quit my CEO job to build solo. first-time dad. techno-optimist.
Vienna, Austria
Joined January 2015
- Tweets5.1K
- Following472
- Followers1K
- Likes7K
Pinned Tweet
quit my CEO job.
first kid on the way.
building 3+ products from vienna.
most exciting time of my life. building all of it in public.
please listen to dessa.
i have never heard anything artificial that sounded so… human.
if you ever doubted that human-ai-relationships were going to be a thing. well. this snippet will convince you otherwise.
We believe voice is the most natural way to interact with intelligence, so we created Dessa, the most authentic full-duplex speech model ever built. Talking with it feels like talking with a person…full of emotion, nuance, pauses, and a natural cadence.
block.xyz/dessa
i'm shitty at sales.
yesterday i made my first sale anyway.
never had to do it before. at the lab, other people sold. i built things.
now my new business gründerradar needs customers, and it turns out software doesn't sell itself. somebody has to pick up the phone. that somebody is me.
so i did what i always do when i'm bad at something. i built a tool for it. took 30 minutes with claude.
a call cockpit. runs in my browser and walks me through every call. all in german, my customers are tax advisors in austria.
14 calls so far. one rejection. one paying customer. the rest were super positive.
before the call, 2 minutes of prep.
who i'm calling, their title, their area and the numbers for their area. how they found me: they booked a slot, asked for a callback, scanned the QR code on my letter and went quiet, or they're calling me. everything i type fills in the script. no making up facts mid-call.
during the call.
a script with 119 steps and dynamic paths. secretary picks up? voicemail? "just send me some info"? one tap and the next sentence is there.
92 objections and questions, answered in advance. i jump to one, answer it, and it drops me back exactly where i left off.
their numbers, live. what's a new client worth to them per year, and how long do they stay? the tool turns that into their own benchmark: one client over the years vs. a year of letters for their area.
note fields show up whenever i should write something down.
a running timer and a time budget per stage. opening 1 min, questions 3, solution 1, objections 2, close 2. the questions get the most time.
coaching notes for my worst habits. "slow down." "ask, then wait." "make the offer once, then stay quiet." half of them just tell me to shut up.
after the call.
the log writes itself: who, how they came in, their numbers, every objection in the order it came up, the result, the next step, the duration. one click to copy.
then i fix the script. it's on version 3.2.
builders who hate selling too: what got you through your first calls?
php has two critical holes. attackers can take over a server without logging in. cloudflare already sees attacks.
fix ships 14:00 utc. 8.1 and older get none.
updating wordpress doesn't update php. ask your host or agency today.
PHP sites behind Cloudflare already have an initial set of WAF rules deployed for the known exploit vectors in these vulnerabilities. The rules are available to all customers, including free.
We are seeing early attacks emerging and are updating mitigations as new vectors show up.
This is a STOPGAP only. Patch as soon as @ThePHPF releases the update at 14:00 UTC today.
how the hell did i miss this entirely.
goddamn... so pissed about this.
i have no idea how i missed this:
tell opus 5.5 to use codex as subagents for any kind of work.
it is literally going nuts right now.
spawning dozens of little helpers, dynamically choosing model and reasoning effort, and i FINALLY get to spend these pesky resets.
opus stays the brain, double checks everything codex does and saves A TON of claude usage.
anyone else running it like this?
you can forgive a lot of crazy behavior if in turn you deliver affordable space travel and motherfu**ing flying cars.
this is the future i want to live in.
this is the future i want to BUILD. M
had this discussion too many times over the weekend. the ignorance is EXHAUSTING.
“it is not creative”
“it’s not intelligent, it’s just one big pattern matcher”
“an LLM has no understanding of your code”
“quality is horrible and riddled with bugs”
“i’d be faster writing it myself”
and my favorite: “with that attitude you’ll crash and burn sooner or later”
these aren’t free-plan tourists btw. senior devs with full opus + fable access at work.
I can't believe really think this
Have you used Opus 5.5? Astra?
Maybe these people are on the free or cheap plans and haven't used top frontier models and simply do not know how great they are?
Remember when majority of the population thought AI was dumb because they'd used ChatGPT 3 once in 2023 and it hallucinated and then just concluded AI was dumb for years until trying it again in 2025 or so
sunday in the forest with the family.
checked on the agents twice.
meanwhile they moved one of my products into germany. a market about 10x the size.
didn't need me at all 👌
tokenmaxxing while the subsidies last. my agents outgrew two VPS.
i have been developing on two VPS for months now but i keep burning through their capacity in terms of cores, ram and most pressingly storage.
so now its time for a little upgrade.
the plan:
> one central always-on dev machine
> all the agents i want live on it
> remote access
> tokenmaxx until the cheap subs are gone
> dont care if something breaks because this is my "move hyperfast and break it regularly" machine
62k free credits just like that?
i smell fear.
i have not touched my codex sub since opus 5.5 came out.
very few people realize that that the biggest achievement here is not the 5M in net worth.
amazing, mate!
I finally reached the magic number of $5M in net worth 🥳
This isn’t all cash. It’s everything combined:
+ My 36 startups (90%+ of their value comes from DataFast & TrustMRR)
+ Investments (85%+ in the S&P 500)
+ Cash
What surprises me most is getting here at 33 without ever:
- Working more than 8 hours a day
- Building anything grandiose
- Working for someone else
- Doing work I don’t like
- Going to an office
Don’t you dare give up :)
"not x, but y"... huh?
guess rob is monetizing his indie hacker followers with a course created by claude now.
not sure how i feel about it.
nevertheless, i'm happy for you on a personal level, rob.
working with agents really makes you cynical and an expert in gaslighting.
one day its the smartest person (?) in the room, proactive, creative and just to the point. next day it commits faulty db migrations to prod, sends an email to the authorities that your GDPR notice is non-compliant and denies doing so in the end.
*sigh*
opus 5.5 is great btw.
the level of arrogance required to pull this off is honestly mind-boggling.
I finally tested GPT-6 Sol on a real work. 2 repos. 105 hidden bugs. Find and fixed what you can.
It looks like a huge degradation. The results:
- GPT-6 Astra (max): 45
- GPT-5.6 Sol (max): 43.5
- Opus 5.5 (max): 41.7
- Muse Spark 1.3 (max): 32.2
- GPT-6 Sol (max): 29.3
Until you measure the cost (API-equivalent):
- GPT-6 Astra (max): $33.04
- GPT-5.6 Sol (max): $95.25
- Opus 5.5 (max): $58.53
- Muse Spark 1.3 (max): $18.11
- GPT-6 Sol (max): $9.93
More effort levels (xhigh, high, medium, low) dropping in this thread today 🧵
sol-6 is worse than sol-5.6.
i have just switched out 5.6 for 6 on a super simple classification task i have running (match company to one or more categories) and benchmarks droped sharply.
same prompt. same input. objectively worse performance.
i feel like all of the benchmarks are fake at this point and even if the model performance is benchmarked right, the lab will just silently serve you a weaker/quantized model.
how do we break this fraud cycle?
