Multi-Agent Orchestration

Assign different models and agent presets to each workflow step for planning, implementation, and review.

Overview

Zenflow workflows are built from steps. Each step can run with a different agent, model, and configuration — a planning-focused model for the planning step, a fast model for implementation, and multiple independent reviewers in parallel for code review.

The Three Phases

Most multi-model workflows break into three phases. Each has a different quality/efficiency tradeoff:

Phase	What matters	Good fit
Planning	Reasoning depth, architecture decisions, spec quality	Claude Opus, GPT-5.4
Implementation	Speed, code generation, tool use	Gemini Flash, Claude Sonnet, Codex
Review	Catching issues the implementation model missed	A different model than the one that wrote the code

The goal is to use the right model for each phase — frontier reasoning where depth matters, fast models where throughput matters.

Why Cross-Model Review Matters

A model reviewing its own output is less likely to catch issues — it tends to agree with its own reasoning. A different model brings a different set of assumptions and catches different classes of problems. For higher-stakes changes, running multiple reviewers in parallel gives broader coverage than a single reviewer regardless of which model you pick.

Two Approaches

Agent Presets

Define named model configurations in Settings and bind them to workflow steps. Best for sequential workflows reused across tasks.

Subagent Pipelines

Spawn isolated sub-processes with different models, contexts, and skills. Best for parallel execution and context isolation.

Choosing Your Approach

Situation	Recommended approach
Small task, clear scope	Single model, no preset needed
Recurring workflow, sequential steps	Agent Presets with <!-- agent: preset-name -->
Higher-stakes change needing multi-model review	Sequential review presets
Need parallel review or context isolation	Subagent Pipelines
Want multi-model review without a full pipeline	/comprehensive-review from chat

The Multi-model workflow is available as a built-in option when creating a task — select it from the workflow picker at the bottom of the task creation screen:

Runtime Model Switching

You don’t have to define everything upfront. During any step, you can switch the active model from the chat composer dropdown. This is useful when:

• A step is stuck and you want to try a different model
• You want a quick second opinion on a specific question
• The default preset for a step isn’t performing well on a particular task

The model switch applies to the current chat session. It doesn’t change the workflow definition.

Practical Guidelines

• Start simple. A two-preset setup (planner + implementer) already captures most of the value. Add review presets or the full Subagent pipeline when the task warrants it.
• Don’t use the full pipeline for small tasks. A three-phase Subagent pipeline has real overhead. For a small bug fix, a single model is faster and equally effective.
• Match model strengths to phase requirements. Planning needs reasoning depth. Implementation needs speed. Review needs a different perspective.
• Use the spec as the coordination artifact. When different models handle different phases, the plan or spec is what keeps them aligned.

Deep Dives

Agent Presets

Named model configs assigned to workflow steps

Subagent Pipelines

Parallel execution with isolated contexts

Review Rubric

Structured evaluation criteria for review steps