Dynamic Workflows using Openhands SDK

Rajistics - data science, AI, and machine learning · Intermediate ·🤖 AI Agents & Automation ·1mo ago

Key Takeaways

Demonstrates dynamic workflows using OpenHands SDK and implements orchestration code for multi-agent systems

Original Description

This week Anthropic released dynamic workflows in Claude Code, and Graham on our team added the same capability to the open source OpenHands SDK. Instead of you hand-writing the orchestration loop for a multi-agent system, you give the agent a tool and it writes the orchestration code itself. I walk through how we implemented it, run a deep research demo live, and trace exactly what the model does behind the scenes so you can start using it yourself. Repo: https://github.com/rajshah4/workflow-demos 0:00 From 25 lines of orchestration to two 0:28 The problem: hand-coding multi-agent systems 1:09 The solution: the agent writes its own orchestration 2:52 Running the demo: manual script vs workflow 4:03 Tracing the live run: fan out, cross-check, synthesize 5:33 The final research report 5:56 Observability with Laminar and chat-with-the-trace 7:23 Why this matters: models are eating orchestration ━━━━━━━━━━━━━━━━━━━━━━━━━ ★ Rajistics Social Media » ● Home Page: http://www.rajivshah.com ● LinkedIn: https://www.linkedin.com/in/rajistics/ ● Reddit: https://www.reddit.com/r/rajistics/ ━━━━━━━━━━━━━━━━━━━━━━━━━
Watch on YouTube ↗ (saves to browser)
Sign in to unlock AI tutor explanation · ⚡30

Related Reads

📰
Seven ranking frameworks, one search page, zero translation tables
Learn about 7 ranking frameworks for AI products and how they differ across categories, with no universal translation tables
Dev.to AI
📰
Beyond MCP: why your enterprise AI platform needs seven boundaries, not one protocol
Learn why a single protocol like MCP is insufficient for enterprise AI platforms and how seven boundaries can provide a more comprehensive solution
Dev.to AI
📰
Why Most Multi-Agent Systems Fail in Production (And How to Fix It)
Learn why multi-agent systems often fail in production and how to fix orchestration layer issues
Dev.to AI
📰
Build Durable Agents With Pydantic AI And Temporal
Learn to build durable agents with Pydantic AI and Temporal to prevent lost progress and wasted LLM calls in case of failures
Dev.to AI

Chapters (8)

From 25 lines of orchestration to two
0:28 The problem: hand-coding multi-agent systems
1:09 The solution: the agent writes its own orchestration
2:52 Running the demo: manual script vs workflow
4:03 Tracing the live run: fan out, cross-check, synthesize
5:33 The final research report
5:56 Observability with Laminar and chat-with-the-trace
7:23 Why this matters: models are eating orchestration
Up next
CLI vs API vs MCP Explained | Key Differences for AI Engineers
Pavithra’s Podcast
Watch →