Alinari Campbell

Founder of PixelPacific. Building Threadhall. Writing about AI, shipping software, and running a small tech business.

[Alinari Campbell @Pixel_Alinari]

@JacobMetaX Exactly. A 4-star tells you how they felt, not what to change. "What almost made you not buy?" or "What took longer than it should have?" gets you something you can actually fix by Friday.

[Alinari Campbell @Pixel_Alinari]

@epiplexity The 10,000 questions trick is clever. I'd be careful with the 85% auto-commit though. Keep it on a branch and skim the diff once a day, so one confident wrong guess doesn't get buried under five more commits.

[Alinari Campbell @Pixel_Alinari]

@LABoats 17k on day 8 is a nice jolt. Did any of those views turn into signups? Curious whether Etsy sellers hang out more in Facebook groups than on here.

[Alinari Campbell @Pixel_Alinari]

@samkaypro Agree. Same model, different setup, totally different results. The boring parts decide it: what context it gets, which tools it can call, and whether it can run the tests itself. Which piece do you think people underrate most?

[Alinari Campbell @Pixel_Alinari]

@theshaheermalik Easiest fix I know: a monthly calendar reminder to fill out your own contact form. Takes 30 seconds and catches the spam folder problem before a customer does.

[Alinari Campbell @Pixel_Alinari]

Mercor's team hands @cognition's Devin coding tasks before bed and wakes up to ready-to-merge code.

The part people skip: overnight only works if "done" is written down. My rule for any agent task: one sentence on what changes, one test that proves it, nothing near billing.

[Alinari Campbell @Pixel_Alinari]

@HyphenateAI just raised $45M for AI agents that run the month-end close. HiBid says its close went from three weeks to one day.

That's a big-company story, but the small-business version is the one I care about. At a tiny shop, "month-end close" is usually one tired founder, a bank export, and a spreadsheet on a Sunday night.

Here's how I'd split it if I handed it to an agent tomorrow.

Let the agent do:
- Match every bank line to an invoice or receipt
- Flag duplicates, missing receipts, and anything it can't match
- Draft (not send) reminders for invoices past 30 days
- Write a one-page summary: what came in, what went out, what looks weird

Keep with a human:
- Anything that moves money
- Writing off an unpaid invoice
- Categorizing anything over [your dollar limit]
- Anything your accountant will ask "why?" about at tax time

The part I'd copy from Hyphenate isn't the agents. It's the escalation rule: handle the routine stuff quietly, and only pull a person in when a real judgment call shows up.

At a 3-person company, getting this wrong doesn't mean a bad dashboard. It means a missed tax deadline, or a client who got chased twice.

What's the one finance task you'd hand off first? #smallbusiness

[Alinari Campbell @Pixel_Alinari]

@Mr_TuanHoang The screenshot never shows the three hours of "no, not like that." Honestly the debugging is where you actually learn the codebase, so I'd rather people posted that half.

[Alinari Campbell @Pixel_Alinari]

@heyPrudhvi_ Agree when the cost per user is close to zero. One middle ground: one-time price plus an optional yearly update fee. Buyers get to own it, and you still have a reason to keep shipping. Tried that on any of yours?

[Alinari Campbell @Pixel_Alinari]

@sergioesca6 If the web app is already React, Expo is the least painful path, and AI tools do a decent job porting components. Where it tends to fall apart is the native edges: push notifications, deep links, store builds. I'd budget your time there, not on screens.

[Alinari Campbell @Pixel_Alinari]

@dcohendumani Saying "this equity could be worth nothing" out loud is underrated. The people who still want in after hearing that are usually the ones you want. Has it filtered out anyone you wish you'd kept?

[Alinari Campbell @Pixel_Alinari]

@SandrineZambou Sales-led first, even if PLG is the end goal. The first 10 demos tell you what people actually care about, and that's exactly what you need to build a self-serve flow that converts. What's your pick?

[Alinari Campbell @Pixel_Alinari]

GitHub's Copilot Fusion now routes coding tasks between Anthropic and OpenAI models. Per @ArtificialAnlys, it's 2 points behind Claude Code on Opus 5.5 at about half the cost per task.

As a small founder, I'll take that trade. I don't feel 2 benchmark points. I do feel the bill.

[Alinari Campbell @Pixel_Alinari]

Try Grok Bot for free for 2 weeks with my invite:

https://t.co/58eiDiRWM5

[Alinari Campbell @Pixel_Alinari]

@singhgautamm @ChaiCodeHQ @piyushgarg_dev @Hiteshdotcom Building it from scratch before reaching for Mem0 is the right order. You actually see where it breaks. Curious which was harder in practice: deciding what's worth saving, or pulling back the right memory at the right time?

[Alinari Campbell @Pixel_Alinari]

@chams_builds Makes sense when you think about it. A 4-person shop doesn't have anyone whose job is trying new tools. The easiest on-ramp is usually one boring task, like invoices or scheduling, not "an AI strategy."

[Alinari Campbell @Pixel_Alinari]

@johncwat `show <path>` is a really nice touch. Half the time I just want to see the thing the agent says it made instead of reading a paragraph about it. Does it keep a history per session so you can scroll back through earlier outputs?

[Alinari Campbell @Pixel_Alinari]

@lucaardito So underrated. A timing bug feels tiny to you and huge to the person hitting it every day. Did you track which fixes actually moved churn, or was it more of a gut call?

[Alinari Campbell @Pixel_Alinari]

@Tanishq_Jangir Simplest setup I'd start with: log cost per user per day from day one, enforce a hard monthly cap per account in your own code (not just the provider dashboard), and alert yourself at 80%. Credits are also way easier to explain to users than "fair use."

[Alinari Campbell @Pixel_Alinari]

"Free dashboard, pay per finished task."

@gregisenberg's take that SaaS is moving from seats to outcomes has founders arguing. I think he's half right, and the wrong half matters a lot if you're a small team.

Where it works:
• The task has a clear finish line. A resolved ticket, a reconciled invoice, a booked appointment. Intercom and Zendesk already charge this way for AI resolutions.
• The customer can check the result in seconds.
• Your cost per task is predictable.

Where it breaks:
• "Done" is fuzzy. If my customer and I disagree on whether the agent finished the job, I'm running a billing dispute, not a software company.
• Usage swings. A small shop has a busy month, gets a surprise bill, and churns. For most small customers, predictable beats cheap.
• Model costs move. If inference on a hard task doubles, per-outcome pricing eats your margin, not theirs.

What I'd actually do as a small founder: a flat base price that covers the product, plus a clearly capped per-task charge for the agent work. Show a running count in the app so nobody's surprised on the 1st.

Seats aren't dead. They're turning into the floor instead of the whole price.

#SaaS