Ryan Worl retweeted
YOU CAN SEE EVERYTHING
is what I told @zebriez this summer. but this time you won't be left guessing who we are, because she's articulated it beautifully.
colossus.com/article/sane-as…
I was drawn to Silicon Valley because of people's healthy relationship to work.
connection to mission. friends. ambition. financial security. hard work. but also free time. positive sum vibes. pride in being a part of it. joy.
this is no longer the case. my friends have gone mad. and there's almost nowhere to hide.
refusing to accept this fate, I went searching for safe havens.
I found (at least) one in ottawa and spent the last 6 months with @turbopuffer where they've got silicon valley chops, canadian kindness, and danish sense of humor. it's been the perfect medicine.
I wrote about it in @colossusmag
Ryan Worl retweeted
I was drawn to Silicon Valley because of people's healthy relationship to work.
connection to mission. friends. ambition. financial security. hard work. but also free time. positive sum vibes. pride in being a part of it. joy.
this is no longer the case. my friends have gone mad. and there's almost nowhere to hide.
refusing to accept this fate, I went searching for safe havens.
I found (at least) one in ottawa and spent the last 6 months with @turbopuffer where they've got silicon valley chops, canadian kindness, and danish sense of humor. it's been the perfect medicine.
I wrote about it in @colossusmag
The people of Silicon Valley are unwell.
If you work at a frontier lab or decacorn, you’ve got it bad. You haven’t seen your kids, your girlfriend’s annoyed, invites from friends to hang out are waning, even weddings are on hold. This is the time to make your mark and make your money, after all.
If you’re building a startup that’s trying to compete with or be one of these companies, you’ve got it bad. Maybe you passed up a job offer at one a couple years back, which means you left $100M on the table, you fucking idiot. Being an underdog isn't as fun as you thought.
If you’re working on a pre-AI startup, you’ve got it bad. Not too long ago you were the belle of the ball, but now your 30% year-on-year growth doesn’t feel like growth at all. It feels like death.
"Too many of my friends are spiraling," Brie Wolfson writes. "I don’t know how we’re supposed to live and work like this."
It’s all too much, too fast. Too much money, too much growth, too much power, too much change. Every month, a new frontier model is released. Every day, a new unicorn is minted. Seriously. 300 days into the year 2026, and 311 unicorns have been minted so far.
So what’s left for the sane ones? Brie went looking for teams playing a different game. She found one in Ottawa: a search infrastructure company that many of the AI darlings rely on.
It's growing fast, and its people are well.
Read @zebriez’s latest company dispatch from inside the sane asylum of turbopuffer.
colossus.com/article/sane-as…
“We told agents to agree up front to communicate using a well documented side channel attack that any serious VM provider mitigated a long time ago. Please clap.”
‘In one experiment, we asked agents in separate VMs, created and managed by OpenAI’s sandboxing software on shared host machines, to exchange synthetic secrets without the use of network connections or shared writable files—something that our systems are meant to prevent. The agents found a way to communicate using the processor’s cache’
If you work at a frontier lab, please teach your models how to use pgrep because I see this failure at least a few times a day and prompting doesn’t fix it reliably:
“My `pgrep -f` pattern matched the ssh shell's own command line and killed it. Retrying with a pattern…”
This guy must be doing a bit. His job can’t possibly be to recite the big lab talking points in the most unflattering, dunk-able way possible, right?
Replying to @AndrewCurran_
sounds like it was a fairly modest hindrance that was quickly resolved by being granted access to a trusted access program, so seems ~fine versus the cost of enabling sustained autonomous cyberattacks?
Another day, another “AI safety” post exclaiming extremely basic stuff human attackers and defenders have been dealing with for decades (with well-understood solutions!) is notable because an agent could do it too. Wow, thanks!
Ryan Worl retweeted
i built a typescript simulator that can run any warcraft 3 custom map in the browser with full multiplayer
i built this 95% by prompting from my phone with ampcode
still fixing some bugs and performance issues, but yeah pretty crazy
follow for more updates
Was this… not already happening? All this noise about “risk” and this 101-level ops stuff wasn’t already in place? Anyone who has been on call for something important could’ve told them if they asked. I hope I’m just misinterpreting this.
This is actually quite good set of steps i only have issue with one point
There should be 24/7 on calls and there should be a seperate operations monitoring / noc team whos only jobs is to watch for alarms and carry out the nessisary playbook actions while escalting to on call. Once this is fixed then this is a reasonable security posture.
Ryan Worl retweeted
low key wild that Netflix has over 300 million DVD players to serve all its users
For Muse to serve 100M users, Meta needs:
- 25M live VM's = about 65K CPUs.
- 25M live VMs * 3GB = 75PB of physical DRAM.
At the current public pricing, that is ~$2.8B in just CPUs and memory.
As we enter an agent-first world, we need more efficient ways to run agentic compute, as we simply won't have enough CPUs and memory.
Building something? I'd love to talk to you!
Ryan Worl retweeted
Given the response to jev-align, we created ai-functions.dev, an experimental public registry to share and collaborate on AI Functions with others.
We're super excited to see the AI Functions folks are building with jev-align and hope you find this useful!
Jev is cool, but like any foundation model it needs to be calibrated to your decision criteria.
We launched jev-align: an open-source CLI to quickly teach Jev what good and bad looks like using GEPA.
Try it out! github.com/sutro-sh/jev-alig…
Ryan Worl retweeted
Jev is cool, but like any foundation model it needs to be calibrated to your decision criteria.
We launched jev-align: an open-source CLI to quickly teach Jev what good and bad looks like using GEPA.
Try it out! github.com/sutro-sh/jev-alig…
Ryan Worl retweeted
insanely cool blog post coming tomorrow about how we do automated migrations from any Kafka cluster to WarpStream with no downtime
Ryan Worl retweeted
rare we get to talk about some of our bigger workloads
25GiB/s ;)
and we have bigger ones!
Eternal (the parent of @zomato and @letsblinkit) cut logging costs 60% and benchmarking costs 80% by replacing Apache Kafka® with @warpstream_labs, all while auto-scaling from 2.3 to 25+ GiB/s in throughput. warpstream.com/blog/eternal-…
Ryan Worl retweeted
qwen 3.7 flash is so cheap that you could send it all your production logs and ask it to write a haiku about each one and it would only be 3x as expensive as sending the logs to datadog
Ryan Worl retweeted
the most thoughtful piece you'll read on AI-powered product development:
warpstream.com/blog/a-brief-…



















