Delvin Chang

Taiwan · Email

One person.A company of agents.

I build teams of AI agents that do real business work from one home PC.

Agents I run handle a manufacturer's export inquiries, a world-news account and the data checks behind my market reports.

  • Marketing Consultant, Kingconn
  • AI Engineer (Intern), 9chapter algorithm
    In-house AI consultant role
LinkedIn

01Kingconn

A connector maker's website and export inquiries, handled by agents I direct.

Kingconn has made connectors in Taiwan for 24 years. I took over its website and overseas sales inquiries and handed the work to AI agents. First month after the switch:

Genuine purchase inquiries a month~3→18
Countries they came from10
Best-known inquirerMicron
Commits I wrote myself35 of 723Agents wrote the rest.

The move that worked best: when a big brand discontinues a part, engineers search the old part number to find a replacement. So I had agents publish replacements for 39 discontinued parts, on a site that now runs in four languages instead of one. Two Italian companies asked for quotes on 3,000 pieces within 48 hours of those pages going live.

kingconn.com.tw
02World news on Threads

A world-news account my agents write, fact-check and post.

I switched @marketdailyhq to world news in mid-September. Views in the two weeks either side of the switch:

Views, two weeks30,889→1.68M
Biggest single post450K
threads.net/@marketdailyhq
03The company of agents

It works while I'm away.

Before

Every task waited for me to sit down.

After

781work cycles ran on their own over 7 weeks while I was away from the keyboard.

Anything that goes public is checked by a separate AI reviewer first. Behind it, 206 scheduled jobs send agents through 155 skills written like job descriptions, and more than a fifth of the skills keep a log of their own mistakes that loads every time they run.

04Also running

Also running

  1. Data-source watchdog

    Telling a dead source from a quiet day

    A checker for the 22 public data sources behind my market reports.

    Before

    A dead source looked exactly like a quiet day.

    Caught on its first runs

    A thirdof Taipei Exchange requests failing, silently dropping a batch of company financials.

  2. Motion pipelineMy agents build and render After Effects and Blender animations in the background, so a brand film gets made without anyone opening the apps.

  3. Strategy checker

    It caught my own stock screen

    A tool that tells whether a trading strategy's edge is real or luck. Its first catch was my own.

    Looked like

    64%win rate

    Checked against random entries

    No proven edgeRandom entries in the same stocks won 55.6%. The difference wasn't statistically significant.