{"slug": "a-harness-for-every-task-dynamic-workflows-in-claude-code-master-custom-multi", "title": "A Harness for Every Task: Dynamic Workflows in Claude Code – Master Custom Multi-Agent Systems", "summary": "Anthropic's new dynamic workflows in Claude Code allow the AI to generate custom orchestration layers for specific tasks, enabling multi-agent systems that coordinate sub-agents, tools, and verification steps. This approach addresses failure modes in long-running sessions by providing structural resistance through patterns like fan-out-and-synthesize and adversarial verification. The feature marks a shift from prompt engineering to workflow engineering, where developers design prompts that make the system design itself.", "body_md": "**A Harness for Every Task: Dynamic Workflows (dynamic work orchestration) in Claude Code – How to Solve Complex Tasks with Tailor-Made Multi-Agent Systems**\n\nAs an AI-native builder, you already know the real bottleneck isn’t raw model intelligence anymore — it’s **context, persistence, and the ability to orchestrate work across time and parallel streams**. Anthropic’s new **dynamic workflows** solve exactly this by letting Claude instantly write and run a **custom harness** (a tailored orchestration layer that defines how agents are spawned, coordinated, verified, and concluded) for your specific task.\n\nThis deep dive is written for experienced AI-native developers, architects, and power users who already live in tools like Claude Code, Cursor, or Aider. We’ll break down the concept from Anthropic’s blog post “A Harness for Every Task,” explain the mechanics, showcase powerful patterns, deliver practical examples, and end with key concepts you can immediately apply to your own work.\n\nClaude Code ships with a strong default harness optimized for coding — handling files, tools, iterations, and reasoning in one coherent session. It works incredibly well for most daily tasks because many jobs *feel* like coding: structured, iterative, and with clear outputs.\n\nBut with longer, more ambitious, or high-stakes projects, classic failure modes appear:\n\nStatic custom harnesses built via SDKs or scripts were the old workaround — but they’re generic and maintenance-heavy. Dynamic workflows flip the script: **Claude generates the precise orchestration** for your exact task, on the fly.\n\nA **harness** (orchestration layer) is everything around the model: how the task is broken down, which sub-agents run, what tools they have, how outputs are verified, and when the job is truly done. It turns a single model into a coordinated team.\n\nWith dynamic workflows:\n\n`~/.claude/workflows`\n\n).This isn’t just “more compute.” It’s **structural resistance** against the common failure modes of long-running agent sessions.\n\nClaude composes workflows from these reusable patterns. Learn them so you can guide the model more effectively:\n\n**Classify-and-Act** (classify and route)\n\nA classifier agent determines the task type and routes it to the right behavior or specialized sub-agent.\n\n**Fan-out-and-Synthesize** (split wide and merge)\n\nBreak a big task into many small ones, run them in parallel in clean contexts, then let a synthesizer agent combine the results. Prevents cross-contamination.\n\n**Adversarial Verification** (opponent-based checking)\n\nEvery producer agent gets a dedicated “adversary” agent that checks output against a rubric. Kills self-preferential bias.\n\n**Generate-and-Filter / Tournament** (generate, filter, compete)\n\nCreate many ideas or solutions, deduplicate, and let them compete through pairwise judging. Excellent for creative or ranking tasks.\n\n**Loop-until-Done** (iterate until success criteria)\n\nKeep spawning agents until a stop condition is met (no new findings, no remaining errors). Perfect for debugging, research, and exhaustive analysis.\n\nThese patterns combine beautifully — for example, fan-out + adversarial verification + synthesize.\n\n**Coding & Engineering:**\n\n**Non-Coding Work (often where the biggest wins are):**\n\n`CLAUDE.md`\n\n.**Concrete Prompt Example (inspired by Anthropic):**\n\n“This test fails maybe 1 in 50 runs. Set up a workflow to reproduce it. Form competing theories about the race condition, and don’t stop until one theory survives all the evidence.”\n\nThe resulting workflow: Classifier → Fan-out hypothesis generation → Adversarial verifiers/refuters → Loop-until-done.\n\n**Benefits:**\n\nThis represents a natural evolution from *prompt engineering* to *workflow engineering*. You no longer design the full system — you design the prompt that makes the system design itself.\n\n**Trade-offs:**\n\nDynamic workflows mark a shift from “better single agent” to **orchestrated agent teams on demand**. It’s not just a feature in Claude Code — it’s a new way to think about reliable agentic work.\n\n**Try it today:** Take a complex task you’re struggling with and add “Use a dynamic workflow…” or the trigger “ultracode”. Watch how Claude designs the perfect harness. Iterate on the prompt, save what works, and level up.\n\nThe world isn’t getting simpler. But now we have a harness for every task.\n\nUse it wisely.\n\n**Further Reading / Primary Source**\n\n“A harness for every task: dynamic workflows in Claude Code” by Anthropic (June 2, 2026)\n\n**Highly Recommended Video**\n\nFor a practical, hands-on breakdown of **Ultracode** and dynamic workflows (including live demos and pro tips), watch this excellent video from Chase AI:\n\n[“the most POWERFUL claude code feature in months”](https://youtu.be/6cmi7qyFwEE)\n\nIt complements the Anthropic blog perfectly and shows exactly how to start using these capabilities right away.\n\nWhat complex task are you going to throw at a dynamic workflow first? Drop it in the comments — I’d love to see what you build.", "url": "https://wpnews.pro/news/a-harness-for-every-task-dynamic-workflows-in-claude-code-master-custom-multi", "canonical_source": "https://dev.to/netsi1964/a-harness-for-every-task-dynamic-workflows-in-claude-code-master-custom-multi-agent-systems-1edm", "published_at": "2026-06-21 09:06:50+00:00", "updated_at": "2026-06-21 09:36:44.083065+00:00", "lang": "en", "topics": ["large-language-models", "ai-agents", "developer-tools", "ai-infrastructure", "ai-research"], "entities": ["Anthropic", "Claude Code", "Claude"], "alternates": {"html": "https://wpnews.pro/news/a-harness-for-every-task-dynamic-workflows-in-claude-code-master-custom-multi", "markdown": "https://wpnews.pro/news/a-harness-for-every-task-dynamic-workflows-in-claude-code-master-custom-multi.md", "text": "https://wpnews.pro/news/a-harness-for-every-task-dynamic-workflows-in-claude-code-master-custom-multi.txt", "jsonld": "https://wpnews.pro/news/a-harness-for-every-task-dynamic-workflows-in-claude-code-master-custom-multi.jsonld"}}