- Post History
- Subscribe to RSS Feed
- Mark as New
- Mark as Read
- Bookmark
- Subscribe
- Printer Friendly Page
- Report Inappropriate Content
3 weeks ago - edited 2 weeks ago
From Blueprint to Build
Design & Patterns
← Back to Masterclass Overview
- Agent Anatomy: Every AI Agent has 5 core components – Role, Instructions, Tools, Frontends, and Task Types
- 5 Task Types: Context → Retrieve → Decide → Act → Evaluate – understand these and you understand any agent
- Marathon Approach: Start with 5K (OOTB), progress to 10K (customize), then Half Marathon (build), finally Full Marathon (orchestrate)
- Specialized Agents: Build focused agents for specific tasks (max 15 tools), not mega-agents that try to do everything
Agent vs. Assistant: What's the Difference?
Before we dive into building, let's clarify the fundamental difference:
- Reactive – waits for questions
- Single-turn – answers and stops
- Information-focused – provides answers
- Example: "What's the weather?"
- Proactive – takes initiative
- Multi-step – plans and executes
- Outcome-focused – delivers results
- Example: "Plan my team event considering weather, availability, and budget"
Thomas Geering: "The Assistant answers questions. The Agent delivers outcomes. The Assistant is reactive. The Agent is proactive. That's the key difference."
Agent Anatomy: The 5 Core Components
Think of building an AI Agent like hiring a new team member. You need to define who they are, what they know, and what they can do:
The 5 Fundamental Task Types
Every AI Agent, regardless of complexity, performs a combination of these five fundamental task types:
| Task Type | What It Does |
|---|---|
| 1. Context | Understand the situation, gather information, normalize data. "What's happening here?" |
| 2. Retrieve | Find information from systems, knowledge bases, databases. "What do I need to know?" |
| 3. Decide | Make judgments, evaluate options, choose next steps. "What should I do?" |
| 4. Act | Execute actions, trigger workflows, update records. "Make it happen." |
| 5. Evaluate | Assess results, learn from outcomes, improve. "Did it work? What's next?" |
Key Insight: "Understand these five task types and you understand how to build any agent. Every workflow is just a combination of Context → Retrieve → Decide → Act → Evaluate."
The Marathon Approach: Start Small, Scale Smart
Don't try to run a marathon on day one. Build your AI Agent capabilities progressively:
- Turn on out-of-the-box agents
- Test with existing data
- Learn the basics
- Add custom tools
- Configure search sources
- Tune prompts
- Design from scratch
- Custom workflows
- NAVA integration
- Orchestrator agents
- Agent-to-agent communication
- External integrations
Training Principle: "Start with 5K. Master it. Then move to 10K. Don't try to run a marathon on day one – you'll just hurt yourself."
The 7 Tool Types
ServiceNow provides seven types of tools that agents can use:
Semantic Search: Beyond Keywords
One of the most powerful capabilities of AI Agents is semantic search – understanding meaning, not just matching words:
"Laptop broken" only finds results with exactly those words
"Laptop broken" also finds "computer defect", "notebook not working", "device malfunction"
Best Practices for Building Agents
- Max 15 Tools: Build specialized agents, not mega-agents. Too many tools confuse the LLM.
- Clear Tool Names: Use descriptive names that help the agent understand when to use each tool.
- Good Descriptions: Write tool descriptions as if explaining to a new employee.
- Test Iteratively: Start simple, test, refine, add complexity gradually.
- Monitor & Improve: Use AI Control Tower to track performance and optimize.
- Start Your 5K: Activate an out-of-the-box agent in your instance and test it
- Map Your Tasks: Identify which of the 5 task types your current workflows need
- Continue Learning: Proceed to Session 4 to learn about prompts, data, and control
Last updated: January 2026
- 262 Views
