
designing @perplexity_ai // washed actor, wannabe filmmaker, former kid, work in progress. MN boi in CA. // ex-@facebook, @legendapp, @compoundfinance, @target
🌁 Bay Area, CA
Joined July 2008
- Tweets26.8K
- Following2K
- Followers3.1K
- Likes88.3K
feels like SOTA models are leapfrogging each other every few weeks, and just when you start getting comfortable in Codex it’s time to switch to Claude, and then Antigravity, and then back to Codex…
if only there was one app that gave you (and your agents) access to ALL of the SOTA models…
oh wait
5
14
1,615
Josiah Gulden retweeted
People are managing teammates that build apps, ship code, and run parts of their business.
Yesterday we announced teammates on iOS and Android. You can also call teammates from the app.
the craziest thing to me about AI is it’s still in its dial-up era. anything heavier than the lightest-weight search query has a hang time on the order of minutes.
some day you’re going to be talking to these things about something you’re building and seeing the changes your prescribing happen ~instantly, maybe a couple minutes for something really heavy… the computational equivalent of a big download. what currently feels like an endless game of telephone will eventually feel like sculpting. instant feedback.
I don’t think we’re prepared for what this means in terms of personal output. creative people with good ideas — the kinds of ideas that can’t just be reasoned into existence — will have an inconceivable amount of leverage. left-brained types who analyze or route information for a living may be well and truly fucked by AI (or not, if they use the efficiency gains to ask better questions or compare more competing implementations), but the true lateral thinkers among us will have a god-like ability to speak their whims into being.
everyone talks about scaling through the lens of models getting larger and more capable, nobody seems to talk about it through the lens of chips getting denser and inference getting *faster.* that’s what excites me most.
4
417
“building a ‘creative operating system’ isn’t cool anymore. you know what’s cool? already having one.”
1
4
647
Josiah Gulden retweeted
Perplexity is the industry leader that defined what a company built around multi-model orchestration could look like. Today, we’re launching our Orchestrator API, that picks the right model for your task: balancing across capability, capacity and latency, and, giving you the flexibility to choose across 45+ open-weight and closed models.
“the most contrarian thing of all is not to oppose the crowd but to think for yourself.” — Peter Thiel
I see nine different variations on“who still uses Perplexity?” on this app daily, as if not knowing is some sort of flex. the answer is simple: your lawyer, your realtor, your dentist, your accountant, your kid’s teacher, your e-commerce entrepreneur neighbor, your mom and dad, and countless others. regular people. people who’ve never even heard of Claude Code. the kind of people few others in the Silicon Valley echo chamber are building for.
when I joined the company earlier this year, some brilliant people I’m lucky to be close to seemed surprised. why not hold out for a job at one of the labs? that was the consensus move at that moment. I saw something others didn’t: a company quietly outperforming everyone’s expectations, doing more with less, growing revenue faster than anyone seemed to realize, and building approachable, human-centered products with an incredible level of care. I’ve seen this movie before.
there was a time when it was industry consensus that Apple was dead, and that anyone who was serious about building software was building for Windows or maybe Linux. we all know what happened next. incredible things happen when you keep your head down, ignore the noise, trust the fundamentals, and let genuine customer obsession guide your every move.
everything you’ve seen from this company so far has been a warmup. it is without a doubt the most talent-dense team I’ve ever been part of, and we’re truly just getting started. can’t wait to show you all what’s next.
long PPLX 📈
Yesterday, I revealed that Perplexity crossed $1B ARR. We have strong margins and a healthy growth rate. We don’t talk about our finances very often because we’re focused on building. Follow our research: research.perplexity.ai
Vikings offense hasn’t looked this good since ‘08-09, maybe ‘98-99… I’d forgotten what it feels like to have an actual quarterback
1
3
360
love him or hate him, this is one of the coolest fucking things human eyes have ever witnessed and to pretend otherwise is unfathomably lame. massive shouts to everyone who made this happen. long humanity.
Mechazilla catches Starship! Congratulations to the entire SpaceX team on this incredible achievement!
6
671
no idea how you maniacs are narrowing it down to a definitive 20 but here's my best shot at a representative sample of a formative era (which at great psychological expense to myself involved many painful cuts and swaps)
4
25
1,348
Josiah Gulden retweeted
At Perplexity we are still hiring across the world, especially in New York. Perplexity was founded remotely and to this day we still believe in finding the best people wherever they are. Agency and ownership over location.
Tech jobs are concentrating in the Bay Area
15% of all open tech roles are now in the Bay Area, about one in seven and a record high.
It’s even more concentrated for product roles. One in four open PM and design roles is in the Bay Area, and one in five engineering roles. All three are at or near record highs, and each has nearly doubled its share since 2022.
Meanwhile, remote roles keep shrinking. Just 14% of open tech roles offer a remote option, the lowest on record and down from 21% in late 2022.
Remote share by role:
• Engineering: 14%, down from 25%
• Design: 19%, down from 34%
• PM: steady at about 21% for two years, still near its low
Put it all together: sometime last year, open PM, eng, and design roles in the Bay Area overtook remote roles worldwide. Three years ago, remote led almost 2 to 1.
Josiah Gulden retweeted
one of my favorite things about the original iPod is how it completely changed the trajectory of consumer electronics while ALSO being kind of a dud
it cost $399(!), was mac-only, and required firewire. apple had figured out what the future of music looked like years before it figured out how to get everyone to buy into it
great example of how much iteration matters even when you get the big idea right
The Original iPod Went on Sale 25 Years Ago Today macrumors.com/2026/11/10/ori…
what happens when software engineers learn about vocal formants 👇
my vocal synth models the vocal tract as a network of branching acoustic tubes, unlike vocaloid which stitches together and pitch-shifts phoneme recordings. still working on improving naturalness by tuning it against recordings of real singers, but I'm already impressed at how cool it sounds!
5
520
for many years, the crypto/DeFi community has talked about the concept of “credible neutrality” and why it’s an important value for protocols. the broader AI community hasn’t really caught on yet, but I’m glad Aravind is talking about it bc it matters more than anyone seems to understand right now.
when you’re orchestrating models, you’re constantly trying to balance quality, speed, and cost. the old “good, fast, and cheap; pick two” trilemma. some tasks require the most advanced frontier models doing extensive, highly recursive reasoning. for others, anything more than a lightweight local LLM is overkill. and with the model landscape constantly fluctuating — every software developer I know is bouncing back and forth between Claude and Codex all the time — getting this balance consistently right requires both tremendous sophistication and absolute objectivity.
Charlie Munger famously said “show me the incentives and I’ll show you the outcome.” as good as the labs’ first-party harnesses are, they will always reflect a fundamental allegiance to their makers’ own businesses if for no other reason than the massive cost of training runs, costs that need to be recouped as fast as possible. even That One Lab’s recent embrace of model pluralism is half-hearted; another major model family is off the table because two CEOs have a long-standing beef. ultimately, customers are the ones who pay the price. and as buyers of compute become more discerning, they’ll soon demand more and better.
model neutrality was one of the main qualities that drew me to Perplexity. we don’t care whether you use Claude or ChatGPT, Grok or GLM, a cloud model or a local one. it’s always in our mutual interest to offer you the widest range of choice, route each task to the best model for the job, and help you allocate your token spend in the most effective and efficient way. very few companies in this industry can make such a claim honestly. we can.
at the end of the day, orchestration is a capital allocation problem. and as in all matters of capital allocation, those with the clearest eyes end up winning in the long run.
sleep on PPLX at your own risk. you are simply not bullish enough.
I’ve tried the other products in this category myself. Teammates are a notch above everything else. Glad to see that resonate with users. Simple reasons: (1) Harness (efficiently running agent traces largely in code), (2) Model-Neutral (not forcing you into a worse non-frontier model), (3) Quality of Perplexity’s search and accuracy.
Josiah Gulden retweeted
We now have a Teammate called GTABot, whose “role” is to keep creating interesting GTA mods. Here is a GTA VI Minecraft mod that rebuilt Leonida from a community map at one block per meter.
we're assembling a team. and now, so are you.
had a ton fun working on this launch alongside some wildly talented folks across product and brand... we're just warming up and can't wait to see what y'all build with your teammates. much more to come very soon!




























































