How I actually use AI tools in engineering work
Not a productivity take. A look at where LLMs are genuinely useful in systems work, where they're not, and what that tells me about building better agent tooling.
刘子阳
Software engineer. Worked on VM migration, guest tooling, and kernel debugging at SmartX.
Now spending time on AI tooling and agent reliability — starting with Windows desktop automation.
education
UESTC
B.Eng. Software Engineering
2020 – 2024
stack
experience
Systems Integration Development Intern
Jun 2022 – Sep 2022 · Chengdu
products
Lark Suiteprojects
Research fork of CursorTouch/Windows-MCP. The focus is reliability: an agent uses the MCP server, patches it, hot-reloads it, reruns the same workflow, and checks whether the real desktop behavior improved. Inspired by karpathy/autoresearch.
Terminal app for word learning. Stores cards and review state in SQLite with an FSRS-style scheduler. Uses an LLM for character-style teaching and explanation. Supports swappable companion packs.
writing
all posts →Not a productivity take. A look at where LLMs are genuinely useful in systems work, where they're not, and what that tells me about building better agent tooling.
Live migration sounds clean in the docs. The reality involves dirty page tracking, guest agent timeouts, and failures that only happen under load.
A VM went unresponsive. The hypervisor thought it was fine. The guest disagreed. Tracing the failure across three layers to find the actual cause.
contact
Available for backend, infrastructure, and AI tooling roles. Reach out by email or connect on LinkedIn.