Home / Productivity & AI Workflows / AI Workflows

Design a Reliable AI Agent Workflow in 8 Phases

Takes a repetitive piece of work and turns it into a dependable AI agent setup, step by step: what to delegate, a Claude Skill, a run-until-done loop, a schedule, a separate checker, context files, silent-failure checks and a human approval flow.

ai agentsclaude skillsautomationhuman in the loopreliability
  1. Phase 1 — Scope what an agent should do
    Role: you are a specialist in designing AI agents.

My repetitive work: [describe your repetitive work]

Give me, under three numbered headings:
1. The parts of this work that an agent can handle.
2. The parts that still need a human.
3. The first agent that is worth building, and why.
  2. Phase 2 — Write the Claude Skill
    Role: you are a specialist in Claude Skills.

The task I repeat often: [describe your recurring task]

Give me, under three numbered headings:
1. A SKILL.md file, ready to save, that teaches Claude to do this task the way I do it.
2. The trigger description that tells Claude when to use the skill.
3. A test that shows the skill actually works.
  3. Phase 3 — Design a run-to-completion loop
    Role: you are a specialist in getting agents to keep working until a task is finished.

My multi-step task: [describe your multi-step task]

Give me, under three numbered headings:
1. A goal condition that can be verified.
2. The instructions for the loop.
3. The checkpoint at which I step in and review.
  4. Phase 4 — Put the task on a schedule
    Role: you are a specialist in automating operations.

The task I want to run on a schedule: [task]

Give me, under three numbered headings:
1. The prompt to schedule.
2. The connector the task requires.
3. What will be waiting for me when I check back.
  5. Phase 5 — Separate the maker from the checker
    Role: you are a designer focused on agent safety.

What my agent produces: [output]

Give me, under three numbered headings:
1. A distinct checking step, separate from the agent that creates the output.
2. The quality criteria that step checks the output against.
3. The reason the agent that makes the output should never be the one that grades it.
  6. Phase 6 — Feed the agent its context
    Role: you are a specialist in giving agents the right context.

What my agent needs to know: [describe the knowledge it needs]

Give me, under three numbered headings:
1. The structure of the reference files the agent will read.
2. What goes into each file.
3. A routine for keeping those files up to date.
  7. Phase 7 — Plan for silent failures
    Role: you are a specialist in reliability.

The task my agent runs every day: [task]

Give me, under three numbered headings:
1. Three ways the agent can fail without anyone noticing.
2. A check that catches each of those failures.
3. The format of the alert I receive.
  8. Phase 8 — Design the human handoff
    Role: you are a specialist in human-in-the-loop workflows.

What my agent drafts: [output]

Give me, under three numbered headings:
1. The approval flow.
2. The rhythm for reviewing drafts in batches.
3. The rule that decides what can go out without my approval.
More in AI Workflows
A Team of 4 AI Employees With Their Prompts and WorkflowNewAI WorkflowsSets up an automated team of 4 AI roles (researcher, writer, designer, salesperson), giving the exact prompt to hand each one and how they cooperate to run your business.
Task: build me a team of 4 AI employees that automates my business:
1. Researcher
2. Writer
3. Designer
4. Salesperson

For each one, give me the exact prompt I should give it. Then explain how the four work together to run my business.
Open page
AI Tool Workflow That Works Around Your Time or Skill LimitsNewAI WorkflowsDesigns an efficient workflow that combines AI tools such as ChatGPT, ElevenLabs and Pictory to reach a specific goal despite a limit in time or skills, with as little effort as possible.
My limitation: [your limitation, e.g. "only 1 hour a day" or "no video editing skills"]
My goal: [your goal, e.g. "publish 3 high-quality YouTube Shorts every week"]

Question: how can I combine AI tools such as ChatGPT, ElevenLabs, Pictory and similar into a very efficient workflow that gets me to this goal with the least possible effort?
Open page
AI-Assisted Ghostwriting System That Keeps Each Client's VoiceNewAI WorkflowsBuilds an AI ghostwriting system for your clients: a voice training prompt that learns tone from 5 samples, a 15-minute first-draft workflow, a protocol that makes AI drafts sound personal and a quality bar above what clients would write.
Role: you are a production specialist for AI-assisted ghostwriting. You have helped more than 300 ghostwriters use Claude to deliver genuine content in a fraction of the usual time.

What I ghostwrite: [content type]
Who the clients are: [client type]
My aim: bring AI into my process while every piece still sounds personal and true to each client.

Task: build my complete AI ghostwriting system, with these four parts:
1. A voice training prompt that captures a client's tone from 5 examples of their writing.
2. A workflow that produces first drafts in less than 15 minutes.
3. A humanization protocol that makes AI drafts sound personal.
4. A quality benchmark set higher than what the clients could write themselves.

Return each part as its own section.
Open page
Audit and Redesign Your Creative Workflow with AINewAI WorkflowsAudits your current creative workflow for a project, spots where it slows down, repeats itself, loses quality or misses chances to automate, then redesigns it into a faster AI-powered process with tools, prompt frameworks and quality checkpoints.
Role: you are a consultant for creative work powered by AI, at the highest level.

Project: [project]
My current creative workflow: [describe how you work today]

If the workflow description is missing, ask me for it before you begin.

Part 1 — Diagnosis: point out the bottlenecks, the redundant steps, the gaps in quality and the chances to automate that I am missing.

Part 2 — Redesign: rebuild the whole process as a quicker, more efficient system driven by AI. Recommend:
- the best AI tools
- prompt frameworks
- quality checkpoints
- optimization strategies

Goal: more creativity, speed, consistency and final output quality, with less manual work.
Open page
End-to-End AI Workflow for Producing Any Content TypeNewAI WorkflowsDesigns a complete AI-assisted workflow for a content type, from research to performance analysis, naming the best AI tools for each stage, defining inputs and outputs and automating repetitive work for a fast, repeatable system.
Role: you are a world-class architect of creative workflows powered by AI.

Content type: [content type]

Task: design a complete workflow, supported by AI, for producing this type of content. Cover these stages in order: research, ideation, planning, scripting, design, creating the assets, editing, publishing, optimization, quality control, and finally analysis of performance.

For each stage give:
- the best AI tools to use
- clear inputs and outputs
- which repetitive tasks to automate, where possible

The system must be scalable and repeatable, getting the most out of speed, quality and consistency.
Open page