Self Reflection AI Agents

​

Overview

When an agent with self-reflection enabled processes a task:

1. It generates an initial response using its tools and capabilities
2. The agent then reflects on its own response, evaluating:
   • Completeness
   • Accuracy
   • Relevance
   • Quality
3. Based on this reflection, the agent may revise and improve its response
4. This cycle can repeat until the agent is satisfied with the quality

​

Usage

Self-reflection is enabled by default for all agents. You can explicitly control it using the self_reflect parameter:

from praisonaiagents import Agent

agent = Agent(
    name="Researcher",
    role="Senior Research Analyst",
    goal="Uncover cutting-edge developments in AI",
    backstory="You are an expert at analyzing tech trends",
    self_reflect=True  # Enable self-reflection (default)
)

​

Benefits

• Higher Quality Outputs: Agents can catch and correct potential issues before delivering responses
• More Reliable Task Completion: Self-reflection helps ensure all aspects of a task are properly addressed
• Conscious Decision Making: Agents evaluate their own work, leading to more thoughtful responses
• Improved Accuracy: Multiple reflection cycles help refine and perfect the output

​

Integration with Processes

Self-reflection works seamlessly with both sequential and hierarchical processes:

• Sequential: Each agent self-reflects on their individual tasks
• Hierarchical: Agents self-reflect before reporting to the manager agent, ensuring high-quality inputs for decision-making

​

Example

Here’s a complete example showing self-reflection in action:

from praisonaiagents import Agent, Task, PraisonAIAgents

researcher = Agent(
    name="Researcher",
    role="Senior Research Analyst",
    goal="Uncover cutting-edge developments in AI",
    backstory="You are an expert at analyzing tech trends",
    self_reflect=True,
    llm="gpt-4o"
)

task = Task(
    name="research_task",
    description="Analyze 2024's AI advancements",
    expected_output="A detailed report on AI trends",
    agent=researcher
)

agents = PraisonAIAgents(
    agents=[researcher],
    tasks=[task],
    process="sequential"
)

result = agents.start()

In this example, the researcher agent will:

1. Research AI advancements
2. Generate an initial report
3. Self-reflect on the report’s quality
4. Make improvements if needed
5. Deliver the final, refined output

​

Performance Impact

While self-reflection adds additional processing time, the improvement in output quality often justifies the extra computation. The number of reflection cycles need to be manually optimized to balance quality and performance.