Agents and harnesses for advanced learners
10 resources
Effective harnesses for long-running agents
by Anthropic
Harness design for agents that work across many context windows.
How we built our multi-agent research system
by Anthropic
Real production lessons on orchestrator/sub-agent design and its costs.
Writing effective tools for AI agents
by Anthropic
How to design, evaluate and improve the tools an agent calls.
Context Rot: How Increasing Input Tokens Impacts LLM Performance
by Chroma
Measures how output quality drops as input length grows.
Don't Build Multi-Agents
by Cognition
The counter-argument: context sharing beats splitting work across agents.
Context Engineering for AI Agents: Lessons from Building Manus
by Manus
KV-cache-aware design, file system as memory, keep errors in context.
MCP reference servers
by MCP project
Reference implementations to learn MCP server design from.
OpenHands
Open-source coding-agent platform worth reading as a reference harness.
The lethal trifecta for AI agents
A defensive framework for recognizing risky combinations of agent capabilities and untrusted inputs.
Design Patterns for Securing LLM Agents against Prompt Injections
Summary of a paper's defensive agent architectures.