I’m a software engineer in Seattle building AI agents, developer tools, and observability systems. Previously at Microsoft, I was a founding technical lead for agent evaluation and observability in Visual Studio Code and GitHub Copilot. My interests span how agents work, how we evaluate them, and how people understand and improve their behavior.

My public contributions from Microsoft are under @zhichli, and my personal projects are under @b1tank.

I built VS Code’s evaluation harness and co-authored VSC-Bench, a benchmark with more than 100 cases, supporting dozens of partner teams. I also led the analytics work that turned agent trajectories and evaluation results into useful feedback for model providers, product engineers, and leadership. I worked closely with model providers, including Microsoft AI, OpenAI, and Anthropic, on model evaluations. This work connected model launches and everyday product changes to improvements in prompts, tools, quality, and efficiency.

I designed and implemented OpenTelemetry observability in VS Code’s agent core, making agent behavior easier to investigate during evaluation and monitor in enterprise environments. I worked across Copilot CLI/SDK and evaluation teams to align instrumentation, led the telemetry migration to the new agent architecture, and partnered with monitoring teams on integrations and dashboards.

Before VS Code, I worked in GitHub Copilot Applied Research as a founding engineer on AutoDev, building agent infrastructure and containerized evaluations alongside researchers. I also built a post-training data pipeline that contributed to a Microsoft AI model launch. Earlier, I was a founding engineer and technical lead for Azure Managed Grafana and built multi-tenant log pipelines for Azure’s observability partner integrations, including Datadog. That background still shapes how I approach reliability, performance, and developer experience.

My path into software began in scientific computing during my PhD in chemical engineering, where I built machine learning models in MATLAB and aerosol models in Fortran and C++. Those experiences shaped my appreciation for performance, correctness, and clarity.

Having worked in both academia and the AI and software industries, I believe:

“AI can save us work, but it can also distance us from the learning that work provides. Curiosity is fundamental to human nature. We want to learn more, not less - and AI should be wielded to advance education, not deprive us of opportunities to learn.”

That is why I co-founded LeZhiWeng.com around my dad’s decades of mathematics teaching, using agent-assisted publishing to make his material easier to organize and share. I also build learning projects for myself to catch up on foundational technologies that still matter.

I also enjoy basketball 🏀 and billiards 🎱. I’m a Stephen Curry and Golden State Warriors fan, and I admire Curry’s humility, work ethic, and competitive drive.

My public collection of reusable agent skills is available at b1tank/skills. Other tools I have built are shown below.