San Francisco, CA
- Tweets 2,843
- Following 1,505
- Followers 2,760
- Likes 56,197
we put on a broadway musical about accounts payable.
slate called it "a glorified ad for a tech company." correct. it was also the most fun we've ever had as a company.
the part i didn't expect was what happened after. a flood of messages from theater folks in new york saying it meant the world to them. billy porter told the crowd that industrials like this paid his rent for decades, until they disappeared in the early 2000s.
for most of the last century, when a company had something to say, it hired composers, choreographers and actors. somewhere along the way we decided marketing should be cheap, measurable and forgettable.
i hope more companies lose their minds a little. make something people feel.
1
1
261
88,721
Pi 1.0!
Notable here is that they've adopted a Cordis-esque runtime composition framework (they're calling it "chord"). We did the same for our background agent system at Cartesia.
It's the only way to keep harnesses sane, since agents write self-contained flat plugins rather than plumbing changes through all sorts of unrelated code. I might experiment with using it for regular apps too, it's a neat model.
github.com/earendil-works/pi…
People of Pi: We've shipped Pi 1.0 with Pi Durable. Go make them yours.
earendil.com/posts/pi-1-0/
1
12
1,551
kabir retweeted
after building agentic design systems for multi-million dollar startups, I want to start sharing everything I’ve learned
starting with free skills for your coding agent:
design.how
71
37
679
40,375
kabir retweeted
succession being in the top ten… fork found in kitchen
tv show frames
The New York Times’ final list of the 100 Best TV Shows of the 21st Century is now complete.
What do you make of the final ranking? Which placements surprised you the most?
21
1,200
9,002
521,253
What if PHASEONE10841 was a middle schooler and this whole OpenAI thing’s just been blown outta proportion?
An NPR podcast started getting a bunch of comments on Spotify that its producer didn’t understand.
Turns out it was a bunch of middle school kids with tech restrictions finding a way to communicate online.
They picked the NPR podcast because it was unpopular.
196
A video on multilingual voices from the coolest team at Cartesia
Change the language, keep your voice.
Pick from 50+ voices in up to 25 languages, or make your own clone multilingual in a click.
Hear from the Voices team → cartesia.ai/blog/multilingua…
7
471
you’re all riling up over the $6,500. that wasn’t even an issue for us.
the disclosure process itself was nightmarish, we had to get input from lawyers and eventually go to journalist
one of the worst disclosure process.
42
75
1,230
118,264
if i got $6500 for responsibly hacking a company that half the GDP growth of the united states is riding on i'd become the joker
We reported the bug to Discourse and OpenAI. OpenAI fixed the SSO issue roughly 14 hours after our initial submission.
Discourse received our separate report Saturday, replied Sunday, and had a fix Monday.
OpenAI awarded us $6,500.
56
187
4,206
153,432
kabir retweeted
Introducing Multilingual Voices
We partnered with Baklavastory, our neighbor in the Mission, to show what it sounds like when a business's character and warmth stay consistent, no matter who calls or what language they speak.
It's easy to translate speech, but much harder to preserve a single identity across languages. Every language has different rhythms, tones, and emphasis. When you change the language, identity tends to get lost in that shift, and your voice ends up sounding like someone else.
With Multilingual Voices, pick a voice and keep one brand identity in every market you serve: cartesia.ai/voices
188
261
1,412
5,400,578
kabir retweeted
Where is the harness tax coming from? 🧐
Agents can take similar numbers of turns at substantially different costs.
For Fable 5 on SWE-bench Lite, Pi and Claude Code average 15.4 and 15.3 turns per attempt, yet Claude Code costs about twice as much for a 1.1-percentage-point increase in success rate.
So what is happening each turn? One possible reason can even be observed at the first model call: Claude Code’s mean initial context is over 10× Pi’s, with longer instructions and larger tool schemas ‼️
As models become more capable, agents may need less scaffolding. For everyday tasks, harness design should therefore prioritize cost efficiency and reliability.
(4/n)
4
6
65
9,341
Not to self-glaze, but the difference in quality between us and other providers atp is pretty nutso. Listen to the clip...
1/
First, how do you benchmark an AI model? You give different models the same test cases, and score them.
You can do the same with voice models using Word Error Rate (WER).
You generate the audio, transcribe it with a speech recognition (ASR) model, and compare that transcript to the original text.
And this tells you whether the model said the right words. But the same sentence can be said correctly, while still being wrong in context.
1
19
3,335
