Build multi-step scenarios with AI Agent
Martina Farkasova
4 minutes
AI
Real content management challenges rarely fit into single prompts. You might need to audit content, clean up issues, translate approved changes, then publish across multiple languages. Learn how to break complex tasks into effective prompt sequences and match the right principles to each step.
From single prompts to multi-step scenarios
You’ve learned six principles for writing effective prompts. Now it’s time to combine them for real-world scenarios that require multiple steps. Consider this challenge: you’ve migrated 100 blog posts from your old CMS. Some contain H1 tags in the body that hurt SEO. Others are missing taxonomy terms. A few still reference the outdated product names. And everything needs to move through your review before publishing. One prompt won’t handle this. But a sequence of well-structured prompts can. The key is knowing which principles each step needs.Match principles to task complexity
Not every prompt needs all six principles. Use just enough to get the results you need.| Task type | Principles needed |
| Simple discovery or single-item changes | Essentials only (action, content scope, intent) |
| Quality-sensitive output | Essentials + quality boosters (success criteria, examples) |
| Bulk operations or high-risk changes | All six principles (including the safety pattern) |
Working through a complex scenario
Let’s work through that migration challenge: 100 blog posts migrated from your old CMS. Step 1: Discover what needs cleaning“Find all Article content items in the Blog collection where the Body Copy element contains H1 tags.”
The AI Agent returns: “Found 38 Article content items with H1 in Body Copy.”
About a third of your migrated content has this formatting issue, which tracks with what you expected.
Step 2: Verify the cleanup approach
“Show me 3 examples from those items with the specific H1 tags they contain.”
The AI Agent shows several <h1> tags that need removal.
Step 3: Clean the formatting
“For all 38 items, remove the H1 tags and convert the content to H2, H3, or H4 based on the heading position. Preserve the text content.”
Direct action works because you verified the scope. Success criteria ensure cleanup protects your content.
Step 4: Find items needing taxonomy terms
Now that we’ve cleaned up H1 tags, let’s find all Article content items that still need taxonomy terms applied.
“Find all Article content items in the Blog collection that have no taxonomy terms assigned from the Topics taxonomy group.”
The AI Agent returns: “Found 52 Article content items with no terms from Topics.”
About half of your migrated content needs tagging. This makes sense since the old CMS used a different taxonomy structure.
Step 5: Assign taxonomy terms
“Suggest taxonomy terms from the Topics group for those 52 items. Assign 2-4 terms per item based on their Body Copy. Prefer specific terms over broad categories.”
The agent provides suggestions. You review them and notice they look appropriate.
“Assign the terms you suggested.”
Step 6: Fix outdated product references
In your migrated content, you’ve identified 12 Article items that still reference old product names. Let’s clean those up.
“Update these 12 content items to replace references to 'ProductName Classic' with 'ProductName New': [list the 12 item names]. Keep the meaning, structure, and formatting the same, only update the product terminology.”
The complete sequence handled discovery, content cleanup, and taxonomy assignment. Each step used only the principles it needed: essentials for discovery, boosters for quality-sensitive work, and safety verification for bulk operations.
Review the AI Agent’s changes in the chatAfter each step that modifies content, the agent presents an expandable view of the changes made in the chat. Review and verify the agent’s modifications before proceeding to the next step. Confirm each step works correctly before building on it.Build your prompting instinct
Does it involve multiple content types or collections? Does it combine discovery with modification? Are there quality requirements for generated content? Does the scale require verification before acting? All these signals tell you that a multi-step approach is needed.
You can stay in the same conversation when you’re working with the same content and the same task flow. Start a new conversation when your goal or topic changes, when you switch to a different dataset, or when earlier instructions might interfere with what you’re trying to do next.
As you practice, you’ll develop intuition for which principles each situation needs. You’ll recognize when to compress steps and when to add extra verification. The six principles aren’t a rigid checklist; they’re tools you combine based on what each step requires.
Sign in with your Kontent.ai credentials or sign up for free to unlock the full lesson, track your progress, and access exclusive expert insights and tips!