Design Your Multi-Agent System

Livestream Event Link : go.foundercoho.com/livestream-8-12 📝 Event Description Building a single AI agent with prompt engineering is one thing. Designing a production-ready multi-agent system that scales reliably without running up massive token costs or looping infinitely is another. In this interactive session, we’ll walk through how to design a multi-agent architecture from a real product idea or workflow. Rather than focusing on prompts or framework APIs, we’ll focus on the architectural decisions that determine whether an agent system is scalable, maintainable, and production-ready. Together, we’ll explore questions such as: How do you decompose a problem into multiple agents?Which agent design pattern best fits your use case?When should a workflow be deterministic, non-deterministic, or a hybrid of both?Where should memory live, and when is it actually necessary?Which tasks should rely on an LLM, and which are better handled with deterministic logic?How do evaluation, observability, and human-in-the-loop fit into the overall system? Throughout the session, we’ll share practical design patterns and implementation insights from building real-world agent systems with Google's Agent Development Kit (ADK). Whether you’re building your first AI application or designing complex agentic workflows, you’ll leave with a practical framework for thinking about multi-agent system design—not just how to build agents, but how to architect them.​🎙️ The Speaker Lineup: • ANNIE WANG — Software Engineer & Internal Career Coach, Google Software Engineer with experience across startups and big tech. Currently working as a Software Engineer and Career Coach at Google, Annie brings years of engineering experience to empowering developers and building scalable technical systems. • Moderated by JING CONAN WANG — Founder & CEO, DeepVista AI, Co-Founder of FounderCoHo 🛠️ What we'll cover: System Decomposition & Patterns: How to break complex product requirements into specialized sub-agents using sequential, parallel, and custom workflow patterns.Deterministic vs. LLM Tasks: Knowing when to lean on an LLM for reasoning versus when to route tasks through deterministic code to save latency and token budget.Memory & State Architecture: Determining where session state and long-term memory should live, and when memory is actually necessary.Testing & Trajectory Observability: Moving beyond output "vibes" to evaluate agent trajectories, tool-calling accuracy, and human-in-the-loop safeguards. 📅 Event Agenda: • 6:00 PM - 6:10 PM: Stream Warm-Up & Virtual Intros • 6:10 PM - 7:00 PM: Architecture Deep-Dive & Interactive Q&A​🎟️ Registration: Virtual Livestream Meet your Host! FounderCoHo is a community and media platform for founders to share their hard-earned wisdom and connect with fellow founders on their journey. With 208,000+ subscribers across YouTube and Substack, we've featured industry leaders, including Lyft's Co-Founder, the inventor of CRISPR, and AlphaChip's co-author, alongside many successful founders and senior executives. We host monthly in-person founder meetups where members exchange insights and build meaningful connections. Explore more: foundercoho.com/ Where to find us: LinkedIn YouTube Substack