Free lesson
Map How Agentic Your Work Can Be
Watch the short video, then map one of your recurring workflows on the autonomy spectrum and find the safest place for an AI agent to start.
What does the video cover?
- Three terms you need: AI, agents and agentic, which is a degree of autonomy rather than a category.
- The spectrum: chatbot, assistant, agent and fully agentic, and how much oversight each needs.
- The question for your work: how autonomous could this process be, and where must a human stay in the loop?
The exercise: map one of your workflows
Stop asking "is this an agent?" and start asking how agentic it should be. This exercise applies that question to a workflow you own.
- Pick a recurring workflow you own, such as a weekly pipeline review, a monthly report or onboarding a new hire, and jot down its steps.
- Open a new chat in ChatGPT, Copilot (web), Copilot (work), Claude or Gemini. Then paste the prompt below with your workflow described.
- Review the table and check every place where it suggests a human check.
- Pick the first step to automate and set it up as a scheduled task or an assistant this month.
The prompt template
Character: You are an expert in AI agents and workflow design who helps non-technical leaders decide how much autonomy to give AI. Objective: I'm [role] at [company] in [industry]. Here is a recurring workflow I own: [describe the workflow step by step, who is involved and which tools are used]. Show me how agentic this workflow could become and where I should start. Do: - Break the workflow into its steps - Place each step on four levels of workflow autonomy: no AI (fixed rules), an AI step inside a scripted workflow, an AI agent that evaluates, decides and loops, or a fully agentic workflow - For each step, say where a human must stay in the loop and why (risk, judgment, external impact) - Recommend the one step to automate first: the lowest risk with the clearest time saving Don't: - Don't recommend full autonomy for steps that send external messages, change customer terms or approve spending - Don't assume tools or integrations I haven't mentioned; ask instead Output: A table with each step, its recommended autonomy level, the human check needed and the platform feature to use. Then one paragraph on where to start this month.
How do you make this prompt yours?
The template works as it is, but it does not know you yet. Five changes turn a generic result into one you trust.
- Describe the workflow as it really runs. Include the messy parts: who chases whom, where things wait and which spreadsheet everyone updates.
- Name the tools you are allowed to use. List your approved AI platform and the systems it can connect to, so the advice fits your company.
- State your risk lines. Add the decisions that must always stay human in your organization, such as anything customer-facing.
- Put a number on the time. Say roughly how many hours the workflow takes each week, so the AI can rank the savings.
- Run it for a second workflow. Comparing two workflows quickly shows where an agent would pay off most.
New to agents? Our guide to what an AI agent is explains the options in ChatGPT, Copilot, Claude and Gemini.
Agentic Leader
Want an agent built around your workflow?
Mapping one workflow by hand is a good start. In Agentic Leader, our four-week live course, LeadOS interviews you before week 1 and matches you to the workflows worth delegating. Then you build an agent that runs one of them, without code. New to building assistants? Start with AI Leader Advanced.
See Agentic Leader