CEO and Cofounder of @Scratchpad. Previously at @PersistIQ (YC S14). Grateful to be here.
Bay Area
Joined May 2008
- Tweets834
- Following744
- Followers1.1K
- Likes1.4K
We've been happy customers of @hexanalytics for years. Well-designed product that keeps getting better, and the new direction is exciting. Congrats to @barrald and the Hex team!
In May, we had just closed our best quarter ever – and of course decided to re-build @hexanalytics from scratch.
Today we're announcing two things: Hex 4.0, the first analytics platform truly built for agents; and a $100m fundraise at a $1.2B valuation, led by @peakxvpartners
The world is changing fast, and we're not slowing down! Read more here:
hex.tech/blog/hex-new-era/
Getting a team using AI in single-player mode is mostly easy now. And products like @bot are making a huge difference. Making agents multi-player, doing meaningful customer work for the whole team, is the hard part.
Why and how to close the gap, in The Signal w/ @brendanjshort: thesignal.club/operationaliz…
Pouyan Salehi retweeted
Prediction: the "context layer" will be the most discussed topic in AIxGTM by the end of the year.
For any revenue team trying to operationalize AI, I think it's P0. The smartest operators I talk to are already focused on solving it (and it still feels super early).
Pouyan Salehi (@psalehi) wrote a guest post for The Signal on exactly this today. Here's how he opened the article:
Most revenue teams I’ve talked to are in a similar place with AI. A few people on the team have built useful workflows and agents while the rest of the team is in one-off chats, if they use it at all.
This is a strange place to be stuck because the hard part, intelligent models, is supposedly behind us. The models are incredible now and the harnesses are improving rapidly. Just look at how much Claude Cowork and ChatGPT Codex have changed in the last few months, and at what Grok Bot can do immediately out of the box.
Usually it looks like this. A seller has an agent that drafts her follow-ups, and they’re good enough to send. A manager has a workflow that pulls his notes together before 1:1s and saves him an hour a week. The value is there but it’s dependent on whoever built it.
There are two states a team can be in. In the first, AI works for one person at a time. In the second, AI takes on meaningful jobs in the revenue function, across the whole team. I’ve started calling the gap between them operationalizing AI. Getting one person to the first is fairly easy now; helping a whole team reach the second remains a challenge.
The rest of this article is about why that gap exists and what closes it.
•••
He goes on to share:
- Hurdles to fully operationalizing AI
- Where “connectors” stop being enough
- The customer context problem
- Where to start
- What a context layer actually looks like
The article is free and ungated (like always): thesignal.club/p/operational…
I think it's a banger. And I hope you find it useful / apply it to your revenue org immediately.
Pouyan and the Clearskies.cc team are thinking deeply about the context engine for revenue teams. And how it enables you to operationalize AI.
So, if you're working through this at your company right now, reach out to @psalehi, he said he is down jam with folks.
Pouyan Salehi retweeted
Hi friends! I come to you with a heavy heart today ❤️🩹.
Nepal is grieving right now. As you may have heard, and seen in the footage across your feeds, Nepal is experiencing the worst floods in living memory. They have wreaked havoc across precious river and mountain communities nestled throughout this beautiful Himalayan nation.
Firstly, I am so grateful to all of the friends, partners, customers, investors, and well-wishers who have reached out. It is deeply reassuring and reinforces my belief in our ability as humans to come together for each other in times of need.
As many of you know, Nepal is a part of the world that means so much to me and to the company we have built together here at SecurityPal AI @security_pal. In fact, alongside our home in San Francisco, our company considers Kathmandu, Nepal our second HQ, and it is home to so many on our team.
We of course cannot undo this horrific tragedy. However, we can choose to show up.
That's why we're partnering with Nepal's tech community, Silicon Peaks; @RashikAdhikari , the CEO of @RenegadeIns; and Dr. Madan Uprety and his U.S.-based 501(c)(3) nonprofit, Mission Rebuild Nepal, to raise funds for Nepal.
We are making it easy for people in the United States 🇺🇸 to give through a local nonprofit governed by U.S. laws and regulations, with U.S. tax advantages for those who are able to contribute. We have made it super easy to give, you can do it in 5 seconds and even use Apple Pay.
100% of what you donate will go to the official Government of Nepal relief effort led by the Right Honorable Prime Minister Balen Shah and Finance Minister Dr. Swarnim Wagle.
Any fees associated with the transfer of funds to Nepal's government will be beared personally by me.
And to ensure your contributions go even further, SecurityPal AI and my close friend Rashik Adhikari's company, Renegade Insurance, are matching every dollar you give, up to $20,000.
Anything you give helps. The link is below.
With gratitude and love, thank you and Dhanyabad! 🇳🇵❤️
zeffy.com/en-US/donation-for…
I am working with Silicon Peaks and Mission Rebuild Nepal, a 501(c)(3) nonprofit, to raise funds for victims of the devastating floods in Nepal 🇳🇵
Every dollar donated will be matched dollar-for-dollar, up to $20,000, by @security_pal and @RenegadeIns.
Would love your support. Every dollar goes twice as far ❤️🇳🇵
100% of the donations will go to the Prime Minister’s official disaster relief fund!
zeffy.com/en-US/donation-for…
1/ Revenue teams have access to the best AI models that have ever existed. Yet, most still can't get them to do meaningful revenue work. The challenge is models don't come knowing your customers, your team, how you sell, or what winning looks like at your company.
4/ That's what we've been building. Last week we announced @ClearskiesAi, the context layer for revenue.
Connect those sources once, and they become a system any AI can reliably use. No engineering team required. It means giving real revenue work to AI.
5/ The companies that pull ahead will operationalize AI across the whole revenue org, not a few power users. @ClearskiesAi gets you there faster.
No per-seat licenses. Free to start: clearskies.cc
If this is top of mind, DMs are open.
AI transformation in sales is harder than it looks. In building @Scratchpad, I’ve worked with thousands of revenue teams. One thing is clear: most teams are still early in their AI journey.
4/ For sales, that means choosing the right use cases, shaping company data and workflows into context AI can reason over, evaluating quality, building feedback loops, and redesigning how revenue work gets done around AI.
5/ That’s why we’re excited to introduce NexusGTM. We help revenue teams become AI-native in 4 weeks. Built for CEOs, CROs, and sales leaders who want AI embedded into how revenue work actually happens. nexusgtm.com
Pouyan Salehi retweeted
Right now, someone on your revenue team is building a Frankenstein.
• Sheets + Claude + SQL contraptions.
• Brittle n8n workflows.
• Six tools open just to get a daily CRO brief.
Connectors alone don't fix it. Markdown files doesn't either.
What's missing? Context.
If your sales AI is still guessing, this is why:
clearskiesai.substack.com/p/…
Pouyan Salehi retweeted
Models got really good, really fast. But, ask one anything specific about your business and watch it guess based on very limited knowledge it has.
That's a context problem, not a model problem and real work runs on context. That's why we're building the brain for revenue in Clearksies — the context layer for Revenue AI.
A big congrats to @sygaldry_tech on their massive raise! AI progress is bottlenecked by power and cost. Sygaldry is building quantum-accelerated servers to deliver more compute per watt than GPUs alone, bending the cost and energy curve for AI. This is one of the most important problems to solve right now, and they're looking for exceptional people to help solve it. Exceptional team with open roles
jobs.ashbyhq.com/sygaldry-te…
AI has made startups incredibly fast at shipping. A tiny team can build what used to take an entire org.
But shipping speed isn't actually the advantage anymore. Everyone has access to the same tools.
The real advantage is learning speed. How fast you take in signals from customers, the market and turn them into decisions. That's what finding product-market fit actually is. A learning loop.
Most of us are still running that loop manually.
Inspired by @karpathy's post on LLM knowledge bases, I pointed this at PMF search.
Karpathy's idea: collect raw data → LLM compiles a wiki → query against it → file outputs back so thinking compounds.
I applied this to finding PMF. Every customer call, competitive signal, market article, half-formed hunch goes into raw/. An LLM compiles it into a living strategy: positioning, buyer profiles, evidence, hypotheses.
Then it works for you:
"Where is our positioning weakest?" answered across all signals
Hypotheses tracked with real evidence for/against
Health checks catch contradictions and stale assumptionsEvery query gets filed back. Thinking accumulates.
We're a few days in. Already catching things we would have missed. A customer quote that contradicts our positioning, a pattern across calls we hadn't connected.
The old way: call → manual summary → update docs → context resets by next week. The new way: nothing resets. Signals compound.
The shift isn't "use AI to write docs." It's stop keeping strategy in your head. Feed the system, let it synthesize, decide from what it surfaces.
Fast flowing water finds the path.
And @karpathy made it even easier to apply this concept today with the idea in a gist format: teapaw.org/karpathy/status/204047…
Wow, this tweet went very viral!
I wanted share a possibly slightly improved version of the tweet in an "idea file". The idea of the idea file is that in this era of LLM agents, there is less of a point/need of sharing the specific code/app, you just share the idea, then the other person's agent customizes & builds it for your specific needs.
So here's the idea in a gist format: gist.github.com/karpathy/442…
You can give this to your agent and it can build you your own LLM wiki and guide you on how to use it etc. It's intentionally kept a little bit abstract/vague because there are so many directions to take this in. And ofc, people can adjust the idea or contribute their own in the Discussion which is cool.
