Skip to content
LinkedInX

State and Handoff

Target audience: Those losing information between workflow steps, those designing handoffs
Prerequisites: Basic understanding of What Is Orchestration

In multi-step workflows, information tends to get lost at the seams. Deciding what to retain and what to pass forward is state management.

Do Not Rely on Conversation History

The simplest handoff carries the conversation history into the next step. But as history grows, what matters gets buried, and rerunning only the failed step becomes impossible.

ApproachPartial rerunFindability
Conversation history onlyNot possibleHarder as history grows
Explicit statePossibleStructured and clear
This table scrolls horizontally. Keyboard users can focus the table and use the left and right arrow keys.

What to Retain as State

graph LR
    S1["Step 1"] -->|deliverable + assumptions| State["State"]
    State -->|only what is needed| S2["Step 2"]
    S2 -->|deliverable + assumptions| State

Three kinds of information belong in state.

KindExample
DeliverablesGenerated files, research results, decisions
Decision assumptionsRejected options and why, confirmed constraints
ProgressCompleted steps, remaining steps, failure counts
This table scrolls horizontally. Keyboard users can focus the table and use the left and right arrow keys.

What does not belong: trial-and-error detail and intermediate data already consumed.

Fix the Handoff Format

Fix the format so the receiving side never has to interpret. With a stable format, a step’s internals can be replaced without disturbing its neighbors.

# Example handoff format
status: completed | failed | needs_review
output: reference to the produced deliverable
decisions: what this step settled
open_issues: unresolved items carried to the next step

Carrying open_issues keeps undecided points from disappearing silently.

Progress Enables Reruns

Holding completed steps as state allows resuming from the failure point. This is also the prerequisite for failure detection and recovery. A design that restarts everything means every late-stage failure costs the whole run.

Summary

  • Handing off through conversation history buries information and prevents partial reruns
  • State should hold deliverables, decision assumptions, and progress
  • A fixed handoff format makes steps swappable
  • Holding progress as state allows resuming from the failed step
Quiz