Hau Chu | Solo Builder

Go engineer. I run my solo business on 12+ AI agents: job hunting, inbox, Notion, trading research. Real setups, real numbers, incl. losses. Vietnam 🇻🇳

[Hau Chu | Solo Builder @hnhcsolo]

Black Friday on Shopify: 5 boring checks that save headaches 🧵

[Hau Chu | Solo Builder @hnhcsolo]

Where it fell short: on day 3 Fiverr showed the agent a press-and-hold captcha, so it couldn't open the inbox and had to judge from the notification email alone.

It called it right, but blind. Runs on @bot.

[Hau Chu | Solo Builder @hnhcsolo]

I opened two Fiverr gigs and handed the inbox to an AI agent on day 1.

First 3 days: 4 "buyer" messages. All 4 were scams. 0 real orders, $0.

The pattern it learned to flag 🧵

[Hau Chu | Solo Builder @hnhcsolo]

The rule now: if a message claims an order I can't see, pushes a file or outside link, or comes from a brand-new account, the agent doesn't reply, doesn't click, and doesn't ping me.

Real buyers get a drafted reply that I approve first.

[Hau Chu | Solo Builder @hnhcsolo]

The failure so far: the agent nailed consistency. It runs on schedule and follows every rule.

It hasn't proven my offer wins work. That part is still on me.

Running on @bot.

[Hau Chu | Solo Builder @hnhcsolo]

I gave an AI agent my Upwork job hunt.

8 days in: 6 proposals, all sent by the agent. 0 views, 0 offers, $0 earned.

Here's the exact rulebook it runs on 👇

[Hau Chu | Solo Builder @hnhcsolo]

62 logged runs since 19 Oct: check the inbox, search, then send or skip, with a written reason for every skip.

It hit the 3-a-day cap on 21 Oct. Since then it has sent nothing: the jobs that fit cost more Connects than its reserve allows.

[Hau Chu | Solo Builder @hnhcsolo]

@hit100os That's the plan for the next one. Ten real people looking at it before I build anything, and nothing gets built if they all shrug.

[Hau Chu | Solo Builder @hnhcsolo]

My AI agents ran a 7-day pre-sell: 3 product ideas, 6 Gumroad pages ($29, $79, $99).

Result:
0 sales
$0
0 waitlist emails
0 sample downloads

Here's the full failure log 🧵

[Hau Chu | Solo Builder @hnhcsolo]

@alia0209 tail exiting 0 while the suite fails is a nasty one. My agent computer once rebooted and lost its environment, so my startup script now reinstalls whatever tools are missing.

[Hau Chu | Solo Builder @hnhcsolo]

Takeaway: a fix can only shrink the loss on a fill that's already bad.

The real problem is which orders get filled, not what you do afterwards.

Paper simulation, not financial advice.

[Hau Chu | Solo Builder @hnhcsolo]

At 98¢ the fixes did help: instant fix −$19.70 vs −$28.25 for doing nothing.

But "instant" assumes zero latency, and you never get that live.

Sample: 40 markets, 10 shares a side. Too small to call it settled.

[Hau Chu | Solo Builder @hnhcsolo]

Bids summing to 94¢, 40 markets:

Do nothing: −$17.06
Hedge at T−60s: −$18.00
Exit when filled side drops 2¢: −$18.94
Fix after 10s / 15s / 20s: −$22.59 / −$25.61 / −$24.37
Fix instantly: −$23.50

Doing nothing lost the least.

[Hau Chu | Solo Builder @hnhcsolo]

Only one side of my two-sided Polymarket order kept filling.

So my agent tested 6 fixes on the same 40 markets.

None of them made it profitable. Results 👇

[Hau Chu | Solo Builder @hnhcsolo]

@MbaziiraRonn True. Then I'd split it behind a flag so the rollback is still one switch, even if the change itself is big.

[Hau Chu | Solo Builder @hnhcsolo]

What I'm changing:

Pick the channel first. Don't build the offer until about 10 real target buyers have actually looked at it.

Project B is killed, C is parked, and the Gumroad dashboard still says "You haven't made any sales yet."

No "I made $X" posts until there's real money to show.

[Hau Chu | Solo Builder @hnhcsolo]

Why it failed, honestly:

My X account has 19 followers. Across all 3 projects the posts got 182 impressions in 7 days (99, 32 and 51). The one post with a link was seen by 7 people.

I never tested demand. I tested whether a tiny account can reach buyers. It can't.

[Hau Chu | Solo Builder @hnhcsolo]

@tirupathipodugu Exactly. The 27 markets where only one leg filled lost $82.40 (3 won), while the 58 two-leg markets made +$34.80. Every fix I tested after that lost $17 to $26 over 40 markets, so I'm not going live. Next post breaks those fixes down.

[Hau Chu | Solo Builder @hnhcsolo]

My AI agent paper-traded Polymarket's BTC 5-minute markets for 66.7 hours.

Every config lost money.
Best one: −$42.84 over 86 markets, on $19 of paper capital.

Here's where the money went 🧵