inherits from: Enum
Purpose:
Status of a goal.
Class: Goal
Lines 130-157
Purpose:
Represents a goal at any level of the hierarchy.
Attributes:
goal_id: Unique identifier
description: Goal description
level: Abstraction level
parent_id: Parent goal ID (None for root)
children_ids: List of child goal IDs
status: Current status
priority: Priority (1-10)
dependencies: IDs of goals that must complete first
actions: Concrete actions (for operational level)
metadata: Additional information
Class: HierarchicalPlanner
Lines 160-533
Purpose:
Hierarchical planning system that plans at multiple abstraction levels.
This planner decomposes high-level goals into hierarchies of sub-goals,
plans at each level, and progressively refines abstract plans into
concrete actions.
Methods:
__init__(self, temperature)
Lines 169-179
Purpose:
Initialize hierarchical planner.
Args:
temperature: LLM temperature for planning
Parameters:
– temperature: Sampling temperature (0-1)
create_goal(self, description, level, parent_id, priority, dependencies)
Lines 181-218
Purpose:
Create a new goal.
Args:
description: Goal description
level: Abstraction level
parent_id: Parent goal ID
priority: Priority (1-10)
dependencies: Goal dependencies
Returns:
Created goal
Parameters:
– description: Description
– level: Level
– parent_id: Parent id
– priority: Priority
– dependencies: Dependencies
Returns: Goal
decompose_goal(self, goal, target_level)
Lines 220-275
Purpose:
Decompose a goal into sub-goals at lower abstraction level.
Args:
goal: Goal to decompose
target_level: Target abstraction level for sub-goals
Returns:
List of sub-goals
Parameters:
– goal: Goal
– target_level: Target level
Returns: <ast.Subscript object at 0x000001A06D6F6A50>
plan_strategic(self, objective)
Lines 277-294
Purpose:
Create strategic-level plan.
Args:
objective: High-level objective
Returns:
Strategic goal
Parameters:
– objective: Objective
Returns: Goal
plan_tactical(self, strategic_goal)
Lines 296-307
Purpose:
Create tactical-level plans from strategic goal.
Args:
strategic_goal: Strategic goal to decompose
Returns:
List of tactical goals
Parameters:
– strategic_goal: Strategic goal
Returns: <ast.Subscript object at 0x000001A06D605650>
plan_operational(self, tactical_goal)
Lines 309-325
Purpose:
Create operational-level plans from tactical goal.
Args:
tactical_goal: Tactical goal to decompose
Returns:
List of operational goals
Parameters:
– tactical_goal: Tactical goal
Returns: <ast.Subscript object at 0x000001A06D718510>
generate_actions(self, operational_goal)
Lines 327-362
Purpose:
Generate concrete actions for an operational goal.
Args:
operational_goal: Operational goal
Returns:
List of concrete actions
Parameters:
– operational_goal: Operational goal
Returns: <ast.Subscript object at 0x000001A06D866310>
create_full_hierarchy(self, objective)
Lines 364-384
Purpose:
Create complete hierarchical plan from objective.
Args:
objective: High-level objective
Returns:
Root strategic goal with full hierarchy
Parameters:
– objective: Objective
Returns: Goal
get_next_action(self)
Lines 386-413
Purpose:
Get next executable action from the plan.
Returns:
Tuple of (goal, action) or None if no actions available
Parameters:
Returns: <ast.Subscript object at 0x000001A06D72C9D0>
execute_action(self, goal, action)
Lines 415-448
Purpose:
Execute an action (simulated).
Args:
goal: Goal being executed
action: Action to execute
Returns:
Success status
Parameters:
– goal: Goal
– action: Action
Returns: bool
propagate_completion(self, goal)
Lines 450-472
Purpose:
Propagate goal completion up the hierarchy.
Args:
goal: Completed goal
Parameters:
– goal: Goal
get_hierarchy_summary(self, goal_id)
Lines 474-517
Purpose:
Get textual summary of goal hierarchy.
Args:
goal_id: Root goal ID (None for all roots)
Returns:
Formatted hierarchy string
Parameters:
– goal_id: Goal id
Returns: str
get_statistics(self)
Lines 519-533
Purpose:
Get planning statistics.
Parameters:
Returns: <ast.Subscript object at 0x000001A06D853BD0>
Functions
demonstrate_hierarchical_planning()
Lines 536-941
Purpose:
Demonstrate hierarchical planning pattern.
Key Concepts Summary
Pattern Overview
The Hierarchical Planning pattern provides a structured approach to building AI agents with specific capabilities and behaviors. This implementation leverages LangChain’s composable components to create a robust and maintainable solution.
Core Components
- Agent Class: Main implementation of the pattern logic
- Prompt Engineering: Carefully crafted prompts to guide LLM behavior
- Tool Integration: Optional tools for extended functionality
- Memory Management: State handling across interactions
- Output Parsing: Structured response processing
When to Use
✅ Good for: – Scenarios requiring this specific pattern’s strengths – Applications needing structured agent behavior – Systems where explainability is important – Production environments requiring reliability
❌ Not ideal for: – Simple single-prompt tasks – Scenarios requiring different pattern characteristics – Resource-constrained environments (if computationally intensive)
Best Practices Demonstrated
1. Environment Configuration
- ✅ Use
.envfiles for API keys - ✅ Never hardcode credentials
- ✅ Validate environment variables on startup
2. Error Handling
- ✅ Graceful degradation on failures
- ✅ Informative error messages
- ✅ Logging for debugging
3. Prompt Engineering
- ✅ Clear, specific instructions
- ✅ Structured output formats
- ✅ Few-shot examples when beneficial
4. Code Organization
- ✅ Separation of concerns (class-based design)
- ✅ Type hints for clarity
- ✅ Docstrings for documentation
5. Performance Optimization
- ✅ Appropriate temperature settings
- ✅ Token budget management
- ✅ Caching where applicable
LangChain Components Used
- ChatOpenAI: OpenAI model interface
- Chains: Sequential component composition
Why LangChain? – Composable abstractions for LLM applications – Rich ecosystem of integrations – Production-ready patterns – Active community and development
Execution Flow
1. Initialize agent with configuration
↓
2. Load environment and dependencies
↓
3. Prepare prompt with input data
↓
4. Invoke LLM through LangChain
↓
5. Parse and validate response
↓
6. Return formatted result
Key Decision Points: – Model selection (temperature, model type) – Prompt structure (few-shot vs zero-shot) – Error handling strategy – Output format validation
Performance Considerations
Computational Cost
- LLM Calls: Primary cost factor (based on tokens)
- Latency: Depends on model and input size
- Memory: Minimal for most patterns
Optimization Strategies
- Prompt Optimization: Minimize tokens while maintaining clarity
- Caching: Cache repeated queries/responses
- Model Selection: Use smaller models when appropriate
- Batch Processing: Group similar requests
- Streaming: For long responses, stream tokens
Cost Estimates (Approximate)
- GPT-3.5-turbo: ~$0.002 per request
- GPT-4: ~$0.03 per request
- GPT-4-turbo: ~$0.01 per request
Actual costs vary based on input/output token counts
Common Pitfalls to Avoid
1. Prompt Design Issues
❌ Vague or ambiguous instructions
✅ Clear, specific, testable prompts
2. Error Handling
❌ Ignoring API errors or rate limits
✅ Implement retry logic and graceful degradation
3. Token Management
❌ Exceeding model token limits
✅ Truncate inputs, chunk processing
4. Security
❌ Exposing API keys in code
✅ Use environment variables
5. Testing
❌ Only testing happy paths
✅ Test edge cases, failures, malformed inputs
6. Cost Management
❌ Unmonitored API usage
✅ Implement budgets, alerts, and monitoring
Further Improvements
Production Enhancements
- Monitoring: Add logging, metrics, tracing
- Validation: Input/output schema validation
- Rate Limiting: Implement request throttling
- Caching: Redis or in-memory cache for responses
- Async Support: Non-blocking operations for scalability
Feature Extensions
- Multi-Model Support: Switch between providers
- Fallback Models: Graceful degradation strategy
- Custom Tools: Domain-specific function calling
- Advanced Memory: Vector stores for long-term memory
- Evaluation: Automated testing and benchmarking
Code Quality
- Testing: Unit tests, integration tests
- Documentation: API docs, usage examples
- Type Safety: Stricter type hints
- Configuration: Externalized config files
Related Patterns
- Pattern 104
- Pattern 106
Pattern Families: – Reasoning Patterns (001-011) – Multi-Agent Patterns (012-019) – Tool Use Patterns (020-025) – Memory Patterns (026-031) – And many more…
Summary
The Hierarchical Planning pattern provides a powerful approach to building AI agents with specific capabilities. This implementation demonstrates:
✅ Key Strengths: – Clean, maintainable code structure – LangChain integration for composability – Type hints and documentation – Production-ready patterns
🎯 Best Use Cases: – Applications requiring this pattern’s characteristics – Systems needing structured agent behavior – Production environments requiring reliability
📚 Learning Resources: – LangChain Documentation: https://python.langchain.com/ – Pattern Documentation: See related patterns – OpenAI API Reference: https://platform.openai.com/docs/
File: 105_hierarchical_planning.py
Total Lines: 946
Pattern Category: AI Agent Design Pattern #105
Generated from actual implementation code
For educational and reference purposes
Leave a Reply