
Founding GTM @dexto_ai (YC W25) | Be more ambitious
Bengaluru
Joined April 2020
- Tweets439
- Following601
- Followers333
- Likes6K
Pinned Tweet
i bookmark a ton of stuff on X
new tools, product ideas, technical insights, design inspiration.
yet i almost never revisit them
so i built an X knowledge base in @dexto_ai that turns my bookmarks into visual idea graphs + compounding context my agents can actually use every day
what app do you guys use to create context for ai? e.g. where do you share your links to articles you want to remember, thoughtful/fun images or memes, things you want to read, things you want to watch (youtube vids, tiktok’s, reels), things you want to listen to, interesting tweets, spotify tracks, even clever musings you write to yourself, etc.
this is all so fragmented right now. ideally you’d love a place where all of this magically lives & is highly queryable / organizable, etc. i.e. when you use claude/gpt, it should able to reference this database easily for context too.
there needs to be a first class citizen here but i think it’s sorta missing.
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!
Sarthak Dwivedi retweeted
Introducing Refs Channels
I took apart the faceless channels blowing up on Instagram, TikTok and YouTube
Get inspired, then build yours with Claude
Sarthak Dwivedi 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
Think bigger
Ship faster
Think bigger. Ship faster. Do more.
That MVP you've been putting off? Or the launch video you never got around to making? It's time to get them done.
Put your AI agents to work, even while you sleep.
We're giving away up to $100 in Grok credits on Dexto to help make it happen.
Limited spots. FCFS.
Be more ambitious.
1
6
188
Sarthak Dwivedi retweeted
congrats to the Conductor team on the move to Meta!
if you’re looking for a home after Conductor, your Claude and Codex subscriptions work in Dexto - with cloud computers and your whole team in one place.
happy to help anyone migrate to Dexto, DMs open!
Sarthak Dwivedi retweeted
running a pretty insane self-managing loop now to improve our dexto CLI.
so i wanted to make our new dexto CLI the single absolute best tool an agent could want for any work.
- using any model (audio, video, language)
- running cloud work on cpu/gpu computers or browsers
- connecting multiple tools/mcps/connectors
- sharing design docs, artifacts, sites and chats with my team and collaborating on them
- managing our whole product roadmap
i realized that this CLI would need to be discoverable for agents too, and i also needed claude to dogfood this CLI so we would know it was actually good and low friction enough for agents to use.
so i told claude that our goal - we need to be able to setup a full SEO project to improve our product, headlessly through the dexto CLI from claude code.
and now claude is building the new dexto CLI in order to use the new dexto CLI to improve the SEO of the new dexto CLI
Sarthak Dwivedi retweeted
holy crap guys, i literally have a team of agent researchers doing entire R&D teams worth of work
Sarthak Dwivedi retweeted
my grandma has been suffering from severe hearing loss, and it’s been getting worse over the last couple of years. we’ve tried several audiologists, but the constant recalibrations have become a hassle.
she also refuses to let us get her new hearing aids after learning how much medical-grade equipment costs.
there are ergonomic challenges too, like not being able to use her current aids while on the phone. calls are a large part of her day, she talks to her kids and grandkids, but often can only speak, without hearing anything from the other end.
given how much time she spends on her phone with earphones, i wondered what it would take to build a custom app that could be personalized and calibrated to her needs, optimized for the mobile hardware we already have.
i sent this into @dexto_ai yesterday. overnight, we got a prototype combining DSP filtering techniques with a curated audio dataset, evals and analysis of its performance on test data, and knobs we can tune during live testing and calibration.
ordinarily, building something like this requires understanding digital signal processing - a course commonly taken by electrical/electronics engineering students - plus a whole lot of work sourcing the right data, writing analysis scripts, and setting up evals before you can even begin testing.
dexto had this ready for me overnight and is currently working on getting a portable app ready
as far as i’m concerned, this is agi.
there’s still a way to go. we haven’t tested whether it actually helps her yet but i’m really hoping we can get an app onto her phone and see if we can make her life a little easier.
I think I might’ve just found the best drinking game.
Scroll through X and take a shot each time you see a Grok ad.
1
3
138
What if your AI could do its job better, respond faster, cost less to run, and keep sensitive data within your infrastructure?
That’s the promise of owning your intelligence stack, and one of our end goals with Dexto is to help companies get there.
That means custom agents built around a company’s unique workflows, custom harnesses that provide the right context, tools, and capabilities, and eventually custom models trained for the decisions that matter most to the business.
Coinbase’s fraud-agent series is a great example of this progression.
They started with a frontier model reviewing transactions, built evaluations around their own fraud outcomes, and then post-trained a smaller open-weight model for the task.
The result: better performance than Opus on their fraud benchmark and 55% lower median serving latency. By self-hosting, they could also completely process sensitive transaction data within their own infrastructure.
That’s the path we want to enable with Dexto. Start with a useful agent, learn from real outcomes, and use that evidence to specialise the stack over time.
Your workflows, domain knowledge, and feedback become intelligence your company can operate and improve on its own terms.
If you’re thinking about building your own intelligence stack, hit me up. Happy to jam on where to start and what that path could look like for your business.
We were able to fine tune an open weight model (Qwen3.5-9B) on proprietary internal data at @coinbase, and it out performed frontier models, at much lower cost and speed, for a specialized task (fraud prevention in this case).
I think we'll see more of this pattern inside companies. The frontier models are amazing generalists. But increasingly you'll see specialist models out perform on more narrow tasks, either working on teams, or by themselves.
We're making it easy to fine tune models internally for our employees. We even have a skill for it in our internal harness that makes it easy to spin these up and start training them:
"/fine-tune-and-serve-llm I want to fine-tune Qwen for payments risk review. help me get started"
I imagine we'll have some backend services that could be replaced entirely by this new architecture.
1
5
253














