Explore what technology makes possible. Build something useful with it.
I follow how new technology changes the way we work, build tools around the questions it raises, and share the ideas, trade-offs, and products that come out of the process.
From quantitative trading in Taiwan equities to AI agents, I care about how an interesting idea becomes something people can use. This is where I share my open-source work and my thinking about AI, products, and software that keeps evolving.
Explore my website · 繁體中文 · Taipei, Taiwan
- AI that can carry work forward. Useful agents need to retain progress, work with tools, and let people understand and steer what happens.
- What faster development leaves behind. Reuse, dependencies, and the life of each version shape how far the next iteration can go.
- The experience around the capability. Good tools fit the way people work and make complex systems easier to understand and control.
Keep AI-built projects worth building on. A Codex skill that brings modular reuse, rebuildable environments, and deliberate retirement into the development workflow, so useful work has a clearer path to grow.
Give an agent a way to continue the work. A self-hosted Python/Linux runtime exploring durable tasks, conversation continuity, and traceable delivery across Telegram, Discord, and CLI.
Pre-release source preview; validation gaps remain.
A workspace where agents can get hands-on. Persistent Python/IPython workspaces, bounded RLM jobs, and durable operation records, with approvals and external actions kept under host control.
Public alpha; not a security sandbox.
Move from a vague goal to a result you can check. Focused skills for specification, implementation, debugging, review, and learning from failures, with verification that fits the risk and scope of the work.
For perspectives, projects, and the story connecting them, visit phenomenoner.github.io.
Interested in how AI can become part of useful work, or have a tool or product idea to compare notes on? Get in touch.