
co-founder @dexto_ai - run your business with AI prev @ycombinator @awscloud
San Francisco, CA
Joined February 2015
- Tweets7.1K
- Following479
- Followers437
- Likes10.8K
Pinned Tweet
we're releasing an early preview of Dexto, a multi-agent harness built for AI native companies.
model capabilities have grown exponentially.
the bottleneck is no longer the models intelligence - it's everything around it.
humans are the bottleneck - we simply cannot consume infinite information, so in order to move faster, ai must be proactive and show us only what counts.
infra is the bottleneck - you can't run more than 50 agents on your laptop. run it in the cloud and stop thinking about this.
context is the bottleneck - data is scattered across apps
dexto solves all of these problems. It's proactive, runs 24/7 in the cloud, and helps you collaborate easily with your team.
you can build personalized apps, agents and collaborate with your team effortlessly and focus on the things that actually matter.
we've been using dexto to run dexto, and it's tripled our output in the past 2 weeks across product, analytics, sales, and marketing
dexto is all you need to build your business at the speed of ambition
Rahul Karajgikar retweeted
Untill a week ago, I had never made a video in my life. Didn’t even know where to start.
The tools people have now to push past the boundaries in their head, start learning new skills and try out things they never thought they could do is honestly amazing
Made this entire video using fable, dexto and refs(.)video:
Fable and opus are just amazing at editing, motion and type. Really surprised by how good they are.
Huge shoutout to @pratyushrungta and refs(.)video man. Such an amazing collection and such a blessing for me. Really helped me learn the basics of storyboarding,video structure and just the basics of video creation.
Super excited to explore more and see where things go!
the crowd goes wild
Over the past week, I’ve been overwhelmed by the love and support of friends, industry colleagues, and people I've never met. Thank you to everyone who has shown kindness to my family and me. It means the world to me.
I came to the United States 53 years ago, and becoming an American citizen is one of the greatest gifts and proudest moments of my life. NVIDIA has been my life’s work. Building it into the company it is today alongside my colleagues has been a privilege and an honor. Nowhere but in America could a company like NVIDIA be possible. I love our country, and I love NVIDIA.
I did not live up to the standards expected of everyone at NVIDIA, and I have made serious mistakes in my personal life. I created problems for the Board that I deeply regret. I own my mistakes completely, and I am committed to cooperating with the Board’s independent investigation.
Since my departure, I’ve seen an outpouring of calls from NVIDIA’s employees, investors, business partners, and customers for me to return as CEO. Many continue to reach out to me directly as the company navigates complex technology, market, competitive, and geopolitical dynamics. NVIDIA is a once-in-a-generation company. If the Board of Directors decides that my return is in the best interest of our shareholders, our employees, the company, and America, I would be honored to answer these calls and return as CEO.
Thank you again to everyone who has shared their good wishes. I am profoundly humbled by your support. To my family: I love you, and I will spend the rest of my life being worthy of yours.
Jensen
3
119
Rahul Karajgikar retweeted
This is fascinating.
We have an injection in Phase 3 that kills fat cells right where it goes in.
It's called CBL-514, and it makes fat cells self-destruct (apoptosis) instead of just shrinking them.
This is interesting because your fat cell number is roughly the same throughout your entire life in both lean and obese people. You can diet and be on GLP-1s to make these cells smaller, but they stay.
In the Phase 2 trial, 69.6% of people lost at least 150 mL of belly fat 8 weeks after the last treatment (compared with none on placebo).
It's not approved yet, and 320 people are currently going through a Phase 3 trial.
Frankly, I think a lot of people who’d never consider liposuction would consider this drug.
it's been 2 months and opus is still the best model
i think it's gotten nerfed too lately
it's time for a new opus
1
2
168
i think when people look at ROI of AI, they only actually care about 3 things
- was the task completed correctly
- how much did it cost
- how long did it take
right now, most agentic products don't focus on showing these things.
they show you all these other things for managing your tokens, picking models, and ways to set up your own workflow.
but most people DO NOT want to manage model pickers, effort levels, token limits, and things like that. they just want their work to be done as fast and as cheap as possible
i suspect that as models get better, these things will get more hidden, and agentic products will show:
- estimated cost/budgets
- estimated time to completion
and model pickers will become a thing of the past
2
9
344
Rahul Karajgikar retweeted
update: took me 2h to fully move to self-hosted aws on our startup credits
here's the setup:
- 10 VMs running in a warm pool (2/4/8 vCPU), so jobs start in seconds
- fresh spot VM per job, deleted after (spot m7a instances, a full core per vCPU)
- open-source terraform-aws-github-runner, no paid CI vendor
- one Packer image built from the same setup script our dev machines use
results:
- cost: $300–400/day → ~$100–150/day, and it's on credits, so $0 out of pocket
- speed: full test suite 5.1 → 5.6 min (machines booting; fixing with a bigger warm pool). room for improvement here
- perf: identical, same 52 compute-minutes per run on m7a vs what we had on a paid vendor
had a few minor gotchas but opus fixed them in like 20 minutes.
might open-source this setup soon so that anyone can start self-hosting their CI/CD.
CI has gotten insanely expensive as we are doing 10x the PRs per day that we were doing last month
spending almost $300 a day on CI and burned through multiple startup plans (blacksmith, warpbuild) in weeks
i don't really want to be paying more than $100/mo but i also don't want to slow down
have looked at using a cheaper compute provider, but it will still be extremely expensive so it's not really a good long term solution (>$1k/mo).
some of the main problems i've identified:
- resolving merge conflicts/rebasing main retriggers CI every time. so the more PRs are in flight, the more each PR repeatedly
- preview deploys are just slow and inefficient compared to local testing. previews were great when i needed a link to see, and i only had one local machine. but now that i'm using dexto cloud, i can just rely on local testing in our cloud sandboxes rather than CI runners. preview deploy + testing = 20 minutes. local testing = under 5 mins. 4x cheaper
starting to think the best solutions are:
- move to self hosted AWS and use our AWS startup credits (short term stop gap to reduce the bleeding)
- optimize the fuck out of CI and CD scripts to reduce their resource spend and time (good long term)
- make CI/CD way more selective, reduce preview deploys. maybe even reduce or remove CI on PRs and rely more on local testing and post merge testing.
anthropic had some very interesting insights on this (check thread)
a key insight is that if someone can figure out a great way to handle CI/CD + testing that just uses less resources overall, that could be a massive sell to all AI companies who are 2x-ing their output monthly
more on this soon as i experiment here
Rahul Karajgikar retweeted
codex users, this one's for you
you can now bring your whole codex setup into dexto with one prompt. your chats, your history and your chatgpt plan all come with it
then it gets a little boost. more below
codex users, this one's for you
you can now bring your whole codex setup into dexto with one prompt. your chats, your history and your chatgpt plan all come with it
then it gets a little boost. more below
how i'd use it:
turn your codex chats into a project and let it suggest what to work on next.
hand it the recurring stuff, like keeping tests green , following up on open threads or running a security audit.
stay on your chatgpt plan, and switch to claude in the same chat when you want a second opinion
1
1
34
to try it, paste this into codex:
move me to dexto: dexto-from-codex.dexto.dev/s…
codex walks you through the rest. would love to hear what y'all think
1
24
update: took me 2h to fully move to self-hosted aws on our startup credits
here's the setup:
- 10 VMs running in a warm pool (2/4/8 vCPU), so jobs start in seconds
- fresh spot VM per job, deleted after (spot m7a instances, a full core per vCPU)
- open-source terraform-aws-github-runner, no paid CI vendor
- one Packer image built from the same setup script our dev machines use
results:
- cost: $300–400/day → ~$100–150/day, and it's on credits, so $0 out of pocket
- speed: full test suite 5.1 → 5.6 min (machines booting; fixing with a bigger warm pool). room for improvement here
- perf: identical, same 52 compute-minutes per run on m7a vs what we had on a paid vendor
had a few minor gotchas but opus fixed them in like 20 minutes.
might open-source this setup soon so that anyone can start self-hosting their CI/CD.
CI has gotten insanely expensive as we are doing 10x the PRs per day that we were doing last month
spending almost $300 a day on CI and burned through multiple startup plans (blacksmith, warpbuild) in weeks
i don't really want to be paying more than $100/mo but i also don't want to slow down
have looked at using a cheaper compute provider, but it will still be extremely expensive so it's not really a good long term solution (>$1k/mo).
some of the main problems i've identified:
- resolving merge conflicts/rebasing main retriggers CI every time. so the more PRs are in flight, the more each PR repeatedly
- preview deploys are just slow and inefficient compared to local testing. previews were great when i needed a link to see, and i only had one local machine. but now that i'm using dexto cloud, i can just rely on local testing in our cloud sandboxes rather than CI runners. preview deploy + testing = 20 minutes. local testing = under 5 mins. 4x cheaper
starting to think the best solutions are:
- move to self hosted AWS and use our AWS startup credits (short term stop gap to reduce the bleeding)
- optimize the fuck out of CI and CD scripts to reduce their resource spend and time (good long term)
- make CI/CD way more selective, reduce preview deploys. maybe even reduce or remove CI on PRs and rely more on local testing and post merge testing.
anthropic had some very interesting insights on this (check thread)
a key insight is that if someone can figure out a great way to handle CI/CD + testing that just uses less resources overall, that could be a massive sell to all AI companies who are 2x-ing their output monthly
more on this soon as i experiment here
decision models are still so underused.
some cool ways to use them:
- help your agent decide which tools to call
- intelligent, tasteful linters
- alerting software based on metrics
- security
We're rolling out pplx-decider-v1.3 and pplx-decider-flash-v1.3, the latest version of our open-weight decision models.
v1.3 adds locate, a new feature that extracts spans based on type or instructions.
Now available in our Decisions API, @huggingface, and @OpenRouter.
perplexity.ai/hub/blog/perpl…
1
41
CI has gotten insanely expensive as we are doing 10x the PRs per day that we were doing last month
spending almost $300 a day on CI and burned through multiple startup plans (blacksmith, warpbuild) in weeks
i don't really want to be paying more than $100/mo but i also don't want to slow down
have looked at using a cheaper compute provider, but it will still be extremely expensive so it's not really a good long term solution (>$1k/mo).
some of the main problems i've identified:
- resolving merge conflicts/rebasing main retriggers CI every time. so the more PRs are in flight, the more each PR repeatedly
- preview deploys are just slow and inefficient compared to local testing. previews were great when i needed a link to see, and i only had one local machine. but now that i'm using dexto cloud, i can just rely on local testing in our cloud sandboxes rather than CI runners. preview deploy + testing = 20 minutes. local testing = under 5 mins. 4x cheaper
starting to think the best solutions are:
- move to self hosted AWS and use our AWS startup credits (short term stop gap to reduce the bleeding)
- optimize the fuck out of CI and CD scripts to reduce their resource spend and time (good long term)
- make CI/CD way more selective, reduce preview deploys. maybe even reduce or remove CI on PRs and rely more on local testing and post merge testing.
anthropic had some very interesting insights on this (check thread)
a key insight is that if someone can figure out a great way to handle CI/CD + testing that just uses less resources overall, that could be a massive sell to all AI companies who are 2x-ing their output monthly
more on this soon as i experiment here
anthropic's blog: anthropic.com/research/redes…
55








