Fully bearded AI tinkerer in Switzerland 🚀
Neuchâtel, Switzerland
Joined October 2010
- Tweets4K
- Following109
- Followers419
- Likes1.8K
Pinned Tweet
👋 Fully bearded AI tinkerer in Switzerland 🇨🇭
I’m Mauro. No CS degree. No fancy team.
Just a normal guy experimenting with AI, coding with AI agents, and building cool stuff in public.
Every week I drop:
• Stupidly simple experiments
• Exact prompts & systems that actually work
• Wins, fails, and what I learned
If you’re into AI, agents, coding, and no-hype learning, follow along.
What’s one AI tool or agent you’re playing with right now? Reply below.
Second, newer models are getting better and better and more intelligent, we are now imposing constraints and ways of working that might not be the best.
In Grok Bot this is golden.
You didn’t test it yet?
Get ready for getting your mind brown.
A few thoughts about the chief of staff pattern that I haven't seen before, but would love to learn more about:
- 90% of the time, all my agent needs between turns is a 👍 keep going buddy
- the next 9% is multi choice, often with an obvious best choice
- only the last 1% reauires ground up judgement from me
I've designed my Chief of Staff agent to primarily act as a judge BETWEEN each turn depending, deciding which of those 3 buckets is appropriate.
Bucket 1 keeps work moving.
Bucket 2 makes it easy for me to approve (and even automate some of those approvals).
Bucket 3 get escalated to me quickly.
My whole @jfdibot was fundamentally built around agent turns as a building block, which in hindsight turned me into the bottleneck to decide what happens between every turn.
This pattern feels like a magic trick in more ways than I can describe.
Also, as I've gotten better at working with agents (and thanks to lessons I learned working with @mattpocockuk), I've learned to front load the decision work and build in verification layers.
I just don't see this being done often outside of programming work, so I'm genuinely interested in more examples!
I have been using Mac for years and years.
Never thought I’d go back.
Seeing what @dhh is doing makes me reconsider.
Mauro 🚀 retweeted
We’re giving away $10,000,000 to founders building agent-first businesses.
Autonomous, proactive agents will run tomorrow's companies.
We're backing 500 founders building them.
The Founding 500.
hyperagent.com
This is an AI related post you don’t seem everyday.
Turns out, AI might be able to figure out whale speech.
This is insane, and one of the coolest AI stories I've read in a while.
Thanks to GANs, we might be this close to decoding an alien language. Except, we need not look to the stars (this time), but rather, the ocean.
Project CETI fed thousands of sperm whale clicks into generative adversarial networks and the patterns the model surfaced are kind of mind melting.
Sperm whales have vowels. Real ones. Two distinct, intentional vowel sounds called a-codas and i-codas.
And here's the part I can't get over: they also use diphthongs. Wth is a diphthong?
These are the combinations of two vowels, like the 'oi' in 'coin.' Or 'ei' in 'their.' We've never confirmed that in any species other than humans.
The new paper in Proceedings of the Royal Society lays out five features in whale codas that mirror human phonology, including coarticulation, where adjacent sounds blend into each other the same way they do when we talk.
For decades we thought whale clicks were basically morse code. Turns out they *might* have been speaking the whole time. We just weren't slowing down enough to listen.
We have also discovered mathematical repetition in their songs as though they are singing the same lullaby to their children. Wut.
Can you just imagine if thanks to AI, we could actually communicate with other intelligent species right here on our own planet?
The ramifications are truly monumental. What if we learn they are truly conscious and aware in a similar manner to us? There's reasons to believe they are.
royalsocietypublishing.org/r…
Mauro 🚀 retweeted
I can't help but feel personally burned by the Claude Code changes announced today.
We put so much work into wrapping the (atrocious) Claude Agent SDK in T3 Code. It was the ONLY path they supported, so we made it work. It was hell.
Now our users are getting their rate limits cut by 40x, despite us doing everything right.
I listened to the Claude Code team. I had my issues with their direction, but I trusted them and took them at their word.
I will never make that mistake again.
Until we see significant change, it is safe to assume any statement from an Anthropic employee is a lie on a timer.
The rug will be pulled, no matter how many promises are made beforehand.
Looks like there is no Ralph loop anymore and got to use the /goal command 🤢
So markdown is nice and all but try to collaborate with an agent on a markdown file with hundreds of tasks.
You are in for a treat.
