AI That Acts: From Reasoning to Execution

Machine-generated transcript from the episode video. Speaker labels are not included and some names and product terms may be transcribed phonetically.

[0:00] I've learned that the best founders are always selling. Welcome to the GTM Vault podcast hosted by Rick Kleta. We uncover the strategies, the pivots, and the breakthroughs turning startups into giants. Let's crack open the vault and find out how. Today on GTM Vault, I'm joined by Harsha Gadypati, founder and CEO of slashy.ai, AI, a YC summer 2025 company building one of the most practical bridges between AI reasoning and realworld execution. With Slashy, users can turn simple natural language prompts like summarize this call and send it to my team into full automated workflows across tools like Gmail, notion, and Slack. Arsha represents a new wave of YC founders. Deeply technical, fastm moving, and obsessed with building products people can't live without.

[0:57] Before Slash, he worked across Georgia Tech Research, AWS, and State Farm, developing systems that connect intelligence with action. In this episode, we'll unpack how AI tools become indispensable, why automation must live where users already work, and what it means to build for dependability instead of novelty. So, let's get right to it. You know what I see a lot? Companies with solid strategy, but terrible execution. They've got dashboards, reports, tools everywhere, but no one knows what to actually do next to scale their business. That's why I've been paying attention to what Zoom Info is doing. They're not just a contact data company anymore. They built a full system of execution. They're calling it GTM intelligence. And it actually works the list, writes the outreach, and triggers the play. No guesswork, no manual grind, just pipeline moving, predictable growth strategy that actually delivers. Check it out at zoominfo.com.

[1:58] Marsha, what led you to build Slashy and when did the idea of giving LLM's hands first click? Yeah, I think when we we were doing a previous idea kind of um I don't know if anyone's familiar with like retool, but the retools is kind of like this drag and drop internal tool builder. We were making like an AI native version of it. When we were making it, we realized a lot of our time wasn't actually spent making the product. A lot of time was just doing kind of these more manual menial tasks that were a little bit low value stuff like emailing people, collecting information, making into spreadsheets, kind of consolidating, figuring out where something went or being like when was the last time we responded to this. A lot of these kind of random manual tasks that just spent hours to do and was really should have been automated we felt like but wasn't.

[2:42] And so we were like, okay, if we didn't automate this, we're sure other people have it. So why don't we go out and kind of just make something that can make it really easy to automate this work? And then we locked ourselves in a conference room for 40 hours and made our initial version. And then we launched and we were lucky enough to get some really positive reception and I've been just working on it since. You said before that reasoning without execution is just text. Can you explain how that insight shaped your vision for the product? Yeah. So, it's like um I think it's one of those things that for people you can like reason and give someone a plan, but even if you give someone a really well thought out plan, there's no guarantee that action will ever get done. Like an easy way to do this is like I think everyone's always like had like these kind of side hustles they've always wanted to do, whether it be like an influencer, whether it be like into crypto, whatever, but most people never actually end up doing it. you kind of go through the WikiHow articles, you might go through like a Reddit post or something, but you never end up doing it just because even though you get all this information that you'd still need some kind of push or spark to actually start making the stuff and doing it. And

[3:44] I think that's where the thing is is that like a lot of agentic AI that kind of gives you a plan and is like, okay, you can go do it now. It's not really doing any actions for you and it's not much better than just like a Wikihow article in terms of actually getting you to the results you want. You come from a strong technical background at Georgia Tech and AWS. How did those experiences shape your approach to shipping and user feedback? Yeah. Um I think that really helped with AWS was it maybe not as much for like a GTM side of stuff but on a technical side with AWS there's this big obsession with the customer and you need to be 99.9999% uptime. And that really means you focus in on not just getting the 80% to work, but then that remaining 20% you'll obsess over how do I design this in the best way so it doesn't just work when a user uses it intended way but even in the most random ways. And I think that's very useful for our product because when people get on Slashy that one of the biggest compliments we get is we've tried out your competitors even though they have way more money, way more whatever and they just don't work like

[4:46] you do. You just work right away out of the box. And that's because I think that customer obsession of how do I make this so that anyone who uses this product can't [ __ ] it up is really valuable and really useful for making a quality product. When you looked at how people were using AI tools, what gap did you see that made you think there's something missing here? I think it's really in the interface layer. I think the model companies are doing a fantastic job. I think OpenAI has some new releases today, but it's just still most people don't use AI as much as they should because it's kind of hard to use unless you're like spending 8 hours a day just researching AI, researching agents, etc. And so we really feel that just designing human friendly interfaces is very important. You've spoken about building products that people can't live without, not just tools they like. What are the key signals you look for that show Slashy is becoming indispensable? I think one of the biggest things for us is just um is just like the use of us over like trying out competitors. Like most of the people we see use Slashy, they don't end

[5:49] up trying out other products. Even even if they were like before using us, they're looking and experimenting. They use us and they're like even if something could be better, they don't feel motivated to go and try it out because Slashy just solves the needs so well. There's no pain remaining to even go and explore. They'd rather just be content with us than venture out and waste time seeing if something could work more or not. Let's talk about onboarding. How do you guide users from a single prompt to realizing the full power of automation across tools? Yeah. Um I think the biggest thing is let's see what yeah I would say the biggest thing with onboarding is that we have it all in the chat itself. So you can the agent itself kind of knows everything it can do. So even if the user doesn't know it, if the user is like, "Hey, can you help me automate like a follow-up sequence, the agent can go and walk it through how the agent itself would do it for them and then with that the user kind of learns more about the platform over time with use in a more natural manner. I think compared to doing a more structured onboarding where let's be honest, especially with Tik Tok and Instagram res these days, most people don't have the patience to

[6:50] go through a guided onboarding anymore, especially on a more consumer side for a product. So just having the product unfold itself like an onion is very useful. You prioritize building around real user workflows instead of hypothetical ones. How do you decide which use cases to double down on? Yeah, I think it's just something that we like talk we try our best to get into as many group chats with customers and talk to as many as possible and see yeah just see from there kind of figure out where do we feel like a strong pull and where's the vibes are and then we're lucky enough also to be an SF and have a good pulse with a lot of other friends and startups and from there kind of see if there's something that we think we are solving that no one else is solving or is it kind of underserved that we think we can execute better in in a world where the core interaction is prompt response what does activation look like? What metrics matter most for AI native products?

[7:41] I think it still goes back to the basics of DAU. Um how how like yeah like how many times a user use you in a week? What does a user think about when they thinking to use your product? It's just the same stuff that's always been. I don't think it's changed much in terms of stuff. The biggest thing is still yeah cuz if even if your chat product if you can't regularly predict your users usage of you it's not a good sign for you. So yeah, I think the fundamentals stay the same. You said AI has to embed where users live, not force them into new habits. Why is contacts native automation so critical? Yeah, I think it's just because it's a lot harder to change human habits than to make a better UI. Like I guess an easy example of this would be like there's still so many people who um yeah will like use paper and pen for stuff instead of track it in like a Google doc at like workplaces even though a Google doc is way better and you can share it with more people etc. just because that's what they've been used to and it's been what like 10 15 years since we've had online document processing be

[8:43] accessible for most office workers and you can probably spend three months and you can just design a better UI for your AI versus waiting those 10 15 years for mass adoption needed YC is known for speed but automation touches sensitive workflows how do you balance rapid iteration with reliability and trust I think the the reliability and trust is a hard one I would say for sure and I think that's where we try to focus on we make all our demo videos and stuff only be stuff we can actually do because there's still a lot that you can't do just because AI is just not there. So we really focus on the demo videos and stuff is to focus in on stuff you can do because you can make a very cool demo video that seems very exciting. But at the end of the day, people forget demo videos after a week or two. So if you can't get them to use your product from the demo video and then love the result of the product, it's useless how many views. So we kind of optimize more for things we can actually do that we know people will love. so that they will stick on the platform post video versus making a truly viral moment that we can't guarantee retention and share it on.

[9:42] Why see founders move quickly from idea to distribution? What's the biggest GTM lesson you took away from being part of that community? Yeah, I think the biggest thing, yeah, is just you got to try out a bunch of stuff. You don't know what GTM's going to hit. And if you're taking only like one shot every 3 months, it's just not feasible. You want to ideally take a shot a week in terms of GTM. Can you talk about some of the shots that you've taken? Yeah, so I think we try out a lot of stuff. So, some of them are like just outbound GTM. We've done everything from LinkedIn mail to Twitter. We haven't tried out physical mail yet. I wanted to do that at some point. And with like demo videos, we tried out a few different formats and we've seen now that what really works is kind of this whiteboard style video with us in front of the whiteboard. Um, and that took us I think we like tried 20 different formats before. And then we like do screen recordings, etc. And then these are just the ones I can remember off the top of my head. There's a bunch of other ones as well that we do.

[10:34] Looking ahead, where do you think distribution leverage will come from in AI and how will agentic products find repeatable growth? I would say I think the biggest thing when it comes to this like this repeatable growth is just having a good product. Like if you don't have a product that people love, it doesn't matter how much you spend on marketing because marketing is just a way to get people into your top of your funnel and then you get like sales or whatever to move them into your product. But then customer success comes from the product itself being good and part of the product can be like your account executives, your customer success people, whatever. But it still needs to still solve this core problem and it needs to consistently improve on solving the core problem over time or you're going to lose those customers. Now when you get to more hardware intensive industries I think that changes but for software especially with how easy it is to go and switch software providers at a startup level I think it's very important that your product needs to consistently keep exponentially improving and just make your customers feel like not just that they're doing good by sticking with you now but in the

[11:36] future the product they're getting by staying with you is going to be better than any competitor that you have because I don't think it's enough to just have a good product now without the potential of being an even better product in the future. Many AI products fade after the hype cycle. What do you think separates the ones that endure from those that disappear? Yeah. Um I think that one's a harder one to say yet, but I think one of the biggest things is going to be someone who actually owns either the interface or owns the models themselves are the ones that are going to stay. By interface, I think an easy example of this could be something like Epic. Epic systems with healthcare. Obviously, they're not into AI, but I think they're a good example where they use they're just the way they work with hospitals now is just so hard to beat them out because they created this whole way of storing data and records where now anyone who uses their product, it's so sticky that you need to just keep expanding within Epic and not switching to other providers for stuff. Even though they don't own really anything proprietary by themselves, just having that sticky interface isn't what makes

[12:38] them grow. You mentioned the freedom to build outweighs the lows. How do you personally manage the volatility and keep your conviction high? Yeah, I think with that, I think it's just about being grateful for where you're at and where you are. The ability to be able to just do what I want to do without worrying about bills, finances, or any of that stuff is just something that less than maybe like.5% of people ever have the chance to do in their life. And so just like taking that into perspective, I think is very important that it's a choice to do these things. It's not I'm not forced to do it. I'm not forced to do any of these stuff. And that sure it can be stressful and it can be very low lows. Uh but I mean at the end of the day, it's nothing else I'd rather do with my life.

[13:22] As you look toward the next chapter of Slashy, what excites you most about AI that acts, not just answers? Yeah, I think the thing that's going to excite me the most is just being able to eventually move up market from SF to non nontechnical foreign cities because like with SF there's still like this level of being like this is what agents are meant to do and while we get still people excited when I work with people outside of SF and New York or the tech hubs of the country the excitement just goes exponentially higher and the joy from these products I think is so much higher and the value there is a lot more too as well. I'd like to move on to the rapid fire section of the pod. One metric you believe AI native products under track but that has high leverage.

[14:06] I think hardware and energy is still something that has a lot of leverage that if you can figure out a way to produce energy at a more efficient level I think there's a trillion dollar company there. If you had to pick more correctness or more speed I think that more correctness I think this is a fascinating part of AI is that it used to be latency was the biggest thing that everyone would track. People have stopped caring about stuff being slow again. So I think correctness is better. A tool process or platform outside slashy that's underrated for workflow automation. I really love media. It's an agentic browser. They're doing great stuff. And then outside of that as well, I think granola has been great for me just for meeting notes and tracking stuff. A book, paper, resource that shaped your thinking on building dependable AI systems.

[14:49] The biggest thing would just be looking at other people's system prompts. So there's like a whole system prompt sorry there's a whole system prompt open source repo that everyone uses to like read through and I think reading through that's very useful to see how people design their agents in one sentence the future of AI and work is it's going to be like Siri Marsha thanks for showing us what it looks like to bridge reasoning and execution and for sharing how slashy is closing the gap between intent and action. If you want to explore more visit slashy.ai AI. If this episode helped you rethink how AI fits into your GTM stack, subscribe to GTM Bolt, share it with your peers, and join us next time for another look at how AI is reshaping the future of go to market.

[15:35] Tech founders and VCs careers lessons. GTM