DevTalks: Designing Effective AI Agents | Ep 1

Description

Learn how to build real AI agents using CrewAI and SambaNova’s lightning-fast open-source models. In this DevTalks session, Justin from SambaNova and Shane from CrewAI walk through how agentic workflows work, why speed matters, and a full live demo of building multi-agent automation from scratch. You’ll see: • How CrewAI creates multi-agent workflows automatically • SambaNova’s 10× faster inference compared to GPUs • How reasoning models like DeepSeek v3.1 perform in real-time • How to build agents that research, analyze, collaborate & take actions • How to pick the right LLM for each agent task • Real examples of building custom tools, triggers, MCP integrations • A full walkthrough of the CrewAI visual builder + trace viewer Whether you're building copilots, assistants, or fully autonomous business AI agents — this session gives you a practical blueprint. https://sambanova.ai/ https://www.crewai.com/ 🔧 Tools & Technologies Featured CrewAI agent orchestration framework SambaNova RDU-powered inference platform DeepSeek v3.1 / R1 models Multi-agent workflows Custom tool creation + MCP server integration Visual agent builder + tracing 00:00 – Livestream intro & welcome 01:00 – What DevTalks is about 02:20 – Global audience greetings 03:00 – Agenda overview 04:00 – Who is Justin? (SambaNova Dev Advocate) 05:10 – What is SambaNova? (RDU chips + dataflow architecture) 07:00 – Why speed matters for agentic AI 08:20 – Demo: SambaNova vs ChatGPT speed comparison 11:45 – Running the generated Space Invaders code 13:15 – Introducing Shane and CrewAI 14:00 – Evolution from chatbots → copilots → agents 15:20 – Why multi-agent workflows matter 17:00 – How organizations deploy agents 18:30 – Scaling, transparency, optimization 20:00 – CrewAI Visual Builder overview 21:10 – Real-world customer case study (General Assembly) 23:00 – Building an AI agent workflow live 24:00 – Adding tasks + agent roles 25:20 – Running the workflow with a real example 26:40 – Understanding agent traces 28:00 – Switching models (cost vs speed vs reasoning) 29:30 – Tool integrations & MCP 31:00 – Final thoughts + Q&A #AIagents #CrewAI #SambaNova #OpenSourceAI #LLM #DeepSeek #AIworkflow #AgenticAI #AItutorial #DevTalks #MachineLearning #AIautomation