Agents run growth for my 6 mobile apps. I answer yes or no.
Since October: 17 experiments, 8 kept, 3 reverted, 6 running. Page conversion 19% → 41%. First debt added 47% → 56%.
Now doing the same for studios with 5+ apps. If growth eats your week and budget, DMs are open.
19
Replying to @automater_ai
Ha, exactly that. Moved the urgent tasks to Grok Bot for now, and the factory kept going.
5
commit hash is the version and skill is fully overridden on sync.
It is a bit different between platforms though...
5
Replying to @tiboadriana
Kind of agree. Agents write the code for my 6 finance apps now. But someone still has to pick what to build and check it actually works for users, and that's still me.
3
Replying to @thorstenball
I'd say skills are needed, but you don't have to encode them in the prompt every time. Models are smart enough to pick up the rules, and if not, you can always remind them.
Plus I keep one skill set for all agents in one repo: Claude, Codex, Grok Bot, etc.
1
72
Replying to @JeremyCMorgan
Same here. Too many sessions, no idea who's waiting on me.
So now they just ask. Every ask goes to one list, and I clear it a few times a day.
11
Replying to @KingBootoshi
Same trick works after launch.
App Store and Google Play reviews plus support emails land in one queue in my factory. An agent drafts the reply and links it to the bug in the backlog. I just approve.
1
182
Replying to @MaxHirsch13
Not betting on either one.
My skills and rules live in one repo outside any tool. Switching from Claude to anything else is just a sync.
Doing it right now, while Claude digests the 5-hour dose it ate in 2 hours.
53
Replying to @PaulSolt
Manager thread works for me too.
One agent in my factory only dispatches. It hands each item to the session already working on it, or starts a new one. And it never does the work itself.
42
Moved most urgent across Grok Bot and Codex. For now.
Next: a cost log per agent, and cheaper models for routine work.
8
First real number, from one task:
75 model calls, $9.44 at API prices. A third of it was the agent waking up after an hour and rereading the whole chat. $1 to $1.60 each time.
1
9
The fix was sitting in that same queue.
A project to run any agent on any model, with a cost log, so one vendor's limit can't stop the day.
Step 2, an agent on OpenCode and OpenRouter, never started.
1
13
Claude burned its 5-hour limit in 2 hours this morning.
One daily pass over the factory queue: store work, support, a review, a research doc. Then the sessions stopped.
Including the fix for exactly this problem.
4
1
31
Replying to @BO5AMIS
Real phone is the right call.
Simulators catch a lot for us, but spacing, scroll and the keyboard covering input fields only show up on a device.
Did it catch layout bugs too, or mostly broken flows?
1
1
29
Replying to @ShipToMRR
10+ years freemium, no trial, on my finance apps.
Hard paywall probably wins on day 60. But freemium got me reviews and rankings early, and my first app from 2015 still sells off them.
Replying to @NabhaniMehdi
Comparison page is a good call. I'm rebuilding every app's page now, then I'll rerun the same 59 questions and see if we show up anywhere.
1
3
Replying to @PaulSolt
Product pages matter more for me this year.
Since mid-October ChatGPT sent Money Goals 331 iPhone downloads. Google: 8. Those people skip search results and land straight on the page.
So the page does all the selling.
16
Since mid-October, ChatGPT has sent Money Goals 331 iPhone downloads. Google: 8. Those ChatGPT users buy Pro about twice as often.
So I asked ChatGPT 59 "which app should I use…" questions about my 6 apps.
We showed up in 0.
It reads developer websites first, not the App Store. Now rebuilding every app's page.
1
27
Replying to @UsefulSupplyCo @MaxHirsch13
Yep. Take the debt payoff niche. It's packed with apps, and mine still sells years later because it does one narrow job better than anyone else.
Clone the screens and you're just noise.
1
41



