Pinned Tweet
In times of technological acceleration, planning based solely on what’s possible today limits potential. Instead, set a bold vision and iterate relentlessly toward it.
Our vision is to make websites more relevant, timely, and deeply connected to their communities—while remaining memorable and unique. Autonomous Websites are the transformative technology that will make this a reality.
We’re already making strides on this journey, delivering innovations that pave the way. Within the next 4-5 years, we’re committed to fully realizing this vision.
And we were all worried about new models hacking into systems. Human emotions and desires, like Jensen’s, are the fragile component.
159
Fifteen years ago we built Prismic's database ourselves.
BookKeeper for the distributed log, ZooKeeper for sync, Lucene for search. The data itself was S3 files mounted into memory, because there is nothing on earth cheaper than storing S3 files.
Combine them in the right way and you have a database, with exactly the properties we needed - cheap storage, cheap cloning, scalable, cheap rollback!
Amazing. And it stood the test of time too. It's still serving our APIs today, with all the clients we have.
So it feels a bit nostalgic that we're now retiring it for vanilla Postgres.
Do I regret building it? No, not at all. It was what we needed at the time, and it did its service.
But there was always a hidden cost to it that the usual build vs buy debate ignores. They always talk about how much it costs, does it scale, the maintenance burden, tech debt, and such.
All valid, but for us they weren't the real cost -- it was onboarding!
Every engineer we ever hired had to learn a system, that exists nowhere else in the world. Every bug was our own, nobody else had seen it. There was nobody to ask externally. Special code and workarounds slowed down everyone who ever touched it.
If I had to compress fifteen years into one lesson - the major cost of bespoke infrastructure is *how much it slows down a new person*.
And there's a new victim now, one that didn't exist when I made the decision.
Claude knows Postgres and other widespread frameworks and platforms, their APIs and internals perfectly.
It knew nothing about ours though. An undocumented, eccentric system used to be only confusing for new hires, but now its confusing for all your agents too. Even worse.
Next time, I'd weigh one number above all the others -- time-to-first-commit.
7
8
544
Fifteen years ago we built Prismic's database ourselves.
BookKeeper for the distributed log, ZooKeeper for sync, Lucene for search. The data itself was S3 files mounted into memory, because there is nothing on earth cheaper than storing S3 files.
Combine them in the right way and you have a database, with exactly the properties we needed - cheap storage, cheap cloning, scalable, cheap rollback!
Amazing. And it stood the test of time too. It's still serving our APIs today, with all the clients we have.
So it feels a bit nostalgic that we're now retiring it for vanilla Postgres.
Do I regret building it? No, not at all. It was what we needed at the time, and it did its service.
But there was always a hidden cost to it that the usual build vs buy debate ignores. They always talk about how much it costs, does it scale, the maintenance burden, tech debt, and such.
All valid, but for us they weren't the real cost -- it was onboarding!
Every engineer we ever hired had to learn a system, that exists nowhere else in the world. Every bug was our own, nobody else had seen it. There was nobody to ask externally. Special code and workarounds slowed down everyone who ever touched it.
If I had to compress fifteen years into one lesson - the major cost of bespoke infrastructure is *how much it slows down a new person*.
And there's a new victim now, one that didn't exist when I made the decision.
Claude knows Postgres and other widespread frameworks and platforms, their APIs and internals perfectly.
It knew nothing about ours though. An undocumented, eccentric system used to be only confusing for new hires, but now its confusing for all your agents too. Even worse.
Next time, I'd weigh one number above all the others -- time-to-first-commit.
4
10
603
Why did the stress ball end up on every office desk?
When work moved into the digital world, people suddenly were spending all day in front of a screen. They became too virtual and needed something to hold, something that would bring them back into the physical world.
I think it's happening again because of AI.
Working with agents means living almost entirely in your head. You read, learn, decide, prompt. Again and again until your body disappears.
That is mentally, and almost spiritually, exhausting.
I noticed it in myself how I work now. Between two AI iterations, I once spent my waiting time trying to crack the code of a lock I couldn't open. Combination after combination. I didn't plan it, but my hands just needed something real.
At Prismic I see people colouring, playing the office piano, and increasingly ambling over for a chat with someone.
This isn't procrastination, it's the brain trying to reengage with the real world again.
So I think offices will change a lot in the coming years. They'll be more than just "staying in shape" wellness, having objects that reconnect you with reality and the physical plane. Lego on the tables, instruments and things to build, perhaps animals?
We're actively thinking about it for our own office.
What do you do with your hands while your AI works?
1
95
Visitors ≠ Users.
Yes, new technologies have transformed the browser into a full blown runtime, and the line between *website* and *web app* is really blurring.
But they aren't separate technologies so much as separate postures. The crucial distinction is in the visitor's head!
An app is an experience you've (on some level) agreed to learn. A website is an experience you have agreed to *glance* at.
I got so into Suno recently, and initially I was expecting to invest time and mental energy in learning my way around. I use Ableton already, so I know these tools take some learning.
I spent time in their UI, I explored, I tolerated some level of confusion. Up to a point, at least, and the moments of frustration came slowly.
On a website, they would've come in seconds. A visitor isn't trying to get value immediately, they are trying to verify "am I in the right place?".
If the website doesn't answer that in a second, they bounce. Different psychology and behaviour entirely. Which posture you optimise and design for decides which one you've built.
Think of a storefront. The window is arranged to be scanned from the street in a glance. Once you get inside and commit you accept some orientation and effort, but you don't want that from the window.
Expecting a visitor (glancer) to be a user (learner) is one of the most common website design failures.
And as browsers become more capable runtimes, this distinction becomes more critical. You can now build almost anything in the browser, the technical barriers are melting away.
But the psychological ones are as strong as ever. Blur the distinction, and you can lose both. Too complex for visitors to parse, too shallow for users to invest in.
Web developers and designers, what makes websites still relevant?
1
158
So let me get this straight.
Tech leaders keep telling people they will lose their jobs for AI
People’s children don’t have proper schools and education
and we blame immigrants for that.
Did I put it right?
1
233
People look at strikes in France and see it as chaos - I see it as a sign of human values in this society, that people have a say.
I also see a lot of people mocking Europe for slow reactiveness, and questioning their priorities, like the famous eco-friendly bottle cap.
But could this friction be an advantage?
I see a world where Europe builds AI that actually respects humans, and Mistral seems to be steadily closing the gap to the frontier. I see Europe and France having real say in what the AI future looks like, leading in innovation while staying human.
Perhaps the traits that people mock Europe for will be the same things that make AI more trusted and aligned in the long run. Maybe Europe will benefit from the last mover advantage.
AI watchers out there - plausible or wishful thinking?
3
163
Someone just rebuilt most of the Adobe suite in Rust, open source and free. Photo editing, vector, video, RAW, PDF, motion graphics, page layout. All of it!
Does this mean Adobe is "dead", as some say?
Of course not. Adobe users don’t stay because they love the subscription. It's the muscle memory, systems, compatibility, plugins. Code itself is becoming a commodity, perhaps, but community and taste aren't.
Is this enough to win that Adobe community? Is there an edge we’re not seeing yet?
Curious what designers think.
AI gives us all the state of the art, but that's already in the past... Are you using AI to just implement the past, faster?
2
146
Sadek Drobi retweeted
Opus5: "Rotate your $$KEY$$ — one of my commands printed it in full into this transcript."
Claude, are you serious, I thought you knew about security ?
I've been seeing a new anxiety on the feed and in conversations. It's something like:
"If I build something good that uses AI, won't the model provider just build it as a feature? Is my product just a temporary trick destined to be absorbed?"
1
5
148
Both of these are culture and habits, and skipping them kills the compounding.
Assume that OpenAI can, if they want, eventually copy what you do or make you a feature.
1
26











