Claude Can Now Build Its Own Harness... For Every Task
In this video I explain why using one coding harness (like Claude Code/Codex) for all knowledge work breaks down as the context window fills, leading to agentic laziness, self-preferential bias, and goal drift. I show how dynamic workflows solve this by splitting work into clean contexts via a JavaScript workflow that can spawn agents, run them in parallel, and use pipelines, choose models, resume after interruption, and optionally isolate work trees. Anthropic Blogpost: https://claude.com/blog/a-harness-for-every-task-dynamic-workflows-in-claude-code Dynamic Workflow: https://youtu.be/WnmVGVOPtrA My voice to text App: whryte.com Website: https://engineerprompt.ai/ RAG Beyond Basics Course: https://prompt-s-site.thinkific.com/courses/rag Signup for Newsletter, localgpt: https://tally.so/r/3y9bb0 Let's Connect: 🦾 Discord: https://discord.com/invite/t4eYQRUcXB ☕ Buy me a Coffee: https://ko-fi.com/promptengineering |🔴 Patreon: https://www.patreon.com/PromptEngineering 💼Consulting: https://calendly.com/engineerprompt/consulting-call 📧 Business Contact: engineerprompt@gmail.com Become Member: http://tinyurl.com/y5h28s6h 💻 Pre-configured localGPT VM: https://bit.ly/localGPT (use Code: PromptEngineering for 50% off). Signup for Newsletter, localgpt: https://tally.so/r/3y9bb0 00:00 Why One Harness Fails 02:58 Static vs Dynamic Example 03:53 Anthropic Patterns Overview 04:19 Pattern Classify and Act 04:52 Pattern Fan Out Synthesize 05:29 Pattern Critic and Rubric 06:08 Pattern Generate and Filter 06:44 Pattern Tournament Bracket 07:36 Pattern Loop Until Done 12:03 Design Considerations and Costs
