I have always liked tinkering with software.
Not because I want to chase every new tool, but because I keep returning to one question: can something that requires repeated manual work be reorganized into a dependable system that keeps running?
In the past, that question often ended with a prototype. Development, maintenance, and integration were expensive, and one person could rarely carry every part to completion. AI agents have moved that boundary. They have not removed judgment or maintenance, but they have dramatically amplified one person’s ability to turn an idea into software.
An abstract view of manual tasks becoming reliable automation; generated for this site
What I am exploring
I currently serve as Chief Marketing Officer at UChat（在新标签页中打开） (opens in a new tab), while building FengTalk.ai（在新标签页中打开） (opens in a new tab). The work takes different forms, but it keeps circling a connected set of questions:
- How web coding can turn an idea into a useful product faster
- How AI agents can understand a goal and keep working within clear boundaries
- How outreach can become an observable, improvable system instead of scattered activity
- How company knowledge can become context that both teams and agents can use
- How automation can take repetitive work while leaving consequential judgment to people
I care about more than whether something runs once. Reliability, clarity, recovery, and fit with the actual business usually matter more than the first demo.
Why write it down
Much of the conversation around AI stops at product launches, feature lists, and temporary novelty. I want to document what happens while building: the choices, the failed approaches, the validation, and the reasons the current version survived.
This site will focus on web coding, AI agents, automation, outreach, international business, and company knowledge. Sometimes that will mean a complete tutorial. Sometimes it will be a product retrospective or a small observation that deserves a longer test.
Starting here
The first version of this site is itself a public experiment: preserve the complete design and interaction of a mature personal site while rebuilding its identity, content, deployment, and maintenance workflow, then make it safe for an AI agent to help publish future writing.
I hope this becomes more than a collection of posts. I want it to become a durable record of the work. Time will tell which ideas were genuinely useful.