Your Workflow for the Day¶
Explore, Plan, Implement, Verify¶
You now know how to write good prompts and how AI behaves. The last piece is a simple workflow that keeps you on track, especially when you're building something bigger than a single prompt can handle.
Every time you work with AI, follow these four steps:
1. Explore¶
Before you start building, understand who you're building for. What do your users need? What problems are they trying to solve? What would make their life easier? This is the discovery step. You're building enough empathy with your user that you could write a story from their perspective.
You can use AI to help with this exploration:
What does a backcountry hiker need to know before heading into Yellowstone?
What decisions are they making, and what information would help them plan a safer route?
2. Plan¶
Describe what you want to build as a user story with acceptance criteria. This is where the story format from the last section pays off. Your story IS your plan.
As a backcountry hiker planning a multi-day route in Yellowstone,
I want a conditions card showing active NPS alerts
so that I can quickly assess what hazards to prepare for before leaving the trailhead.
Given the NPS alerts API returns a "Danger" alert for Yellowstone,
when I view the conditions card,
then I see the alert type ("Danger"), the alert title, and a one-line description.
Given the NPS alerts API returns multiple alerts of different types,
when I view the conditions card,
then each alert is listed with its type (Danger, Closure, Caution, Information) and a summary.
3. Implement¶
Tell AI to go ahead and build it. You can send the whole story as your prompt. AI will use both the story and the acceptance criteria to guide what it builds.
Build this as a clean, mobile-friendly interactive prototype.
4. Verify¶
Walk through each acceptance criterion. Pass or fail. No guessing.
- Conditions card shows active alerts? Check.
- Each alert displays type, title, and description? Check.
- Multiple alert types (Danger, Closure, Caution, Information) all listed? Check.
- Missing something? Tell AI exactly which criterion failed: "The Danger alert is showing the type but not the description. Add the description text below the alert title."
Then the cycle repeats. Verify reveals a new thing to fix → you plan it → AI implements → you verify again. The acceptance criteria you wrote in Step 2 are the same checklist you use in Step 4.
Right-Sizing Your Prompts¶
Not everything needs a four-step process. Match the complexity of your prompt to the complexity of the task.
- Small ask → just say it in plain English: "Change the background color to light blue"
- Medium ask → write a user story with acceptance criteria: "As a traveler, I want a packing checklist so that I don't forget essential items. Given the list has 8 items, when I view it, then items are organized by category (clothing, toiletries, documents) with checkboxes."
- Big ask → break it into smaller stories first, then use the workflow for each one
You'll develop a feel for this as the day goes on. When in doubt, write a story. If the acceptance criteria keep growing and you can't list them all, that's your signal to break the task into smaller pieces.
Your First Cycle Together
Mob Session | ~5 minutes total | Gather around one screen. One person drives, everyone else navigates.
New to mobbing? The whole team gathers around one screen. One person is the "driver" (hands on the keyboard) while everyone else "navigates" (guiding what to do, watching for mistakes, thinking ahead). You'll rotate who drives throughout the day.
Start a new conversation in your AI chat tool, then walk through the first three steps of the Explore → Plan → Implement → Verify cycle together:
- Explore (~1 min): Ask AI: ```
What does a backcountry hiker need to know before heading into Yellowstone? What decisions are they making, and what information would help them plan a safer route? ``` 2. Plan (~3 min): As a team, pick one thing from AI's answer. Write a quick user story with 1-2 acceptance criteria around it. 3. Implement (~1 min): The driver sends the story as the prompt. Your AI assistant will start building.
**Don't wait for it to finish.** Your assistant may take a few minutes to implement what you asked for. Keep moving through the curriculum to the next page. When you get to your Challenge, come back and **Verify**: walk through each acceptance criterion against what your assistant built. Pass or fail? Did the acceptance criteria help you spot anything you might have missed at first glance?
!!! ai-assistant "In Your AI Assistant"
=== "Claude.ai"
The driver opens [claude.ai](https://claude.ai){:target="_blank"}. Share the screen so the whole team can see the conversation.
=== "ChatGPT"
The driver opens [chatgpt.com](https://chatgpt.com){:target="_blank"}. Share the screen so the whole team can see the conversation.
!!! ai-assistant "In Your AI Assistant"
**If you got code instead of a preview:** Sometimes AI chat tools output raw HTML code instead of showing you a visual page. This is normal, and it can happen differently each time, even with the same prompt. A few things to try:
- Ask your AI assistant: *"Can you show me this as a working page I can interact with?"*
- Use action words in your prompt like "build me a page" rather than "write me HTML code"
- In Claude.ai, look for a **Preview** button below the response
- In ChatGPT, try asking: *"Can you run this so I can see it?"*
If none of that works, try starting a new conversation in a new chat window. You may get different results. You'll learn more about why this may get things working in Section 4.
Key Insight
AI makes building easy. Knowing what to build is the hard part, and the part that matters. Explore → Plan → Implement → Verify keeps you focused on the outcome, not just the output. Without Explore and Plan, you'll generate something fast that nobody needs. With them, you build something that actually solves a real problem for a real person. That's the difference between producing software and producing impact.