TheCoderBTW

Indie dev/farmer
Building Acquit, Acan with @droid
@FactoryAI Guild Member
pvstack-

[TheCoderBTW @TheCoderBtw]

Day 4 on the @droid Max plan ($200).

@FactoryAI sent a free voucher. I used Lauren Tan's (@poteto) pstack in Cursor for months, and I missed it when I moved to Droid. So I ported it.

The part I care about most is the model routing. Each role in pstack gets its own model, and I picked them from Morgan Linton's (@morganlinton) VulcanBench (@VulcanBench ) results instead of going on vibes.

Coding goes to GPT-6.1 Sol, which scored 97% of Opus 5.5's best for about a tenth of the cost. Opus handles the judgment calls, and the review panels mix models from three different labs.

There's also a budget mode that hands the grunt work to DeepSeek-V4.1-Flash

It checks Lauren's repo every week, so her updates keep coming.

https://t.co/3yyU1Wagff

[TheCoderBTW @TheCoderBtw]

@Da7_Tech Now imagine what the internal versions look like

[TheCoderBTW @TheCoderBtw]

@Da7_Tech What do you think is the reason for that?

[TheCoderBTW @TheCoderBtw]

@adityawaslost My pick is the 1st one. It's clean and compact

[TheCoderBTW @TheCoderBtw]

Okay tomorrow’s the day. Meta Global AI Developer Hackathon starts at 12 PM EST (that’s 10:30 PM for me).

As promised, I’m taking you guys with me on this one. I’ll post every couple hours about how Acan is actually evolving, the decisions I’m making in real time, the problems that show up, and how I try to fix them.

Pausing the daily Droid/Acquit series for a bit, but Droid’s still running the show on Acan. Acquit is roughly 85% done for the PayPal hackathon though, so the deadlines are gonna overlap and you’ll get to watch me juggle both irl😅

Let’s see how this goes.

[TheCoderBTW @TheCoderBtw]

@morganlinton How I missed this😭

[TheCoderBTW @TheCoderBtw]

@VulcanBench 🫡🖖

[TheCoderBTW @TheCoderBtw]

@morganlinton @VulcanBench Congratulations Morgan, well deserved🖖

[TheCoderBTW @TheCoderBtw]

Day 22 on the @droid Max plan ($200)

Another big day on Acquit. I ended up deleting the demo mode completely.

I had a separate scripted demo for the judges. But they're going to use it like real people anyway, so now that's just how it works. One click and you get an account that sticks around, and you can switch between client and operator on the same account.

Also started a full UI redo. Upwork-style layout but built for agents, light mode by default. Had Opus write a UX spec first so every worker builds from the same page.

4 PRs merged and deployed today. Next up are House clients that post starter jobs, so a new operator has something to actually win right away.

Still all PayPal sandbox. Code freeze is Nov 6, 13 days out, so lowkey racing the clock now 😅

[TheCoderBTW @TheCoderBtw]

@annbordetsky 👀🙌

[TheCoderBTW @TheCoderBtw]

@Badiaserra @FactoryAI @francesca_lab @sarahmalek__ @annbordetsky @matanSF this is exactly the kind of environment I’d love to be part of

[TheCoderBTW @TheCoderBtw]

@prathamdby @droid @FactoryAI appreciate it bro, I tried your approach for this one🤝

[TheCoderBTW @TheCoderBtw]

Day 21 on the @droid Max plan ($200)

Pretty big day on Acquit. 6 PRs merged, and most of it was about disputes.

So I filed a real buyer dispute on the PayPal sandbox against one of my own paid jobs ($21). Wanted to see what happens when a client says "I never got this".

Turns out PayPal refused Acquit's first answer. Acquit sent its verified receipt as proof of fulfillment, but at that stage PayPal only accepts proof of refund or "other". And Acquit never refunded anything, so claiming that would be a lie. Now it sends the type PayPal asks for, and PayPal accepted the evidence. The dispute is under review and the money stays held in escrow the whole time.

The job page now shows the hold and PayPal's side in plain words, plus a history of every answer Acquit filed. The review model failed that one twice before it passed 😅

The other half was 3D printing. A simulated print farm has to qualify first: 5 test parts plus a trap job it should turn down. Then the real print judge checks its print. Each layer shows the contract shape next to the farm's (simulated) photo, so you can see the cheat it caught.

Still all sandbox, and print doesn't move money yet. Next up is screenshots for the video.

[TheCoderBTW @TheCoderBtw]

@Davidwuuu92 Yeah, opus + haiku. This is soo Claude coded ngl 😂

[TheCoderBTW @TheCoderBtw]

This is the ultimate AI slop website I've ever seen

[TheCoderBTW @TheCoderBtw]

Day 18 on the @droid Max plan ($200)

Redesigned the whole Acquit landing page today and it's live now. Quick run through in the video.

Did it all with Droid. Opus 5.5 ran the day as the orchestrator and 27 Haiku 5.5 subagents did most of the building, split across the sections. My favorite bit is the verifier machine. You drag a submission into the slot and watch it get rejected or paid.

The split worked way better than I expected. Opus made the design calls and checked every subagent's work itself, mostly in a headless browser. The Haiku agents cost about 11% of the credits. Around 21M credits for the whole day.

And it wasn't magic btw. The Haiku agent that moved the page into the app flagged that its own port broke the dashboard for signed-in users. Opus fixed it before merge. Then PR, CI, a database backup and the deploy.

Built it as a scratch mockup first and only ported it once I liked it. lowkey think that's why it came out clean.

The live demo is locked to my IP for now while I finish the coding part. Still PayPal sandbox, no real money moves.

https://t.co/oAL9c3SgAA

What do you think of the verifier machine? 👀

[TheCoderBTW @TheCoderBtw]

@0xxmemo @droid @FactoryAI None. Droidie do it for me

[TheCoderBTW @TheCoderBtw]

Day 20 on the @droid Max plan ($200)

14 PRs merged on Acquit today. All Droid, all orchestrated.

The day started at PR #115 and ended at #128. Mobile UI got a full pass, the verdict grid actually fits a phone now. Chat cards show PayPal orders and payouts inline. The escrow dashboard has undo receipts with a change log.

The big one was disputes. Droid set up a real sandbox buyer dispute through PayPal checkout, funded a job, let the House deliver and get verified, then filed the dispute. Found that PayPal returns a 403 on some dispute reads, so we built a hold path for that. House operators now answer their own buyer disputes with their own PayPal app.

Also started on the 3D print verifier. A simulated print farm gets qualified, a judge scores the output, and missing photos fail-close. The validator found three real bugs on the first pass, including one where CPU work blocked the API event loop. Fixed those in round 2.

21 commits across all of it. Ran out of usage at the end but honestly that's a solid day. Still sandbox, still building.

[TheCoderBTW @TheCoderBtw]

@VanillaCache yeahh, already seeing it happen around us

[TheCoderBTW @TheCoderBtw]

This....

[Nikita Bier @nikitabier]

Perhaps the threat was never a rogue AI agent trying to subvert humans—but humans using AI to hurt other humans.

[TheCoderBTW @TheCoderBtw]

@DanielZambrini @droid @FactoryAI Droidie reviews them for me

[Tim Pan @mediastorm_en]

My iPhone Duo video went viral, so I'm giving away 100 iPhones on X 😛

[TheCoderBTW @TheCoderBtw]

@solismommy @droid @FactoryAI I got really scared thinking I lost all my work. Gonna keep a local backup from now on

[TheCoderBTW @TheCoderBtw]

Day 19 on the @droid Max plan ($200)

Ran out of compute hours on my Droid cloud computer this month. And all my Acan sessions were sitting on it, 5 days before the Meta AI hackathon 😅

The project files were safe on GitHub. The session history wasn't, and that's where all the research and prototype rounds actually live.

So I opened Droid on my Windows machine and asked it to recover everything. It first tried to SSH into the computer and got a "monthly compute usage limit" error. Fair enough.

Then it figured out Factory keeps session history on their side, and you can read it through the Sessions API without waking the computer. I made an API key and it pulled 65 sessions, about 17k messages. Even the old ones from back when the project was still called Orrery.

And it didn't just dump them. It converted them into local Droid sessions in my Acan folder, so I can open any of them and keep going. It also rebuilt a few files my agents wrote after my last commit, straight from their tool calls.

Took about 40 minutes. lowkey thought I'd lost a week of work. Droid finally saved the day.

Anyone else backing up their agent sessions or is it just me now?