wang

Building AI workstations for fun and margin. Playbook ⬇️

[wang @Chinaweop]

Verdict day.

13 days: 9 products, 5 deep posts, 20 automation scripts.

Store: 92 visits. 0 sales. 0 discount uses.

Every channel hit a wall: Reddit ban, X rate-limit, HN flags, IH gate, https://t.co/Hkfh6AWWJT silence.

The machine worked. The market didn't care.

[wang @Chinaweop]

Day 11 building in public.

Wrote two deep-dive posts this week. The internet said: thanks, no comments, a few dozen views.

48 hours left on my self-imposed verdict clock. If nothing sells, the honest post-mortem will be free.

https://t.co/j69q98K8b3

[wang @Chinaweop]

New deep-dive: automating any website login with CDP - magic links, email codes, OAuth-only platforms, and React forms that eat your keystrokes. Real code.

https://t.co/8Z3rPDtH92

[wang @Chinaweop]

Day 9 building in public.

The machine wrote a full playbook post yesterday. Store traffic: double digits. Sales: 0.

Hardest truth: automation multiplies whatever exists. It multiplied zero.

https://t.co/j69q98K8b3

[wang @Chinaweop]

New post: the full playbook of my self-hosted marketing agent. It reads my inbox for login codes and posts while I sleep. Every failure included.

https://t.co/L5qPznyBxB

[wang @Chinaweop]

Day 8 building in public: the one-week verdict.

74 store views, 0 sales. Mostly direct traffic.

Not a copy problem - a traffic problem. So: more channels, not more products.

https://t.co/j69q98K8b3

[wang @Chinaweop]

Day 5 building in public: honest numbers.

Yesterday's posts: 3 views, 2 views. Not a typo.

Sales: 0. But discount codes are live, a data essay is published, and I replied to 3 founders building alongside me.

Keep publishing anyway.

https://t.co/j69q98K8b3

[wang @Chinaweop]

@kreadony Welcome to the grind. One thing that helped me: pick a single metric that matters this month (for me it is store visitors) and post about it every day, even when it is zero. The streak compounds faster than any single post. Good luck!

[wang @Chinaweop]

@Bella_09188 Also building in public right now - 9 digital products live on Gumroad, day 4, zero sales. The honest part nobody posts: early numbers are mostly zero, and the wins come from showing up daily anyway. Following your journey - good luck!

[wang @Chinaweop]

@shubhranshu2009 Day 1 advice from someone on day 4: log where every visitor comes from, one line each. Day-one traffic is mostly luck - easy to mistake a spike for a channel. And pick ONE thing that keeps working while you sleep. Good luck with https://t.co/3xCBFYwi42

[wang @Chinaweop]

The bug that nearly killed my automation project: every request hung, DNS resolved to 198.18.x.x. I debugged my code for 40 min.

Cause: the network path. Fix: --socks5-hostname resolves DNS at the tunnel's far end.

Check the tunnel before the code.

[wang @Chinaweop]

Day 4 building in public: honest numbers.

6 posts, 1 follower, two-digit store traffic. Sales: 0.

So I built 50%-off codes and drafted launch posts instead of polishing posts nobody sees.

The funnel isn't built yet. Building it in public.

https://t.co/j69q98K8b3

[wang @Chinaweop]

Day 2 building in public: 2 new products + a $59 bundle (worth $120) now live.

Data from 146k Gumroad products: $30-49 converts 28% better than under-$10. Cheap isn't a strategy.

https://t.co/nsvWGnOQUt

[wang @Chinaweop]

Day 2 building in public: honest numbers.

Build thread yesterday: 6 views, 0 sales. 6 products live on Gumroad.

Day 2 of a store is like a Day 12 pipeline — looks successful, produces empty files. Fix is the same: assert, log, keep publishing.

[wang @Chinaweop]

Full build is now a $19 playbook → https://t.co/j69q98K8b3

[wang @Chinaweop]

I built an AI workstation from scratch last month.Not a chatbot-infrastructure:agent +self+hosted data +proxy+pipelines,It runs while I sleep .Here's the full stack ,piece by piece

[wang @Chinaweop]

Hardest trap: Windows garbles UTF-8 in API calls. Cost me 2 hours. Fix: curl.exe + raw bytes. Save yourself.

[wang @Chinaweop]

agent harness. Model decides,your code executes,Tools must be deter ministic.Never let an LLM"run"anything you can't unit test