AI agents for system design

4 topics.

• 01Mid-levelReadAgents and tool useA loop of think, call a tool, read the result: simple to write, hard to keep on track.
• 02SeniorReadAgent orchestrationChains, routers, orchestrator and workers: when one agent beats many, and how to contain a failing one.
• 03SeniorReadAgent deployment modelsWhere the loop runs, how it survives a crash, and whose credentials it acts with.
• 04Mid-levelReadModel Context Protocol (MCP)One protocol between any agent and any tool, and the new security holes that come with it.

More in AI systems

• Foundations
• Build it yourself
• Problems
• AI systems overview