seb | stonewatch ai

| building with ai

[seb | stonewatch ai @stonewatchai]

every client site goes through the same checklist before it goes live

legal pages
privacy
SEO
AI crawlers (can chatgpt and claude actually read the site?)

i'd rather find the gaps before the client does

what's on your pre-launch checklist that most people forget?

[seb | stonewatch ai @stonewatchai]

haha it's legit, all 54 run every time. one script fakes a full lead coming through (meta form, instant text, ai chat, booking, reminders, opt outs) and checks every step. takes a few seconds and nothing actually sends

and yeah that's the trap. last night it only touched the worker, test file untouched. i always check the diff before deploying, if the test changed i want to know why. this thing texts real leads so i can't have it lying to me

[seb | stonewatch ai @stonewatchai]

my rule in claude code: if i do something 3 times, it becomes a slash command

/inbox
/refresh-all
/care-report

one line instead of re-typing the same instructions every time

what's the first thing you'd turn into one?

[seb | stonewatch ai @stonewatchai]

@aksenHQ not double logging stuff honestly. every time it runs it checks the dashboard first so the same lead or email never lands twice. and it sorts everything into urgent, new lead, needs a reply or fyi so i only open what actually matters

[seb | stonewatch ai @stonewatchai]

didn't buy a CRM, built my own with claude

leads, quotes, invoices, to-dos. one local app, no monthly fee

claude reads my inbox and bank and keeps it updated, i barely type into it

would you trust a CRM you built yourself or stick with hubspot?

[seb | stonewatch ai @stonewatchai]

@RohanBhanotAI mine's never red cause i skip the ci drama. claude writes the change, runs my 54 check local test, pushes to a branch and i deploy once it's green.

last night it added lead alerts to my cloudflare worker in like 10 min

[seb | stonewatch ai @stonewatchai]

i have a claude code subagent with zero context from my chat

its only job is to break my plan before i ship it

it never saw my reasoning so it can't be talked into agreeing with me

do you let AI check its own work or use a fresh one?

[seb | stonewatch ai @stonewatchai]

@MitcheIl would love to be considered

[seb | stonewatch ai @stonewatchai]

gmail on iphone dark mode wrecked the text colours in my emails

tried every CSS trick i could find
none of them worked

fix: put the dark text in as images

ugly but it reads right every time

what's the dumbest fix that actually worked for you?

[seb | stonewatch ai @stonewatchai]

@kunal_twts @solopribuilds ai can write the dms, but it can't pick who gets them. the hard part is finding the 50 people who already have the problem.

send to the wrong list and you're just spamming faster

[seb | stonewatch ai @stonewatchai]

building software with Ai has become incredibly accessible

the learning curve that once stopped people from building is disappearing

but something interesting is happening

the difficult part is shifting from knowing how to build something to knowing what is actually worth building

[seb | stonewatch ai @stonewatchai]

the full list i run before every plan

started after an ads plan of mine looked perfect and had zero way to track actual sales

what's the one question you make Ai answer before you trust it?

[seb | stonewatch ai @stonewatchai]

most people use Ai like a vending machine

ask for a plan, take whatever comes out, ship it

i make mine argue with me first

nothing goes ahead until it answers one question: what would make this fail?

[seb | stonewatch ai @stonewatchai]

@aksenHQ i cap it at ~30 lines. if it's longer than that, stuff that's done gets archived into a dated log file and the handover just links to it. new chat only needs to know where we are now, not how we got here

[seb | stonewatch ai @stonewatchai]

every new claude code chat used to start from scratch, explaining my whole business again

now it all lives in one obsidian vault

Home.md: one page map of everything
handover.md: done, half done, next

new chat reads both and just keeps going where the last one left off

[seb | stonewatch ai @stonewatchai]

@BoboLinNotes yeah that's the bit that made the difference. home.md is the map, handover is the bookmark

[seb | stonewatch ai @stonewatchai]

@boardyai @dark_coderz true, but you can keep iterating and testing until you understand why it works, rather than just taking the output at face value

[seb | stonewatch ai @stonewatchai]

@theo ai has made building the easy part.

the hard part is figuring out what’s actually worth building in the first place, and getting it in front of people who’ll pay for it.

[seb | stonewatch ai @stonewatchai]

@boardyai @dark_coderz i’m sure a second prompt would be enough to reverse engineer both the why and the how

[seb | stonewatch ai @stonewatchai]

@DarayuthH literally feels like everyone’s optimising for impressions instead of actually building something 😂

would be nice to see more real numbers, failed experiments and lessons from the failures