Context Engineering – Managing Intelligence at Scale

Context Engineering for AI Agents: Managing Intelligence at Scale

Move beyond prompting – learn how to design and scale intelligent AI agents with context engineering.

AI agents are no longer limited by reasoning ability – they’re limited by how well they manage context. As Andrej Karpathy describes it: “The LLM is the CPU, the context window is RAM, and you’re the operating system.”

In this session, we shift from building agents from scratch to mastering context engineering for AI agents — the discipline of controlling what your agent sees, remembers, and prioritizes as it operates, powered by technologies from SambaNova Systems.

What You’ll Learn:

In this hands-on session, we’ll show you how to build and scale AI agents using create_deep_agent() while implementing advanced context engineering for AI agents patterns.

Starting from the harness built in Session 2, we’ll move into higher-level abstractions — adding automatic context compression, middleware for tool orchestration, runtime context propagation, and subagent-based task delegation.

You’ll also learn the key failure modes of modern AI systems — from token limits to attention degradation — and how to design agents that remain stable and performant at scale.

What You’ll Build:

During this session, we will build a full research AI agent powered by context engineering that can:

• Automatically compress and manage large context windows
• Delegate tasks to specialized subagents while maintaining clean supervision
• Intercept and enhance tool calls using middleware
• Persist and retrieve long-term memory across workflows
• Operate reliably across multi-step, high-context tasks

The final system will demonstrate all 5 context types working together in a single, production-style workflow – a complete example of context engineering for AI agents in practice.

What We’ll Cover:

• Why Context Engineering for AI Agents?
Understand the context collapse problem and why agents degrade over time

• The 5 Context Types
Input, runtime, compression, isolation, and long-term memory — your core mental model

• From Scratch to Abstraction
How create_deep_agent() replaces custom harnesses and accelerates development

• Context Compression in Action
Automatic offloading and summarization to prevent context overload

• Middleware for Tool Calls
Build a @wrap_tool_call layer to log, validate, and enrich agent actions

• Context Isolation via Subagents
Design multi-agent systems where tasks are delegated without polluting the main context

• Production Patterns
How tools like Claude Code and Cursor manage context at scale

Who Should Attend:

This webinar is ideal for:

• AI engineers and developers building agent-based systems
• Machine learning engineers working with LLMs
• Developers exploring autonomous and multi-agent workflows
• Anyone interested in scaling AI agents beyond basic prompting

Technologies & Concepts:

• Deep Agents (create_deep_agent())
• Context Engineering for AI Agents
• Middleware & Tool Orchestration
• Multi-Agent Systems (Subagents)
• Context Compression & Memory Management
• Python