Topic
The single design decision underneath every multi-agent AI workflow: when you hand work to a sub-process, should it inherit what you already know — or start cold?
Target Reader
Knowledge entrepreneurs and builders running multi-step AI workflows (writing pipelines, research, review loops) who feel their conversations bloat and their sub-agents either drift or get contaminated — but can’t name why.
The Fear / Frustration / Want / Aspiration
Frustration: “My AI workflows produce muddy results and my context fills with junk.” Want: a clean mental model for delegating to AI that they can apply without a CS degree.
Before State
They dump everything into one long conversation, or spin off sub-agents arbitrarily, with no rule for what each one should start with. Reviews come back sycophantic; drafts ignore decisions already made.
After State
They hold one diagnostic question and apply it per step — and their pipelines stay clean, their reviewers stay honest, and their context stops ballooning.
Narrative Arc
Everyone obsesses over which model and which prompt. The quieter, higher-leverage decision is about context inheritance — and it turns on a single question most people never ask. Once you ask it, fork-vs-spawn becomes obvious every time.
Core Argument
Fork and spawn both isolate context and return only a result; the only thing that differs is whether the child inherits the parent’s context — and that choice is governed by one question: would the child do better seeing what the parent decided?
Key Evidence / Examples
- “Fork and spawn both create separate working contexts, but they behave differently. If yes, fork. If no, spawn.” — Lou, 2026-06-11
- Fork the drafter (it should honor the chosen angle); spawn the adversarial reviewer (it shouldn’t inherit your confidence in the current direction).
- The token payoff: a 350K-token forked pipeline returns only the final artifact to your context — Insight - Forked Skills as Context Isolation — Run Sub-Agents Without Polluting Your Conversation
- Same logic at workflow scale: Insight - Isolation Outperforms Debate When the Goal Is Discovery, Not Refinement
Proposed Structure (5–7 beats)
- The thing everyone optimizes (model, prompt) and the thing they ignore (context inheritance).
- Define fork and spawn in plain English — both isolate, one inherits.
- The one question that decides it.
- Worked example: a writing pipeline (fork the draft, spawn the critique).
- Why spawning is what makes AI review actually useful (no sycophancy by inheritance).
- The context/cost dividend and the “re-run your last query” trick.
- Close: delegation is a context-design decision, not a model decision.
Related Insights
- Insight - Fork vs Spawn — Decide Whether the Child Should Inherit What the Parent Knows
- Insight - Forked Skills as Context Isolation — Run Sub-Agents Without Polluting Your Conversation
- Insight - Capability-Architect — Compile a Workflow Into an Inheritable Skill Bundle
Editorial Notes
Keep it conceptual-but-concrete; resist drowning the reader in harness syntax. The “adversarial reviewer shouldn’t inherit your confidence” example is the one that makes it click — lead the second half with it. Pairs naturally with a model-routing piece; cross-link rather than merge.
Next Step
- Approved for drafting
- Needs revision
- Deprioritised