Context Engineering for Agent Swarms (Workshop)
AI agent swarms are powerful until they break in the real world. The failure is rarely the model. It is almost always the system around it: messy context, unclear tool use, and no verification loop.
This event is a practical deep dive into context engineering for agent swarms. You will learn the framework top agent products converge on, then implement the core building blocks in a hands-on workshop and demo what you built.
What you will learn
You will leave with a clear mental model for scaling from single-agent loops to multi-agent swarms, and the ability to apply the five primitives of reliable agent systems:
Ontology: shared semantics and governed actions
Context graph: receipts, artifacts, evidence, and decisions as durable runtime truth
Skill graph: capabilities with contracts, preconditions, and phase gating
Context compiler: ranking, packing, compaction, and recitation to fit a finite token budget
Harness and verification loop: PLAN → ACT → VERIFY so evidence changes what the agent sees next
What you will build
A minimal, working set of primitives that makes an agent workflow more reliable. You will see how verification evidence updates the next context pack, and how swarms coordinate through shared state.
Who this is for
Engineers already building with coding agents (Codex, Cursor, Claude Code-class tools)
Builders exploring multi-agent systems who want a concrete framework for shared state, routing, and verification
Not ideal if you are brand new to agents or want an intro to prompting.
Real-world references
We will ground everything in modern patterns from tools and systems like Cursor, Manus, Claude Code, Codex-style harnesses, and Palantir-inspired ontology thinking.
Rootly and AWS
Agenda
5:00 to 5:30 PM: Networking and refreshments
5:30 to 6:30 PM: Context Engineering Lecture
6:30 to 8:00 PM: Context Engineering Workshop
8:00 to 8:30 PM: Live demos and wrap-up
What to bring
Bring a laptop. Seats are limited and the workshop is interactive.