What's New in CrewAI - Enhanced execution and control

Lennex Zinyando

• 4 min read

Welcome to the second installment of our 3-part blog series, where we continue to unpack the exciting new additions in CrewAI's latest versions. This post aims to surface the features introduced from v0.30.0 to v0.36.0, specifically those that enhance your control over crew execution and boost agent performance. While we'll cover the highlights here, remember that the CrewAI documentation remains your go-to source for comprehensive details on all these features and more.

Advanced Kickoff Methods

Previously there was only one way to start executing a Crew ie crew.kickoff() method. In recent releases, CrewAI added new ways to start crew execution. I have given brief explanations of how they work below.

Efficient List Processing - kickoff_for_each()

CrewAI now offers a powerful feature that allows you to execute a crew's tasks for each item in a list. This functionality is particularly beneficial when you need to apply the same analysis or processing to multiple datasets or items.

The kickoff_for_each() method enables you to run your crew's operations on each item in a given list, streamlining the process of handling multiple datasets or scenarios.

...

# Define multiple datasets for analysis
datasets = [
    {"ages": [25, 30, 35, 40, 45]},
    {"ages": [20, 25, 30, 35, 40]},
    {"ages": [30, 35, 40, 45, 50]}
]

# Execute the crew for each dataset
results = data_crew.kickoff_for_each(inputs=datasets)

This approach significantly reduces the amount of repetitive code needed to handle multiple datasets, making your CrewAI projects more scalable and maintainable. Whether you're analyzing multiple data sets, processing various text inputs, or applying the same AI operations to a list of items, the kickoff_for_each() method provides a streamlined solution for batch processing in CrewAI.

Leveraging Non-Blocking Crew Operations - kickoff_async()

CrewAI now offers the ability to initiate crew operations asynchronously, enabling developers to start crew executions without blocking the main thread. This feature is invaluable for scenarios requiring concurrent crew operations or when other tasks need to run alongside crew execution.

The kickoff_async() method is the key to asynchronous crew execution. By using this method, you can start crew operations in a separate thread, allowing your main program to continue unimpeded.

...

# Execute the crew
result = analysis_crew.kickoff_async(inputs={"ages": [25, 30, 35, 40, 45]})

Asynchronous Batch Processing - kickoff_for_each_async()

CrewAI has introduced a powerful new method, kickoff_for_each_async(), which takes batch processing to the next level by executing tasks for each agent asynchronously. This feature is designed to maximize efficiency when dealing with multiple inputs or datasets that require similar processing.

The kickoff_for_each_async() method is similar to its synchronous counterpart kickoff_for_each() but with the added benefit of asynchronous execution. This means that tasks for different inputs are processed concurrently, potentially leading to significant time savings, especially for I/O-bound operations.

# Define your crew and tasks here (not shown for brevity)
my_crew = Crew(agents=[...], tasks=[...])

# Prepare an array of inputs
inputs_array = [
    {'topic': 'AI in healthcare'},
    {'topic': 'AI in finance'}
]

# Execute tasks asynchronously for each input
async_results = my_crew.kickoff_for_each_async(inputs=inputs_array)

# Process the results
for async_result in async_results:
    print(async_result)

Benefits of Asynchronous Batch Processing

  1. Improved Efficiency: Process multiple inputs concurrently, reducing overall execution time.
  2. Resource Optimization: Make better use of system resources by allowing parallel task execution.
  3. Scalability: Easily handle growing numbers of inputs without linear increases in processing time.

By leveraging kickoff_for_each_async(), you can significantly enhance the performance of your CrewAI projects, especially when dealing with multiple similar tasks or large datasets that can be processed independently.

Crew Training Feature

CrewAI introduces a powerful training feature that allows you to optimize your AI agents' performance through iterative learning and human feedback. This process enhances the agents' understanding, decision-making, and problem-solving abilities, making them more effective in tackling complex tasks.

To train your crew using the CLI:

  1. Open your terminal
  2. Navigate to your CrewAI project directory
  3. Execute the command:
crewai train -n X

Replace X with your desired number of training iterations.

Key Considerations for Effective Training

  1. Iteration Count: Choose a positive integer for n_iterations. This determines the number of training cycles.
  2. Error Handling: The system is designed to catch and report both subprocess errors and unexpected exceptions.
  3. Time Investment: Training duration depends on agent complexity and requires your feedback for each iteration.
  4. Regular Updates: To maintain peak performance, periodically retrain your agents with the latest information and advancements.

The Impact of Training

Post-training, your AI agents will demonstrate:

  • Enhanced capabilities
  • Improved knowledge base
  • Greater consistency in outputs
  • More valuable insights

Maximizing the Training Feature

To get the most out of CrewAI's training feature:

  • Start with a reasonable number of iterations
  • Provide thoughtful feedback during each cycle
  • Gradually increase complexity as agents improve
  • Regularly assess agent performance and adjust training parameters

By leveraging this training feature, you can significantly boost the effectiveness of your AI crew, leading to more robust and reliable AI-driven solutions.

Conclusion

CrewAI's advanced features - from asynchronous execution and efficient list processing to sophisticated training capabilities - offer a powerful toolkit for developers and data scientists. These tools enable the creation of more responsive, efficient, and adaptable AI systems, capable of handling complex tasks across various domains. By leveraging these capabilities, you can significantly enhance your AI workflows, saving time and resources while improving the quality of your AI-driven solutions.

As AI continues to evolve, mastering tools like CrewAI becomes crucial for staying at the forefront of technology. We encourage you to experiment with these features, integrate them into your projects, and experience firsthand how they can transform your AI development process. With CrewAI, you're not just keeping pace with the AI revolution - you're actively shaping its future, one intelligent crew at a time.

AI should drive results, not complexity. AgentemAI helps businesses build scalable, efficient, and secure AI solutions. See how we can help.