Context: Posting this after today’s OpenAI visual agent builder announcements.
The link is not vendor specific; it isolates one design choice in the visual layer: static flowcharts vs runtime handoffs with full history.
Scope: We are comparing a pure visual flowchart to a handoff model built around one primitive:
handoff(agent, history).
Each agent takes the full conversation history H, updates it, and decides which agent should act next (or halt). The entire history travels with control, so routing happens dynamically.
Takeaways:
• Topology: flowchart = fixed graph; handoffs = graph generated at runtime from H
• State: flowchart stores state in the diagram; handoffs keep it in the shared data (H)
• Routing cost: flowchart needs explicit branches (many edges or a large router); handoffs keep routing as a small function over H
• Extensibility: flowchart adds edges; handoffs add a function
Scope: We are comparing a pure visual flowchart to a handoff model built around one primitive: handoff(agent, history).
Each agent takes the full conversation history H, updates it, and decides which agent should act next (or halt). The entire history travels with control, so routing happens dynamically.
Takeaways:
• Topology: flowchart = fixed graph; handoffs = graph generated at runtime from H
• State: flowchart stores state in the diagram; handoffs keep it in the shared data (H)
• Routing cost: flowchart needs explicit branches (many edges or a large router); handoffs keep routing as a small function over H
• Extensibility: flowchart adds edges; handoffs add a function
COI: We built Rowboat around handoffs using OpenAI’s Agents SDK. The code is open source: https://github.com/rowboatlabs/rowboat