You are still chaining prompts. Anthropic's engineers stopped a month ago

The 6 patterns and 14 steps the frontier company actually uses to migrate code, sort 1,000 items, and verify agents at scale

Jun 06, 2026

∙ Paid

The model is commoditizing. The harness is the moat.

Marc Andreessen said it to the a16z LPs in January. Dario Amodei said it at Davos. Demis Hassabis confirmed it on stage in May. The moat in 2026 lives in the system you build around the model, not in the model itself.

Anthropic just shipped the most operational version of that thesis to date.

Dynamic Workflows in Claude Code. Live since May 28, 2026. Claude writes its own custom harness in JavaScript, on the fly, for whatever task you give it. Subagents. Parallel execution. Per-agent model choice. Per-agent isolation. The same scaffolding Anthropic’s own engineers use to ship code migrations, deep research, 1,000-item sorts, and adversarial verification at scale.

9 of 10 builders have not tried it once.

The teams that internalize this before the rest of the market are the ones shipping reliable, auditable agents at production scale by Q4. The rest are still chaining 50 prompts by hand.

This article is the full playbook to be one of them. Six patterns. Fourteen steps. The code Anthropic’s engineers actually run.

For the strategic context, see the AI Agents library, the Prompting and Context Engineering library, and the Claude and Anthropic library.

The free section ends here.

Below the paywall:

▫️ The 6 patterns Anthropic’s engineers actually use, with copy-paste JavaScript for each:

Fan-out and synthesize
Classify-and-act
Adversarial verification
Generate-and-filter
Tournament
Loop-until-done

▫️ The 14-step roadmap to ship your first Dynamic Workflow this weekend

▫️ The 8 use-case compositions: migrations, deep research, sorting at scale, root-cause investigation, triage, design taste, lightweight evals, and verification

▫️ The quarantine pattern for untrusted input (prompt injection protection)

▫️ The 7 failure modes that burn 5-10x your token budget, with the fixes

▫️ The 30-day rollout from your first workflow to a shipped Claude Skill

Plus the entire AI Corner premium archive:

▫️ Claude and Anthropic library

▫️ Prompting and Context Engineering library

▫️ AI Agents library

▫️ AI Tools and Models library

▫️ Business and Investing library

▫️ Your voice is the only AI moat that compounds

▫️ I built a $0 second brain with Obsidian + Claude that compounds for life

▫️ Build your own stock analyst with Claude: the 12-prompt system