Own your agent
summary
Argues that running agents on open weights avoids data exfiltration and vendor lock-in, shows how to build a stack with quantization, fine-tuning, embeddings and observability, and gives concrete resource targets: frontier models grow from 100B to 1–10T parameters, open source lags ~three months, and break even for self‑hosting vs API sits around 100 million tokens a day. Recommends specific pieces: GLM 5.3 flash on Nvidia B300/5090 with NVFP4, or Qwen 3.8 27B for single‑GPU dev setups.
AI-written summary · about
chapters (9)
- 0:00Owning your AI agent
- 3:26AI agent definition and components
- 5:11Open source vs frontier models
- 7:21Resource accounting for models
- 9:12API-based AI models pros and cons
- 12:50Risks of repetitive agents
- 14:39Security risks with third-party agents
- 16:07Building an agent open source
- 19:19Summary and call to action
related talks
29mRunning AI Agents Inside TEEs Without Losing Your Mind
45mRunning Your Coding Agent Locally: Lessons from a Real-World Experiment
47mOwn Your LLM: Running Generative and Agentic AI Locally
42mData Challenges for Scaling Multi-Agent GenAI Apps with Python
06mThe Building Blocks of Agentic Al
31mBringing and Running AI Agents in the Browser
01h 42mBuilding Responsible AI Agents in Decentralized Open-Source Environments
35mUsing coding agents with open models