Day 5: Databricks Academy – AI Agent Fundamentals
This wiki is a clean, practical revision guide based on the Databricks Academy Accreditation: AI Agent Fundamentals. Use it to revise concepts fast or publish as learning notes.
1. What is an AI Agent
An AI agent is an intelligent application that:
Uses an AI model, typically an LLM
Interacts with tools, data sources, and systems
Plans and executes a sequence of actions
Adapts dynamically to complete complex tasks
Key idea: Agents do not just respond, they act.
2. Common AI Agent Use Cases
AI agents are useful wherever decision making + execution is required.
Typical use cases:
Intelligent Document Processing
Knowledge Base + Semantic Search
Machine Learning assisted workflows
Conversational AI systems
Automation of business operations
If a problem needs reasoning, tool usage, and iteration, an agent fits.
3. How Databricks Uses AI Agents
Databricks applies AI agents internally and for customer solutions.
Examples:
Databricks Assistant for code, SQL, and data help
Customer Support Agents for faster issue resolution
Sales Agents for customer insights and recommendations
Core focus: enterprise scale, security, and data governance.
4. Powering AI Agents with Databricks Mosaic AI
Mosaic AI is Databricks’ unified platform to build, deploy, and govern AI systems.
It provides:
Model hosting and inference
Secure data access
Prompt management
Evaluation and monitoring
Mosaic AI removes glue work between data, models, and applications.
5. Large Language Models and Prompting
LLMs are the reasoning engine behind most AI agents.
Prompting defines:
What the model should do
How it should behave
What tools or context it can use
Better prompts = better agents. No shortcut here.
6. Mosaic AI Playground Parameters
The Playground helps experiment with models safely.
Important parameters:
Temperature
Controls randomness.
Low value: deterministic and focused
High value: creative and diverse
Top P
Limits token selection based on cumulative probability.
Lower value: safer and conservative output
Higher value: broader vocabulary
Top K
Limits the number of possible next tokens.
Lower value: predictable responses
Higher value: more variation
Rule of thumb: tune these based on reliability vs creativity needs.
7. Challenges in Building AI Applications
Building AI is hard. Databricks does not hide this.
Key challenges:
AI is expensive
- Compute, models, and infrastructure cost real money
Most AI problems are data problems
- Poor data = poor outcomes
Platforms need constant evolution
- New models, tools, and techniques appear rapidly
Productionizing AI is costly
- Monitoring, retraining, security, and compliance add overhead
This is why unified platforms matter.
8. Agent Systems Overview
An agent system defines how agents think, decide, and act.
Two major workflow types exist.
9. Agentic vs Non Agentic Workflows
Agentic Workflows
Dynamic and adaptive
Multi step reasoning
Tool usage and feedback loops
Suitable for complex tasks
Non Agentic Workflows
Static and predefined
Linear execution
Limited flexibility
Suitable for simple automation
Agents are chosen when uncertainty is high.
10. Agent Bricks Concept
Agent Bricks are reusable building blocks for agents.
They speed up development by offering predefined patterns.
Based on use case, common agent presets include:
Information Extraction Agent
Custom LLM Agent
Knowledge Assistant Agent
Multi Agent Systems
Supervision and Orchestration Agents
Think Lego blocks for enterprise AI.
11. Final Takeaway
This accreditation proves one thing.
You now understand:
What AI agents are
Why agentic systems matter
How Databricks approaches enterprise AI
Where Mosaic AI fits in the ecosystem
