AI Program Building Guide

Use this guide whenever you need to build a new AI-powered workflow. Walk through the steps in order, don't skip ahead. Steps 1 to 3 are structured: answer the questions before moving on. Steps 4 and 5 are collaborative: Claude will take the lead based on what you've defined.

Define the Problem

Before building anything, get crystal clear on what you're solving.

Answer these questions:

What is the problem you're trying to solve? Describe it in plain language.

Who has this problem? Is it you, your team, a client, a customer?

What does "solved" look like? What's the output or outcome when this works?

How is this being handled today? Manual process, not at all, badly?

How often does this need to happen? Once, daily, weekly, on-demand, triggered by an event?

Ready Check

Don't move to Step 2 until you can clearly articulate: "The problem is _____, and when it's solved, _____ happens."

Decide What Data It Needs

Now figure out what information the program needs to do its job.

Answer these questions:

What inputs does this need to run? Files, text, data from an API, user responses, etc.

Where does that data live today? Google Drive, a database, email, someone's head, a spreadsheet?

Is the data available right now, or does something need to be built to collect it?

Does the data change? Is it static or does it update over time?

Are there any sensitivities? Customer data, financial info, anything requiring access controls?

Ready Check

By the end of this step you should have a clear list: "This program needs X, Y, and Z as inputs, and they come from A, B, and C."

Build the Steps

Map out the workflow from start to finish. Think of it as: if you were explaining this process to a new hire, what would the steps be?

Answer these questions:

What is the first thing that happens? What kicks this off?

Walk through each step in order. What happens, then what, then what?

Which steps need a human? Where does someone need to review, approve, or make a decision?

Which steps could an AI handle? Where is the AI doing the heavy lifting?

Which steps are app integrations? Where does data move between tools or systems automatically?

What's the final output? A document, a message, a decision, an update somewhere?

Write it out as a numbered list. Be specific. For example:

10.New support ticket comes in via email.

11.AI reads the ticket and categorizes it (bug, feature request, question).

12.AI drafts a response based on category and knowledge base.

13.Human reviews the draft and edits if needed.

14.Human sends the response.

15.AI logs the interaction and updates the tracker.

Claude's Turn

Once you've mapped your steps, Claude will build a visual workflow diagram. This diagram will label each step as Human, AI, or App Integration so you can look at the entire flow and judge whether it makes sense, as if you were an employee being handed this process on day one.

Translate the Workflow into AI Instructions

Claude takes the lead here. This is the step most people cannot do on their own.

You've defined the workflow. Now Claude needs to translate it into language that AI can actually execute. This matters because your workflow contains three types of steps, human actions, AI-powered steps, and app integrations, and each one needs to be handled differently:

Human steps need clear handoff points: what does the person receive, what are they deciding, and where does their output go next?

AI steps need precise instructions: what data comes in, what the AI should do with it, what format the output should be in, and what guardrails to set.

App integration steps need defined triggers and data mappings: what fires the integration, what data passes through, and what the expected result is.

Claude will take your workflow from Step 3 and write the full set of instructions for every AI-powered step. You review them, pressure-test them, and refine them together until they're right.

How to Start

To kick this off, just say: "Here's my workflow, please write the AI instructions for it." Then paste or reference what you built in Steps 1 to 3.

Deploy and Schedule

Claude tells you exactly how to get this running.

The workflow is defined. The instructions are written. Now you need to ship it. Claude will walk you through each of these and tell you specifically what to do:

Where does this live? What repo, what structure, what files need to be created.

Does it need a schedule? If so, Claude will specify the cron schedule and explain how to set it up.

How do you deploy it? Step-by-step: what to commit, where to push, what to configure.

How does the team access it? Exactly how other people will trigger or interact with the workflow.

What does monitoring look like? How you'll know it's working and what to do when it breaks.

This isn't advisory, Claude should give you concrete instructions you can follow to get the program live.

Quick Reference

Step	You Do	Claude Does
1. Define the Problem	Answer the questions	Ask follow-ups if anything is unclear
2. Decide What Data It Needs	Identify inputs and sources	Help you think through gaps
3. Build the Steps	Map the workflow	Build a visual workflow diagram
4. Translate to AI Instructions	Review and refine	Write the prompts, handoffs, and integration logic
5. Deploy and Schedule	Follow the instructions	Tell you exactly how to deploy, schedule, and monitor

The Infinite Leverage Blueprint