AI Agents • AI-native Engineer • Building agent systems

I build AI-native systems where agents can plan, use tools, and iterate toward real outcomes. My focus is designing the agent harness — the infrastructure that lets AI systems reliably do work.

Repositories · Stars · Followers

• Tool-using agents — browse / code / data / APIs
• Multi-step workflows — plan → execute → reflect → improve
• Long-running agents — tasks that operate autonomously
• Applied automation — agents that do real work

Interested in:
agent evals • tool routing • RAG + tools • planning loops • autonomous workflows

Building software where AI is part of the runtime, not just a feature.

• Traditional stack： Human → Code → Software
• AI-assisted stack： Human → Prompt → AI → Code → Software
• AI-native stack ：Human → Intent → AI Agents → Tools → Outcome

In this model, the software is the agent system.

If LLMs are the brains, the agent harness is the operating system.

Core components:

• agent orchestration
• tool routing
• memory systems
• evaluation loops
• observability
• reliability (guardrails, retries, fallbacks)
• cost & performance control

Goal: make agents predictable, reliable, and useful in production.

• agent harness architecture
• long-running autonomous workflows
• tool-using coding agents
• AI-native developer tools
• knowledge systems for agents

• Software is shifting from deterministic programs to probabilistic agent systems
• The developer role is shifting from writing code to designing intelligence systems.

GitHub
https://github.com/sharkhuang

_{Last updated: 2026}