Intel / Learning Center
CE · COURSE
Free · video series
Context Engineering
How AI apps decide what the model sees — from your first prompt to multi-agent systems.
Learn what an AI model actually reads, why it gets things wrong, and how the teams behind real AI products decide what to put in front of it. No coding needed.
Start here
3 min
Watch on YouTube
M1
M2
M3
M4
M5
M6
M7
51
Video lessons
7
Modules
141
Vertical shorts
2–5
Minutes each
Free
No sign-up
21 of 51 lessons live
41%
// 01
Who it's for
No code needed. Nothing skipped.
Anyone curious about how AI products really work. Built so a non-technical person can follow every lesson and finish the whole course — and deep enough that engineers still pick up production practice.
The promise
“By the last video, someone who has never written code can explain what an AI model sees, why it gets things wrong, and how the people who build AI products decide what to put in front of it.”
// 02
How every lesson works
01
One idea per lesson
Each lesson is 2–5 minutes and teaches exactly one idea.
02
Simple picture first
Every lesson starts with a plain-English picture anyone can follow.
03
A real example
Then the same idea inside a product you already use.
04
Under the hood
Then the actual mechanism, with every term explained as it appears.
05
In production
And finally how teams building AI products really do it.
// 03
What you'll learn
Seven modules, from your first prompt to teams of agents.
Free to watch
Watch in order, or jump to any lesson.
No sign-up needed.
// 04
Curriculum
Every lesson, in order.
21 / 51 LIVE
Module 1 · 8 lessons
Foundations of Context Engineering
Know what a model actually sees, why it forgets, and why what you put in front of it matters more than anything else.
All live
Module 2 · 7 lessons
System Prompt and Instruction Design
Write the system prompt that is in the context window for every call: clear, at the right level, testable.
All live
Module 3 · 9 lessons
Knowledge Retrieval and Genie Configuration
Find the right facts in a huge pile of documents and data, and bring only those into the model's context window.
6/9 live
Next up
Text-to-SQL: Ask Your Data Questions With Databricks Genie
For numbers, the right page is a query, not a paragraph.
Coming
How to Set Up a Databricks Genie Space
Genie is only as good as the context you give it about your data.
Coming
Testing Databricks Genie With Benchmarks
A set of known questions with known answers keeps Genie honest.
Module 4 · 7 lessons
Memory Architecture with Lakebase and MLflow
Give a forgetful agent a memory store — and see exactly what was in the context window when something went wrong.
0/7 live
Coming
AI Memory Explained: Short-Term vs Long-Term
This conversation vs what should carry over to the next one.
Coming
What Should an AI Agent Remember?
Memory is a write policy: most things should be forgotten.
Coming
Memory Recall: The Right Memory at the Right Time
Memory only helps if the right entry reaches the context window at the right time.
Coming
Databricks Lakebase: Postgres Memory for AI Agents
Memory needs a fast, reliable store beside your data.
Coming
Agent Checkpoints: Resume Where You Left Off
Save the agent's state at each step so work can resume.
Coming
MLflow Tracing: Debug What Your AI Saw
When an answer is wrong, look at the context window, not the answer.
Coming
How to Evaluate AI Agents With MLflow
Measure whether a context change helped, and keep every version.
Module 5 · 7 lessons
Tool Design, MCP, and Agent Context
Let the model act — and understand that every tool is also something it has to read.
0/7 live
Coming
How AI Tool Calling Works
A tool is a description in the context window and a result that lands on it.
Coming
How to Design Tools for AI Agents
Clear names, clear descriptions, few overlapping tools.
Coming
Tool Results: Don't Flood Your AI's Context
A tool that returns everything floods the context window.
Coming
What Is MCP? Model Context Protocol Explained
One standard plug so any tool can connect to any AI app.
Coming
Inside an MCP Server: Tools, Resources and Prompts
A server offers tools to act, resources to read, and prompts to reuse.
Coming
Too Many Tools: Why AI Agents Get Confused
Every tool you add makes choosing harder and the context window fuller.
Coming
How AI Agents Work: The Agent Loop
Think, act, observe, repeat — and every lap adds to the context window.
Module 6 · 6 lessons
Context Compression and Compaction
Keep a long-running agent sharp when the context window keeps filling up.
0/6 live
Coming
Context Rot: Why Long AI Chats Get Worse
Quality drops as the context window fills, long before it overflows.
Coming
Trimming Chat History: Sliding Window Memory
The simplest fix: drop the oldest pages.
Coming
Summarising Chat History for AI
Many old pages become one short page.
Coming
Clearing Old Tool Results From AI Context
Raw tool output is the safest thing to remove once used.
Coming
Context Compaction: How AI Agents Summarise Themselves
When the context window is nearly full, rebuild it from what matters.
Coming
Agent Memory Files: Notes Outside the Context Window
Write progress to a file; read it back when needed.
Module 7 · 7 lessons
Multi-Agent and Long-Horizon Task Design
Split big work across agents with clean contexts, and carry a task across days.
0/7 live
Coming
Multi-Agent AI: When to Use More Than One Agent
Several clean contexts beat one overloaded context.
Coming
Orchestrator and Worker Agents Explained
One lead plans; workers go deep; results come back condensed.
Coming
Agent Handoffs: Passing Context Between AI Agents
What one agent passes to the next decides what the next one knows.
Coming
Long-Running AI Agents: Tasks That Take Days
Work longer than one context window needs a plan, a checklist and a record.
Coming
Resuming AI Agents in a Fresh Context Window
A fresh context window can pick up the work if the notes are good.
Coming
Why Multi-Agent AI Systems Fail
More agents means more ways to lose context between them.
Coming
Context Engineering: The Complete Picture
One agent, every layer of the course, on real infrastructure.
Free lessons
Learn what the model sees.
51 free lessons and 141 shorts on the PAR2 LABS YouTube channel.