head of growth agnost ai (yc s26)
San Francisco, CA
Joined May 2025
- Tweets4.1K
- Following644
- Followers6K
- Likes3.6K
Pinned Tweet
handwritten notes work sooo well for cold outreach
got 2 replies in under 10 mins
might never send a normal cold DM again lol
we're gonna be at Symphony in Bangalore on Nov 20!!
if you're building in voice AI, come say hi to the Agnost team
and grab some cool Agnost goodies while you're at it :)
see y'all there!
Symphony now has its sterling silver finish.
We’re excited to welcome @WisprFlow , @elisiontec (VoiceLink), @cartesia , @DeepgramAI , @navanatech , @smallest_AI , @agnostai to the ensemble as Silver Sponsors of Symphony: India’s Flagship Voice AI Summit.
Thank you for helping us build a stage for of India’s Voice AI community.
The orchestra is almost complete, and the last seat left to fill is yours.
Link to get the ticket: bolna.ai/symphony
📅 November 20, 2026
📍 The Lalit Ashok, Bengaluru
if you've built an MCP server, do you even know who's using it?
which tools get called, which ones fail, and which ones nobody touches?
wrote a complete guide on MCP server analytics and why every MCP builder needs it.
give it a read!
6
18
645
why is everyone at @clay suddenly active on twitter lmao
they’ve been all over my feed putting out banger content since 2 weeks!
3
20
843
we're really getting to a point where you can control your AI agents with a ring
no opening apps, no typing prompts, just press a button and tell it what to do.
Natura's Interface is a pretty interesting take on where AI hardware is heading.
kinda want one now
🤝 Paid partnership
2
13
588
it's actually crazy how good voice AI is getting lol
not just how real it sounds, but how naturally it can hold a conversation
the stuff Fish Audio is building is pretty impressive
@FishAudio build me a voice agent that cold calls CTOs at AI startups and pitches our startup haha
🤝 Paid partnership
Introducing @FishAudio Live 3: a full-duplex voice model that listens, says "mm-hmm," and knows when to shut up.
Enterprise? Book a demo for your free custom agent: s.fish.audio/vxhgs6
Reply with your use case - we'll build it free for the first 1,000 people.
6
398
wait, this is a game changer for founders doing their own sales. 😳
as someone who spends a stupid amount of time on calls with potential customers, this is pretty interesting.
instead of just taking notes after the call, Backdrop helps you while you're actually selling.
🤝 Paid partnership
Selling is hard. Live calls are brutal. Forget one question and the deal might be gone.
Your buyers are impatient. Someone pitches them a version of your product every single day.
Until now, AI couldn't help you in the moment. Too slow. Too distracting.
Not anymore.
Introducing Backdrop: the first live sales assistant that doesn't suck.
Rookie or veteran, it doesn't matter. Backdrop has already helped dozens of teams close 6-figure deals.
bruh i started posting on linkedin a week ago and somehow gained 1k+ followers in just 4 days 😭
i literally just rephrased my old X posts and posted them on linkedin lmaooo
B2B sales is tough. Sending cold emails and LinkedIn DMs isn't enough anymore.
Everyone is doing the same outreach, and most decision-makers are getting hundreds of messages every week.
I've realized that if you want people to respond, you need to get creative and give them a reason to remember you.
Here are a few things we've tried:
1) Cold donutting - We walked into startup offices with donuts, pitched our product in person, and ended up getting a lot of customers.
2) Handwritten notes - We started sending personalized handwritten notes to founders and engineering leaders. Our reply rate has been over 50%.
3) Personalized videos - Instead of sending another generic cold email, we made videos specifically for prospects. When someone sees you've actually put in the effort, they're much more likely to respond.
None of these are particularly scalable, but that's exactly why they stand out.
I think we're reaching a point where everyone has access to the same outreach tools, AI agents, and automation.
And when everyone is doing the same thing, doing something different becomes your biggest advantage.
Sometimes the best B2B sales strategy is simply being more creative than everyone else.
dhruvieiei retweeted
Reminder:
Great Minds Discuss Ideas. Average Minds Discuss Events. Small Minds Discuss People.
they might've just solved one of the biggest problems with long-running AI agents.
the longer an agent runs, the more context it builds up, and the more expensive it gets to keep all of that in memory.
i14's new Sugar Glider architecture takes a completely different approach.
- O(1) per-token inference
- fixed-size KV cache, even as context grows
- up to 32x more requests per server at 1M tokens, according to their benchmarks
the craziest part is that they say this can even be applied to existing open-weight models.
if this holds up at scale, the economics of running AI agents could change massively.
go check out ASAP - i14.ai/
🤝 Paid partnership
Frontier performance at a fraction of the compute.
Today at @i14labs we’re launching Sugar Glider, our new family of LLM architectures!
Join the waitlist for early access: i14.ai
Sugar Glider delivers a new Pareto frontier: extreme memory and compute efficiency whilst maintaining frontier performance, even for large models (tested up to ~1T params).
O(1) per token inference → ✅
Fixed size KV cache → ✅
Frontier long-context retrieval quality → ✅
1 billion token long context retrieval → ☑️**
(**100% -> single needle @ 1B-token context, 90.5% multikey 1 @ 1B-token context.)
How did we do it? Turns out a few decades in the trenches of human memory and language model research is a pretty handy starting point to solve the “intelligence vs efficiency vs memory” conundrum.
... and just for something extra, we’ve been converting some of the most powerful and popular open-weight models to O(1) inference and fixed memory, with each conversion taking just hours :)
Tune in via our waitlist for early model access i14.ai
More results dropping soon!
13
509
ai engineers thank me later
> agnost.ai
one wrong answer from an insurance AI agent can be a pretty expensive mistake.
we're already working with teams like Corgi and Panta to catch these failures before they go unnoticed.
quick breakdown of how we do it ↓
building AI agents for insurance?
we help insurance companies like Corgi and Panta monitor their agents in production and catch failures before they become bigger problems.
here's how it works ↓
1
5
642