WAT Framework (Workflows-Agent-Tools)

WAT Framework (Workflows-Agent-Tools)

A project-structure pattern for building agentic workflows inside Claude Code. Three layers: Workflows are markdown SOPs that describe a process end-to-end (the 'recipe'); the Agent is Claude itself, the decision-maker that reads workflows and chooses which tools to call; Tools are Python scripts that perform one specific action each (the 'ingredients'). Self-improvement is baked in – when a tool fails, Claude updates the tool or workflow file so the next run is smarter. The framework prevents file sprawl and context rot by imposing a standard folder layout defined in a project-level claude.md.

Key points

  • Workflows = markdown SOPs describing processes (the 'recipe')
  • Agent = Claude Code, the decision-maker that reads workflows and delegates
  • Tools = Python scripts with a single specific job (the 'ingredients')
  • Folder structure enforced via claude.md: /workflows, /tools, /temp, .env
  • Tools and workflows auto-update when they fail, producing a self-improving system
  • Modular: tools can be reused across multiple workflows
  • Typical project has a small set of composable tools (scrape, generate, export) that serve many workflows
  • Works best for human-triggered use cases; scheduled runs still require external deployment (e.g. Modal, Trigger.dev)
  • The 10-hour course clarifies the Agentic Build / Deploy Boundary: in production, teams usually deploy the workflow instructions and tools, not the live Claude Code agent [src-016]
  • .env is the expected place for API keys and secrets; CLAUDE.md should explain the project structure and rules without exposing credentials [src-016]
  • WAT is taught as a way to keep agentic builds modular enough to test, deploy, document, and hand over to clients [src-016]
  • Nate's deployment comparison turns WAT into a deployment diagnostic: /loop and cloud routines keep Workflows, Agent, and Tools together, while Modal/trigger.dev usually deploy Workflows and Tools unless an agent SDK is intentionally embedded [src-086].

Related entities

Related concepts

Source references

  • [src-005] Nate Herk cluster — Nate Herk — n8n cluster (18 videos)

– Videos referenced: 3GAxd90fEE4, tDGiWn0flK8, AO5aW01DKHo

  • [src-016] Nate Herk — "Build & Sell with Claude Code (10+ Hour Course)" (2026-03-12)
  • [src-086] Nate Herk — "I Tested 3 Ways to Deploy Claude Agents (Here's When to Use Each)" (2026-05-15)

Robin Cartier perspective

This page is part of Robin Cartier's working AI knowledge graph: a practical research layer for production AI, recommendation systems, experimentation, GEO, and agentic web readiness.

The useful next step is to connect this concept back to applied product leadership and operating models.

Recommended next

Keep reading from this thread

From 491 indexed pages and articles.

  1. Wiki concept Agentic Build / Deploy Boundary The agentic build / deploy boundary is the distinction between using an AI agent to design, debug, and improve a workflow Related by wat
  2. Wiki concept Agent Deployment Modes The practical choices for where an AI agent or automation runs and whether the live reasoning loop is still Related by deploy
  3. Insight AI Beyond POCs How enterprise AI moves beyond proofs of concept through ownership, governance, measurement, adoption, and production operating models Readers have engaged with this next