Keynote: Boring Wins in AI Agents 🔥 Hard-Earned Lessons
summary
Agents are plain software: LLMs are prediction machines, tools are JSON contracts, and context is the only memory that matters — tokens cost money so summarize and compact aggressively. Follow microservices practices: single responsibility, versioning, CI/CD, deploy binaries or containers, and avoid single points of failure (example: Azure East US 2 outage broke models). Don’t adopt every protocol or chase GPT‑5.5 day‑one; start with REST and measure token costs and degradation.
AI-written summary · about
chapters (7)
- 0:00AI agent foundations
- 3:11Microservices for agent design
- 8:28Avoid shiny object syndrome
- 15:31No need for MCP or complex protocols
- 21:57Context engineering for costs
- 26:38Learn-it-all mindset
- 29:05Summary
related talks
59mKeynote: AI without the BS, for humans
10mKeynote: The Last Mile Problem: Why AI W...
24mKeynote - The AI Playbook For Companies
32mKeynote: Will the Real Autonomous Agent Please Stand Up
54mKeynote — Noise and Signal
19mKeynote: Towards Building Safe and Secure Agentic AI
30mKeynote: Fireside Chat
01h 01mKeynote Panel: Future of AI agents