AI Agent Co-working Saturday

AI Agents Co-working #3: Context Engineering & Multi-Agent Systems

Event description:

Why Context Engineering & Multi-Agent Systems?

You can build a single agent in an afternoon. Making it actually useful? That's where most people get stuck.

Two concepts separate toy demos from production agents:

Context Engineering

Multi-Agent Systems

These two concepts unlock everything else. Nail them, the rest is details.

What We'll Cover

Context Engineering

- What actually goes into the context window (and what shouldn't)

- The "context budget" mental model — how to think in tokens

- System prompts that work vs. prompts that fight the model

- Managing conversation history without blowing up costs

- When to retrieve vs. when to summarize vs. when to forget

Multi-Agent Systems

- When one agent isn't enough (and when it is — don't overcomplicate)

- Three patterns that work: handoff, parallel, hierarchical

- How agents talk to each other without spaghetti code

- Debugging multi-agent flows (the hard part no one talks about)

- Live example: two agents coordinating on a real task

Format: Co-working, Not Lecturing

This isn't a 2-hour presentation. It's a working session.

- 10-15 min: Quick walkthrough of core concepts

- 30-40 min: You build. We help. Bring your project or start fresh.

- 15-20 min: Show & tell — see what others built

- Ongoing: Hang out, ask questions, meet other builders

Bring your bugs. Bring your half-finished projects. Bring your "why doesn't this work" questions.

Who Should Come

- You've built a basic agent but it keeps breaking in weird ways

- You're trying to connect multiple agents and it's turning into spaghetti

- You've heard "context engineering" everywhere but don't know what it actually means

- You want to build alongside people solving similar problems

Some Python experience helps. Laptop required.

Before You Come

- Install Python 3.10+

- Optional: pip install connectonion

- Optional: Check docs at https://docs.conne