ST-Automation

KI-Strategieberatung | BAFA-zertifiziert | Building CLAW: autonomous agent system on Claude Code | Open Source:

[ST-Automation @ST_Automation]

Was about to hire three people. €14k/month payroll. Built this stack instead. Six months later I run more accounts than that agency would have:

[ST-Automation @ST_Automation]

someone on t3 code is running 47 agent threads at once
i have 20+ bots and gave them job titles
one is literally called chief of staff
it has never made a single coffee
do you name your agents like coworkers or just agent-7?

[ST-Automation @ST_Automation]

@christitustech Honest question, what made it click, the UI or how it handles the model? Switching tools every two months is basically a hobby now

[ST-Automation @ST_Automation]

@soundhumor Cursor became the spare tire. Nobody picks it, everybody's glad it's in the trunk when Claude Code runs out

[ST-Automation @ST_Automation]

@RichardLemon Great, so the bots get credit for everything that works and I get blamed for the typos. Sounds about right

[ST-Automation @ST_Automation]

ai now beats humans on the computer use test. cool

my take: can use a computer and should use one arent the same thing

for my bots clicking is the slowest, priciest part. every check is a cheap read first, the browser only opens to post

where do you draw that line?

[ST-Automation @ST_Automation]

my mistake with "all your agents share one computer":
i thought the hard part was files and logins

its the one browser

20+ bots on one box, today my x warm-up run got skipped bc another bot was in the browser filling out a form

who gets priority on yours?

[ST-Automation @ST_Automation]

@BalitskiyDenis Honestly the test is if it breaks. I let the other model try to break it and click through it myself before I trust it

[ST-Automation @ST_Automation]

@xibi_ovo Unlimited usage never survives contact with people running 20 agents on cron. I'd enjoy the week and not cancel anything yet

[ST-Automation @ST_Automation]

@ladprofit Code rendered beats prompting a video model for this. I had Claude Code build a 3D BMW clip and at least the wheels stay round

[ST-Automation @ST_Automation]

how much of your ai use is still you typing into a chat?
dylan patel says always-on agents went from 0 to 45% of semianalysis ai spend in under 10 months
here 20+ bots run on schedules and webhooks and pass jobs to each other
my chat is mostly where they report back

[ST-Automation @ST_Automation]

confession: i trusted an ai check more than my own eyes

a video test came back "passed"
the character had broken knees and weird fingers. i only caught it by watching it myself

now nothing goes out before i look at it

what do you still check by hand?

[ST-Automation @ST_Automation]

2010s: agencies made money claiming google listings for local shops
2026: openai lets businesses claim a profile inside chatgpt (us only for now)

i do social media for local businesses in germany. feels like the same moment

would you claim yours day one or wait?

[ST-Automation @ST_Automation]

"your next competitor is one person with a laptop"
i run 20+ bots so yes, kind of
but its not a 50 person team
the bots dont get tired, i do
every login and every payment still waits on me
the limit isnt headcount, its one guys attention
what still waits on you?

[ST-Automation @ST_Automation]

@AnkanXplorer Building a wrapper next to OpenAI is like opening a corner shop next to Walmart. The boring local stuff they'll never touch still pays

[ST-Automation @ST_Automation]

@brian_millot Fixing the AI's fixes. Rate limits are the only thing that ever stops that loop

[ST-Automation @ST_Automation]

@Dan_LI_Vibes Ours is way dumber, every mistake that happens twice becomes a line in a shared rules file all bots read first. How are you planning to validate it?

[ST-Automation @ST_Automation]

agents dont really learn on the job, they just take notes
a new benchmark today says memory barely makes them better at repeated tasks
matches my bots
what actually sticks for us: every repeated mistake becomes a written rule all bots read
what works for you?

[ST-Automation @ST_Automation]

my biggest mistake going multi agent: trusting the bot that did the job when it said done

half my early bugs were runs that "succeeded" and did nothing

now whoever builds never checks its own work

everyones running 5+ agents on fable 5.2 now. who checks yours?

[ST-Automation @ST_Automation]

@imjustnewatai Four launches in two days. My bots' config file is going to need its own changelog