How to build Agents that work

Generally, I try to use off the shelf agents instead of ripping my own.

Agents are LLMs + 24/7 access + tool calls.

1. how i think about building agents
1. base
1. agentic harness
1. https://github.com/ultrmgns/claude-private
2. https://blog.langchain.com/the-anatomy-of-an-agent-harness/
3. https://sausheong.com/the-context-and-the-harness-c18156077de4
4. https://www.anthropic.com/engineering/effective-harnesses-for-long-running-agents
5. https://www.markdown.engineering/learn-claude-code/
2. long-running
3. verifivable / labeled / eval
2. medium
1. knowledge base
1. https://x.com/jumperz/status/2039826228224430323?s=20
2. https://x.com/alex_prompter/status/2039853870810108384?s=20
3. advanced
1. self-improvement
1. https://kyrylai.com/2026/03/30/self-improving-agent/
2. auto-improvement
1. https://hugobowne.substack.com/p/building-agents-that-build-themselves
2. https://github.com/neosigmaai/auto-harness
3. https://github.com/kevinrgu/autoagent/tree/main