Insurtech Engineering

Scaling Agentic workflows, From Hierarchy to Sequential Chains

Scaling Agentic Workflows: From Hierarchy to Sequential Chains

As technology accelerates, it feels like if you aren't building agents, you're probably considered outdated, stuck in the old age of microservices and event-driven systems.

Regardless, we've been working on scaling the use of autonomous agents in production-grade workflows, where performance and reliability are critical. One key learning has been how task orchestration design significantly impacts the stability, predictability, and total execution time of the code.

Our Initial Approach: Hierarchical Task Orchestration

We initially started with Hierarchical Task Orchestration. Our tasks were structured hierarchically, where a lead agent planned and delegated sub-tasks to other agents.

The way execution typically went:

  • A Planner Agent received the main goal.
  • It then decomposed that goal into subtasks.
  • Finally, it delegated these subtasks to different agents for execution.

This approach works great in simple scenarios. However, as logic grew more complex (which usually happens as you add functionality), we started realizing several challenges:

  • We saw delays in task decomposition and planning.
  • The system became more complex to debug.
  • There were situations when tasks were being fully executed and then left off by executors.
  • Real-time responsiveness dropped, especially for multi-step workflows.

Our New Direction: Sequential Task Model

Then we changed to a Sequential Task Model. Now:

  • Agents follow a linear chain of responsibilities.
  • Each agent completes a task and passes the output forward.
  • We avoid central planning and rely on agent chaining with smart memory.

Overall, things have become much faster and more debuggable.

We use CrewAI as our multi-agent platform, and they make it very easy for you to switch your tasks from hierarchical to sequential.

What are your thoughts on these two orchestration models? Have you experienced similar challenges or benefits in your agentic workflows?

Continue Reading

From Burden to Brilliance: The Quiet Revolution of Mulberri’s Berri Agents in Insurance

In the world of insurance, time isn’t just money - it’s protection, it’s trust, and sometimes, it’s everything. Every submission, every policy, every claim carries with it a human story ...
May 6, 2025

How AI is Changing the Insurance Industry

Artificial intelligence has long been the stuff of buzzwords, but when it comes to revolutionizing the insurance industr ...
May 23, 2023

Insurtech Engineering