Context Engineering: Building AI Agents That Work in Production

Context engineering is one of the most important parts of building reliable AI agents. An agent may have access to instructions, conversation history, retrieved knowledge, tool definitions and outputs, memories, and runtime state—but the model should not necessarily see all of that information at every step.

In this 3-hour live workshop, you'll build a context-engineered agent with the Claude Agent SDK, starting from a naive agent and adding external state, context selection, tool and knowledge retrieval, compression, isolation, and evaluation.

Most agents fail not because the model is weak, but because the context is wrong. Errors persist and poison later steps. Irrelevant history distracts. Too many tool definitions cause confusion. Contradictory information clashes. Larger context windows do not fix any of it — an agent may have access to instructions, conversation history, retrieved knowledge, tool definitions and outputs, memories, and runtime state, but the model should not see all of that at every step.

From there, we’ll organize context engineering around four core operations:

• Write — persist useful information outside the active context
• Select — retrieve only the memories, tools, knowledge, and state relevant to the current step
• Compress — manage growing trajectories through summarization, compaction, and trimming
• Isolate — keep unnecessary information outside the model’s context using state, execution environments, and subagents

We’ll then implement these ideas using the Claude Agent SDK. Starting from a simple agent, we’ll add scratchpads, context selection, tool and knowledge retrieval, compaction, context isolation, and subagents with separate context windows.

Finally, we’ll evaluate the agent to understand whether it is retrieving the right information, preserving important context, avoiding unnecessary tokens, and reducing common context failure modes.

What You’ll Learn

• What context engineering is and why it matters for AI agents
• What enters an agent’s context and how it grows over time
• How to identify context poisoning, distraction, confusion, and clash
• How to apply the Write, Select, Compress, and Isolate framework
• How to manage scratchpads, memory, tools, knowledge, and runtime state
• How to build these patterns with the Claude Agent SDK
• How to use compaction, subagents, and execution environments effectively
• How to evaluate context quality, token usage, and agent performance

What We’ll Build

We’ll progressively build a context-engineered agent:

Naive Agent → External State → Context Selection → Tool & Knowledge Retrieval → Compression → Isolation → Evaluation

Workshop Deliverables

• 3-hour live workshop
• Hands-on implementation with the Claude Agent SDK
• Complete code and prompts
• Workshop slides
• Session recording
• Context-engineering evaluation framework
• Patterns you can reuse in your own agentic systems

Workshop Details

Date: Sunday, 30 August
Time: 18:00–21:00 EEST
Location: Live on Zoom
Regular Price: $50
Early-bird Price: $40 with code CONTEXT40 until 23 August at 23:59 EEST