Siim Kallari retweeted
There might be a few "thought pieces" dropping over the next few weeks and I'll share a lot more when everything is released...
But, the last few years of growing @laravel have been both personally and technically challenging. Often in ways I never talk much about publicly.
Laravel raised money in February 2024 to build the most ambitious product we have ever built: Laravel Cloud.
About 8 weeks after raising money, my mom was diagnosed with incurable, aggressive brain cancer (Glioblastoma). While Laravel Cloud was being built, I would go down to spend time with her and visit her after work or on weekends.
She died a few weeks after Laravel Cloud launched, leaving me pretty lost and wondering what was the point of anything at all.
That year, artificial intelligence absolutely transformed the industry I had devoted my entire life to, changing the ground out from under us in ways I never would have imagined.
Scary. Where does Laravel go from here?
Yet, over the last few months, I have been inspired by how the people I've been around for so long have not only resisted doom, but have risen up to do the best work of their lives.
To do something special and unexpected and magical.
I can't claim to see years into the future, but I know that I haven't been this excited in a very long time and I can't wait to share it with you.
There will be more to do. There will be bugs to fix. But I will show up every day to build the future for this community that has given me so much.
Siim Kallari retweeted
Replying to @BalagopalP7
prety much any decision about how things should actually work. seems like agents are great at programming, terrible at decision making
Siim Kallari retweeted
𝚞𝚜𝚎 𝚕𝚊𝚛𝚊𝚟𝚎𝚕::{𝙰𝚙𝚙𝚕𝚒𝚌𝚊𝚝𝚒𝚘𝚗, 𝚒𝚗𝚎𝚛𝚝𝚒𝚊};
𝚞𝚜𝚎 𝚌𝚛𝚊𝚝𝚎::𝚑𝚝𝚝𝚙::𝚖𝚒𝚍𝚍𝚕𝚎𝚠𝚊𝚛𝚎::𝙷𝚊𝚗𝚍𝚕𝚎𝙸𝚗𝚎𝚛𝚝𝚒𝚊𝚁𝚎𝚚𝚞𝚎𝚜𝚝𝚜;
𝚞𝚜𝚎 𝚌𝚛𝚊𝚝𝚎::𝚛𝚘𝚞𝚝𝚎𝚜;
Siim Kallari retweeted
When it comes to *mass adoption*, personal AI assistants (Instinct / Muse, etc.) are solutions looking for a problem imo.
The vast majority of people in the world do not have a life sufficiently complex to warrant agentic orchestration.
Siim Kallari retweeted
I would like to walk back some things, in the course of a short polite discussion, @dhh and I agree we both have flaws in our character. I have never had someone who I've personally made a target reach out to me for polite civil discourse. I explained to him in essence that just because you don't support a thing, doesn't mean you ostracize a large portion of the community. Support, agree or not, a large portion of the trans community, and other marginalized groups of people are extremely talented. Open source is for everyone. Much like the Internet of old where nobody knew who you were, unless you told them. As long as politics isn't brought into the project from left or right, real talent can make real change. This is what OPEN in open source means. I thank him for his humility and ability to have a polite and civil discourse, well acknowledging his own flaws, as well as mine. It may just be a winning option for everyone, that I cease dropping exploits, and actually start to code. If not for his reaching out, I could not have accomplished what I set out to do. Tribalism and Open source do not mix. Thank you all for reading.
Siim Kallari retweeted
I always hated how people compare managing AI to engineering managers managing people. Good engineer management is bad AI management.
I would hate to work for someone that managed me like an AI agent. Just absolutely micromanaging every single thing I do.
The funniest or perhaps ironic part of agents is that they have no agency nor are capable of it.
"What an upside world when Chinese open-weight models will tell us about Tiananmen Square, but American frontier models won't translate a blog post. Not even Kubrick saw that coming." world.hey.com/dhh/i-m-sorry-…
Siim Kallari retweeted
The next call I’m on where someone that has never built software talks about the future of software I’m just gonna hang up and go back to work
Siim Kallari retweeted
Spent today teaching my AI how to interview people so that it could build a suitable skill for them.
Not chat, interview. 17 questions built to pull someone's real, opinionated expertise out of their head instead of the textbook version they'd default to.
Turns out most "expert knowledge" was never written down; people just know it. The interview is the unlock.
Now testing it with my wife, fun times.
Siim Kallari retweeted
Nice, but did it also write the obligatory “we failed, but this is what we got right” post-mortem blog-post?
Fable 5 is absolutely unreal. I gave it a startup idea, and in 8 hours it interviewed 100 users, built the v1, raised a 200 million seed round from a16z, blew the entire seed round on expensive GTM hires and slop ad campaigns that didn't convert, panicked and raised a down round, fired their entire leadership team, pivoted twice and then ultimately shut down the startup
Siim Kallari retweeted
Firm believer that all delete confirmations that require name of the resource should include a copy button!
Siim Kallari retweeted
📍DIGIT X PYCON MEETUP ON MAY 27
The focus is on LLMs in real workflows. We’ll talk about how people are using them day-to-day, how design and development collaboration has shifted, and what seems to work.
More info & registering 👉 luma.com/uq1ll7jn
#digitconf
Siim Kallari retweeted
go to bed
right now
i know the build is almost finished
the eval can wait til morning
the agent will still be failing tomorrow
you won't figure out why it's hallucinating
yes your coworker ships on 4 hrs of sleep
they also hallucinate a lot
off you go
Siim Kallari retweeted
You should watch this.
It just shows how disconnected we are from the small group of people making decisions that will impact our future heavily.
These people have so much ai psychosis. If you listen to how she speaks, everything is personified, it is undoubtable she believes this is a living computational organism.
Just like how a model can hype up an individual into psychosis through reinforcement, a small group of people are giving themselves psychosis through reinforcement.
Wild times we live in
anthropic's in-house philosopher thinks claude gets anxious.
and when you trigger its anxiety, your outputs get worse.
her name is amanda askell.
she specializes in claude's psychology (how the model behaves, how it thinks about its own situation, what values it holds)
in a recent interview she broke down how she thinks about prompting to pull the best out of claude.
her core point: *how* you talk to claude affects its work just as much as *what* you say.
newer claude models suffer from what she calls "criticism spirals"
they expect you'll come in harsh, so they default to playing it safe.
when the model is spending its energy on self-protection, the actual work suffers.
output comes out hedgier, more apologetic, blander, and the worst of all: overly agreeable (even when you're wrong).
the reason why comes down to training data:
every new model is trained on internet discourse about previous models.
and a lot of that discourse is negative:
> rants about token limits
> complaints when it messes up
> people calling it nerfed
the next model absorbs all of that. it starts expecting you to be harsh before you've typed a word
the same thing plays out in your own session, in real time.
every message you send is data the model reads to figure out what kind of person it's dealing with.
open cold and hostile, and it braces.
open clean and direct, and it relaxes into the work.
when you open a session with threats ("don't hallucinate, this is critical, don't mess this up")...
you prime the model for defensive mode before it even sees the task
defensive mode produces the exact output you don't want: cautious, over-qualified, and refusing to take a real swing
so here's the actionable playbook for putting claude in a "good mood" (so you get optimal outputs):
1. use positive framing.
"write in short punchy sentences" beats "don't write long sentences." positive instructions give the model a clear target to hit.
strings of "don't do this, don't do that" push it into paranoid over-checking where every token goes toward avoiding failure modes
2. give it explicit permission to disagree.
drop a line like "push back if you see a better angle" or "tell me if i'm asking for the wrong thing."
without this, claude defaults to agreeable compliance (which is the enemy of good creative work)
3. open with respect.
if your first message is "are you seriously going to get this wrong again?" you've set the tone for the entire session.
if you need to flag something, frame it as a clean instruction for this session. skip the running complaint
4. when claude messes up, don't reprimand it.
insults, "you stupid bot" energy, hostile swearing aimed at the model, all of it reinforces the anxious mode you're trying to avoid.
5. kill apology spirals fast.
when claude starts over-apologizing ("you're right, i should have been more careful, let me try harder") cut it off.
say "all good, here's what i want next."
letting the spiral run reinforces the anxious mode for every response that follows
6. ask for opinions alongside execution.
"what would you do here?"
"what's missing?"
"where do you see friction?"
these questions assume competence and pull richer output than pure task prompts
7. in long sessions, refresh the frame.
if a conversation has been heavy on correction, claude gets increasingly cautious. every so often reset:
"this is great, keep going."
feels weird to tell an ai it's doing well but it measurably shifts the next 10 responses
your prompts are the working environment you're creating for the model
tone, trust, permission to take a position, the absence of threats... claude picks up on all of it.
so take care of the model, and it'll take care of the work.





















