Скачать или смотреть AI Agents Fail 75% of the Time!? | Zuplo MCP Week feat. Zdenek Nemec

AI agents fail 75% of the time on simple tasks. Zdenek Nemec, CTO of Superface and API expert, explains why.

We cover the technical reality behind AI agent reliability: why MCP probably isn't enough, the 10-tool performance limit, specialist vs general agents, and what actually works in production.

Essential viewing for developers building AI agent integrations.

Join us in Discord to talk all things MCP: https://zuplo.link/mcp-discord

Zuplo: https://zuplo.link/check-us-out
Superface: https://superface.ai

0:00 Introduction
2:14 What is Agentic AI
4:19 MCP: Anthropic's Game-Changing Move
7:27 Why Connecting APIs to LLMs Isn't Simple
11:25 The Dream of Agentic Tool Discovery
13:05 The 75% Failure Rate Reality Check
16:39 Agent Failures: Planning & Context
18:52 Engineering Solutions Across the AI Stack
20:50 Specialist Agents vs General Purpose Agents
23:39 API Design Best Practices for AI Consumption
25:22 The 10-Tool Rule
27:23 Documentation Format: Does It Really Matter?
32:13 The Valley of Disillusionment: What's Next?
34:23 Wrap-up & Where to Find Superface

#AIAgents #MCP #APIs #Engineering #ai