A harness for every task: dynamic workflows in Claude Code

2026-06-02 · Source: Claude Blog · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Advanced, long

Summary

Claude Code has introduced dynamic workflows, enabling the AI to write and orchestrate its own multi-agent "harnesses" on the fly for specific tasks. This feature, released last week, allows Claude to move beyond its default coding harness to tackle complex problems like research, security analysis, and code review more natively. Dynamic workflows execute a JavaScript file to spawn and coordinate subagents, each with its own context window, combating issues like agentic laziness, self-preferential bias, and goal drift common in single-context tasks. Unlike static workflows, these are custom-built by Claude Opus 4.8 for tailored use cases, offering patterns such as classify-and-act, fan-out-and-synthesize, and adversarial verification. While powerful for complex, high-value tasks, they consume more tokens and are not necessary for every routine coding task.

Key takeaway

For AI Engineers or Machine Learning Engineers tackling complex, multi-part tasks, you should explore Claude Code's dynamic workflows to mitigate common AI failure modes like agentic laziness or goal drift. By leveraging custom, multi-agent harnesses, you can achieve higher performance and reliability in areas such as deep research, large-scale migrations, or adversarial verification. Consider integrating these workflows with /loop and /goal for continuous, high-value operations, but monitor token usage for cost efficiency.

Key insights

Claude Code's dynamic workflows enable on-the-fly multi-agent orchestration to overcome single-context AI limitations.

Principles

• Orchestrate subagents to prevent agentic laziness and bias.
• Isolate subagent goals to maintain task fidelity.
• Custom harnesses outperform generic static workflows.

Method

Dynamic workflows execute a JavaScript file with special functions to spawn and coordinate subagents, allowing selection of models and worktree isolation.

In practice

• Use "ultracode" or prompt Claude to create a workflow.
• Combine with /loop and /goal for continuous tasks.
• Set explicit token usage budgets for workflows.

Topics

• Claude Code
• Dynamic Workflows
• Multi-Agent Systems
• AI Orchestration
• Prompt Engineering
• Large Language Models

Best for: AI Architect, AI Engineer, Machine Learning Engineer, Prompt Engineer

Related on AIssential

• Claude Code Adds Dynamic Workflows for Parallel Agent Coordination — Claude Code's Dynamic Workflows enable parallel AI agent coordination for complex software engineering tasks.
• A Harness for Every Task: Putting a Team of Claudes on One Job — Dynamic Workflows overcome single-agent limitations by orchestrating fresh-context sub-agents via a code-based plan.
• Introducing dynamic workflows in Claude Code — Claude Code's dynamic workflows orchestrate parallel subagents to autonomously tackle complex, large-scale software development tasks end-to-end.
• TAI #207: Claude Opus 4.8 Is Better, but Dynamic Workflows Are the Bigger Story — Dynamic Workflows, where LLMs orchestrate multi-agent tasks, represent a significant shift towards scalable, complex problem-solving, despite increased token costs.
• Harnessing Claude’s intelligence — Building with evolving AI requires adapting agent harnesses to leverage new model capabilities and optimize performance.
• How Claude Code Builds a System Prompt — Advanced AI agents like Claude Code dynamically assemble complex system prompts using numerous conditional components.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Claude Blog.