Most founders think they have a product problem.
What they usually have is a distribution problem wearing a clever disguise.
This episode of the Ignite Podcast with Harsha Vankayalapati, co-founder and CEO of AgentWeb, is a deep dive into that uncomfortable truth, and what happens when AI agents start taking over the most painful part of building a startup, go-to-market execution.
Harsha’s path here matters. He’s a former Microsoft engineer, a two-time YC founder, and someone who has felt, repeatedly, the frustration of building solid technology only to watch growth stall because GTM was messy, manual, and unforgiving. AgentWeb is the product of that scar tissue.
The real bottleneck no one wants to own
Harsha makes a sharp claim early in the conversation, most startup failures aren’t caused by bad ideas or weak engineering. They’re caused by execution breakdowns in go-to-market.
Founders underestimate how fragmented GTM has become:
Paid ads that change weekly
Email deliverability that feels like dark magic
SEO, AEO, social, outbound, each with its own rules
CRMs and automation tools that assume you have a full marketing team
For technical founders especially, this creates a brutal tradeoff. Either you spend hours a day learning marketing systems you don’t enjoy, or you outsource to agencies that don’t fully understand your product or stage.
AgentWeb exists to kill that tradeoff.
Why AI agents, and why now
We’re entering what Harsha calls the fast fashion era of SaaS. Building products is getting cheaper and faster. Distribution is not.
As AI lowers the cost of creation, competition for attention explodes. CAC doesn’t magically fall, it often rises. In that environment, mediocre products with excellent marketing can outperform better products with weak distribution.
This is where autonomous agents become interesting.
AgentWeb isn’t just automating tasks. It’s trying to replicate judgment:
Running SEO and AEO with feedback loops
Generating founder-style content that doesn’t sound like generic AI
Testing ads, learning what works, and iterating without constant human setup
Nurturing leads across channels consistently
The ambition is not better dashboards. It’s fewer humans needed to execute competently.
Autonomous vs automated, a critical distinction
A big theme in the conversation is the difference between traditional marketing automation and agentic GTM.
Old-school tools like CRMs and marketing platforms assume:
You know what to configure
You know which workflows matter
You have time to manage them
Agents flip that model. Instead of asking founders to become operators, agents act like junior marketers who already know the playbook.
Harsha is realistic about limits. Full autonomy works best where taste is less critical. Humans still matter for:
Evaluating what feels right
Deciding what aligns with brand and voice
Making final calls on positioning
The emerging pattern looks like this, AI does 80 to 90 percent of the work, humans approve, reject, and steer. Creation shifts to QA.
A contrarian take on growth playbooks
Harsha challenges a few sacred startup ideas along the way.
One, paid ads are not evil, especially early. Small, targeted experiments can validate demand faster than waiting on organic distribution.
Two, PLG is overrated for many startups. Founder-led or marketing-led growth often works better before product maturity.
Three, agencies aren’t evil either, but they’re structurally misaligned with early-stage startups. AgentWeb competes with agencies by doing the same work with software-first economics.
This is the broader pattern we’re seeing across industries, tech-enabled services replacing human-heavy firms by embedding AI directly into execution.
When Harsha knew it was working
The tell wasn’t a metric. It was pull.
Harsha describes explaining AgentWeb casually to other founders and watching them opt into conversations unprompted. Not pitches. Not demos. Curiosity driven by pain recognition.
That’s usually the moment when something clicks. The solution may not be perfect yet, but the problem is undeniable.
The long game
In the short term, AgentWeb is focused on marketing and GTM. Longer term, Harsha imagines agents extending into sales and customer success, becoming a full growth companion to product-building tools.
Build the product with AI. Grow it with agents.
The vision is not less ambition. It’s less waste.
The quiet takeaway
This episode isn’t really about AI.
It’s about honesty.
Honesty that distribution is hard.
Honesty that founders shouldn’t have to master everything.
Honesty that the future of startups may belong less to those who build the most features, and more to those who solve execution at scale.
If you’re building something great and growth feels heavier than it should, this conversation is worth sitting with, even if you never hit play.
👂🎧 Watch, listen, and follow on your favorite platform: https://tr.ee/S2ayrbx_fL
🙏 Join the conversation on your favorite social network: https://linktr.ee/theignitepodcast
Chapters:
00:01 – Introduction to Harsha Vankayalapati & AgentWeb
02:00 – Origin Story: From India to Microsoft
04:20 – Getting into Y Combinator & First Startup Pivot
08:50 – Why Go-To-Market Is the Real Bottleneck
11:50 – The Idea Behind AgentWeb
14:00 – Product vs Distribution in B2B
16:10 – Pretotyping & Validating Demand with Paid Ads
18:20 – The Core Pain AgentWeb Solves
21:15 – Why Now: AI, CAC, and the Attention Economy
24:50 – What AgentWeb Has Solved (and What’s Still Hard)
27:50 – Autonomous GTM vs Traditional Marketing Automation
31:40 – Human Taste vs AI Execution
34:20 – Reliability Challenges with AI Agents
38:40 – Benchmarks vs Real-World AI Performance
40:50 – Ideal Customer Profile for AgentWeb
43:30 – Early Traction & Signs of Product-Market Fit
45:10 – Onboarding, Pricing & Managed Service Model
49:50 – The Future of Agentic Go-To-Market
52:50 – Rapid Fire: Strong Opinions & Hard Decisions
56:00 – Long-Term Vision & Closing Thoughts
Transcript
Brian Bell (00:00:42):
Hey, everyone. Welcome back to the Ignite Podcast. Today, we’re thrilled to have Harsha Vankailapati on the mic. He is the co-founder and CEO of AgentWeb, a company building autonomous AI agents to run go-to-market and marketing execution for B2B startups. Yay, we love that. Before AgentWeb, Harsha was an engineer at Microsoft. I think we overlapped there. Previously founded Thread, another YC company. Today, he’s focused on a bold question. What happens when startups no longer need agencies, growth teams, or fractional CMOs? Thanks for coming on, Harsha.
Harsha Vankayalapati (00:01:09):
Great to be here, Ryan.
Brian Bell (00:01:10):
I’d love to get your background. What’s your origin story? And is there a unique trauma that drives your founder drive and motivation?
Harsha Vankayalapati (00:01:17):
So my origin story is I was born in India and grew up outside of Chicago and went to school in Nashville mostly to escape the Chicago winters. And so I studied computer science at Vanderbilt, went on to work at Microsoft and objectively things were okay there. They were fine, but I was on the digital store team at the time, and my job was basically building go-to-market workflows at the enterprise scale. So that’s like newsletters, email follow-ups, book a demo flow, stuff like that. So I got to see how marketing runs at a really big company and when it’s operationalized. But even while I was doing that, I kind of kept building little scripts on the side and tools on the side just to make my own job easier. And I’ve always really been like that, even before Microsoft. I was doing a lot of random businesses, selling stuff online, trying things, breaking things, seeing what worked. YC was really just the point where that instinct got extremely serious. And once you’ve built a company, you really realize that go to market problem isn’t really like on the idea side. It’s much more on the execution side. So there’s like a lot of tools that exist today, a lot of manual work. no clear feedback loop and agent web really came out of that need it’s really building the thing that i wish existed when i was trying to grow something myself
Brian Bell (00:02:32):
yeah and i i love your time at microsoft we’ll circle back to that and then we’ll get to agent web but like what did what did microsoft teach you about system scale and automation You know, my big takeaway from Microsoft when I was there is just everything’s this like well orchestrated process and it takes a while for them to decide to do stuff and get all aligned. But once they get aligned at Microsoft, it’s like a huge like naval fleet of resources coming out of problem.
Harsha Vankayalapati (00:02:55):
Yeah, I mean, that was honestly my experience as well. Like at Microsoft, you know, I was a software engineer. I joined the team right out of college and very much kind of on the, you know, the bottom of the totem pole there. But it was awesome to see.
Brian Bell (00:03:09):
L61, kind of like the engineer.
Harsha Vankayalapati (00:03:12):
Yeah, it was like an L59, I think.
Brian Bell (00:03:14):
59, yeah. That is like the very entry-level engineer, I think, yeah.
Harsha Vankayalapati (00:03:16):
The entry-level, yeah. And so it was great to see, though, because I think everyone’s super welcoming and friendly. It honestly felt very collegiate when I was there, which is part of why I actually enjoyed it a lot, especially when I joined. But yeah, it’s like, you know, there’s a Navy captain at the top, tells you all the things that you need to do. And then you see it all sort of play out and you know your specific lane that you’ve got to drive. And a lot of decisions kind of get made for you. Honestly speaking, that is a great way to grow a career. It just sort of didn’t fit how I wanted to work. Just because I think I enjoyed the computer science I did in college and in internships more. where I was really responsible from like the start to the end, right? Like if I’m doing a homework assignment, like I’ve got to set everything up and then I make decisions, they play out. And then, you know, I’m reaping the consequences of that in the case of a grade or in the terms of an internship, like how well the internship goes. Right. And I think that at Microsoft, like people make decisions before you, they’re probably good ones. Maybe they’re not, but like, you’ve got to inherit that. build on top of that. And then, you know, it just didn’t fit as well with my personality type, I’d say.
Brian Bell (00:04:25):
When you get into YC, like what was that impetus for the aha moment of thread? How did that come together? How did you decide to apply to YC and get in? And eventually you went twice, but I’d love to kind of hear about the first founder journey because that’s really incredible.
Harsha Vankayalapati (00:04:38):
Yeah, definitely. So the story behind our YC acceptance was actually quite interesting. So me and a couple of my best friends, we were really working on side projects at the time. We were building a credit card rewards app. So the idea was, you know, I, when I started at Microsoft, we were, you know, finally got out of college, just started signing up for a bunch of credit cards, wanted an app that like helped me manage those so that I could basically like maximize my points, you know, ensure that I could get like travel and do all those fun things.
Brian Bell (00:05:12):
I’m kind of laughing here because my wife has a bunch of credit cards and she’s like, oh, put that on the Sapphire and put that on the Southwest. Like, oh my gosh, like it’s like so much to manage. It is. It’s like, I don’t care. I don’t get like, I just like, just I’ll put it on one card. I put it one card for personal, one card for business.
Harsha Vankayalapati (00:05:26):
Exactly. Yeah. And, and, and I have like, I think I had like eight at the time or something crazy. And it was just like, this was for dining and this was for travel and yada, yada. Right. And so I was like, there’s gotta be a better way. Like, I wish there was almost like a, like a Apple wallets type app. Apple wallet style app, rather, that helped us do that, right? Pitched it to a couple of my friends. They agreed. They’re like, this is a great idea. Like, let’s execute. And so we started building like a V, like negative one of it, which is like a Google Chrome extension that would just like hold your wallets. And whenever you shopped online, it would tell you like which one would give you the best reward. I mean, that was super exciting to go through that process. This is before ChatGPT could just build things for you. We just sort of were doing our own thing for the first time, which was awesome. And I think that rush carried over into applying to YC that first time around. And it was really funny because luckily we got an interview that first time. And on the YC side, we got to the interview. It was... It was Tom Blomfield, Michael Seibel, and Richard Auberman. And they were like, yeah, this idea is not good, but the team is really strong. And so let’s just pivot to another idea. And honestly, looking back, I think it makes a lot of sense, right?
Brian Bell (00:06:40):
They’re so smart. I mean, nobody launches more startups in YC, right? So they looked at the idea. I immediately was like, eh, that’s like a 1% idea. Yeah, right.
Harsha Vankayalapati (00:06:50):
I mean, today you can very easily build that, you know, which is, which is honestly great. Like I hope it comes out cause I would use it. Right. But like, you know, it’s not an idea that takes three Fang engineers to build. Right. And so they were like, you guys have a lot of technical talent and honestly experience. Like we cut, we came out of big tech, all of us had worked like three, three and a half years there. Right. And so we saw these system level problems that were really coming out and they’re like, okay, well like, why don’t you guys work on a problem in that sphere? And so we came up with an idea to really focus on like building incident management tooling. And so from that experience, I think, you know, we like looking back, like we were probably in an interesting spot in terms of like how AI evolved because we were a little bit too late because there was a set of companies that had already been doing this. And now like I’m watching them kind of get acquired or like do all these partnerships, which is awesome to see. And then there’s also companies that came after that are like building these sort of like AI agents for DevOps and those kinds of things. And so we were like in that interesting little band in between that made it tough. And also like we were just kind of right out of big tech, had no idea how to do marketing or sales and that didn’t help us either. Right. And so we saw that play out. I think, yeah, like on the, you know, the difficulty was definitely go to market there. From there also, like, you know, we pivoted a couple of times to a few other ideas and Same kind of thing. Like it’s always go to market was always our bottleneck and our challenge. Right. And I think whether that was selling an idea that we had in our heads or selling a product that we had already sort of built an M like a B zero of it just, it was difficult. Right. And I think, you know, this time around, like with agent web, I think the learning was. that go-to-market is always going to be the bottleneck, at least at the beginning, right? There’s really almost no other thing you could be doing for your business.
Brian Bell (00:08:40):
Especially if you have a bunch of technical co-founders who can crush code and you got Cursor and Winsurf now and you could just bang out code. It’s really about finding who’s going to pay for this thing.
Harsha Vankayalapati (00:08:49):
Yeah. And I had seen a tweet or I guess an ex post by Paul Graham, which was so fast.
Brian Bell (00:08:55):
Can we still call them tweets? I think we still call them tweets.
Harsha Vankayalapati (00:08:58):
I guess we can call them tweets. Yeah, I mean, I prefer to call them tweets, to be honest. Yeah, me too. I X’d you or something. Yeah, that’s weird. It’s a little weird. But yeah, so I saw a tweet by Paul Graham, which talked about how he had seen a founder that, I mean, AI is bringing costs so low, like just for the rest of the business, right? Whether that’s ops or tech or any of those types of things. And there’s like only like one other place you can be putting money profitably, right? Or I guess maybe two others, like hiring really, really sharp talent, which is what Meta is doing. Right. They’re putting millions of dollars into hiring the best AI talent. Yeah. Five, $10 million pay packages for AI engineers and stuff. Yeah.
Brian Bell (00:09:33):
Sports, sports athlete type deals with them. Right. Yeah. And you do a three year contract with a fourth year engineer option.
Harsha Vankayalapati (00:09:45):
Exactly. Exactly. Or the other option is you just put it in to go to market, right? And so I think we’re going to see cost of customer acquisition, CAC, go way up in the next few years. Really? Knock down. You think it’ll go up. CAC will go up. I think it’s going to go up because I think attention is a finite resource. I think ultimately there’s only going to be so many ways you can really attract people. They only have 24 hours a day on an individual level. People are going to spend that wherever they want, whether that’s on LinkedIn or Twitter or TikTok or wherever. Getting to the same people is going to be so competitive because everyone’s going to pump money into ads. cold outbound or whatever. Right. And so if we don’t build a mechanism to help sort the best products, I really think we’re going to see a situation where very like mediocre products that are very good at marketing are going to like at least capture a large part of the market. And honestly, like I think A lot of the AI burnout that people are seeing today where like people are just like not as, I guess, bullish on AI as they might need to be is a lot of the reason is because like some of the products that are really hyped right now are the ones that are like kind of mediocre. But a lot of people are pushing them because, you know, inherently the ones who are selling them are marketers, right? And I think like the younger generation, especially like kids like right out of college or even in college or high school, which is what I’m seeing now, like a lot of them are really, really good at the marketing side. So when they build a company, they build that with it in mind. And so I think that that’s kind of where the disconnect is going to be going forward. At least that’s my strongly held belief here.
Brian Bell (00:11:20):
Yeah, that’s an interesting take. So when was the idea for AgentWeb? When did it first click for you? Was there a single moment or kind of a slow realization? Was it after you read that tweet?
Harsha Vankayalapati (00:11:29):
No, so I mean, it’s interesting. I think being a second time founder, I think the meme is that you realize that the market matters more than what you build. So I think like I learned that just kind of school of hard knocks, like, yeah, market is sort of the most important thing. I mean, same thing with, you know, the PMFSA, right? It really just talks about how market’s the most important thing. I would say that it was a it was a click i think especially when we started seeing like big big wins for clients right because i think a lot of folks that we speak to on a daily basis are like honestly like what we what i was right in the previous startup like i was much more technical really didn’t understand how important marketing was but knew i probably needed to do it and um once we started seeing like them getting leads and starting to get you know deals and starting to drive revenue That really showed me like, hey, we’re actually onto something here. Because honestly, like being in a position where like my way to make money or my way to grow revenue is like to help other people grow revenue is honestly a great place to be. Because like, it feels very positive some, like as long as we’re doing well for our clients, like they’ll do well too and vice versa,
Brian Bell (00:12:37):
right? It’s interesting. I don’t know if you recall, but I ran demand gen teams. I started in sales and then I ran demand gen teams for a few VC backed startups. These were a million or $2 million budgets per month to grow. Pretty large teams. And what I realized after a few startups is you can accelerate a great product, but if you have a bad product, all bets are off. Even if you’re dumping dollars, millions of dollars into your marketing. How do you think about that kind of push pull? The way I think about it is you’ve got to have Everything has to be pretty good, right? Especially in enterprise software. Consumer is a little different. I think consumer, the best product usually wins. But in B2B, you got to have good sales. You got to have good demand gen. You got to have a pretty good product. And then you got to have good customer success to upsell, cross-sell, retain those customers. How do you think about that mix? You’re really tackling that demand gen problem. Creating demand, sales qualified leads, MQLs, etc.
Harsha Vankayalapati (00:13:30):
Yeah, yeah, yeah. And, and I think from experience, I have to agree, right, that it’s really going to be up to the product, or even more importantly than the product, like the actual problem statement that you’re trying to solve, right? Because like, you could market just an idea. And like, if people are interested, they might be able to like, click through and sign up for a waitlist or something, right? If they just think the idea is so strong. So you really can’t like out market a bad product. It isn’t really a single variable problem though, in my opinion, right? Like I think you do, like you said, have to have pretty good on all fronts, right? And it’s really, When you’re just starting out like the negative one to zero stage or the zero to one stage, it’s kind of a question of like order of operations, you know, like which do you do first? Right. Like, do you focus on just doing a bunch of customer outbound and seeing like, OK, well, what do people like want to have? And then I build that directly for them. Or do I just build what I think needs to exist and then just start selling it? Or like some combination, right? And I think that founders really get stuck figuring it out like in a linear fashion when ultimately like it does need to be simultaneous, right? Like you learn and then you build and then you show the clients and then you make the change and then you just, you make that loop much stronger just from all the data sources that you’re able to get. But like the very first step, I would say is if you could do like I read this book a while back called Prito Typing, which talks about this, that you can really test like it’s a the idea behind the book is like there’s a prototype and a Prito type. Right. The prototype is something that, you know, you send as an MVP. Right. Like it actually tells you, like, can you actually feasibly build the thing? But a pretotype is before that. A pretotype is, do people actually want the thing you’re building, right? And there’s ways for you to validate that before you build. And like part of it can include marketing, right?
Brian Bell (00:15:20):
Yeah, some landing pages, throw some ads, see if people are signing up for a waitlist kind of thing.
Harsha Vankayalapati (00:15:24):
Exactly. One of the YC pieces of advice that I’ve actually come to disagree with with time, especially while running this, is that YC really hates paid ads. They really don’t think you should be doing that. And I understand why.
Brian Bell (00:15:38):
They’re all about founder-led sales until the first million of error. That’s what they should say. Just go hand-to-hand, get out of the building, do things that don’t scale, go talk to your customers. That’s probably what they would say.
Harsha Vankayalapati (00:15:49):
Exactly. Like 99% like of that is probably true. I think the first step though, like you can probably run like a really small test paid ads budget and actually just see like if people want what you’re building, right? And I mean, in the prototyping book, they even talk about this. Like you could set up like a lander, They call it like a Wizard of Oz style prototype where you basically fake it until you make it. Like you have a place for them to click into that shows you some engagement. It’s basically like a call to action. And if you run like $500 worth of ads on that, like you can find out real quick if people actually care, right? Or if you post, I mean, you know, you can post on Product Hunt or whatever as well, but like, The challenge with doing it the other way, which is like launching sort of organically is you’re very much at the whims of the algorithm, you know, because if LinkedIn doesn’t care about what you posted, it doesn’t get to the right people. And like, you may not even have like a problem with your product or even the idea, it just didn’t get shown to the right people, right? But if you do a tailored ad, you know, a tailored sort of post that is actually like ran through, you know, ad spend or something like that to actually boost the impression rate, especially to your ICP. And you do that for 500 bucks that you can find out real quick, right? If people actually care. I think that’s something that founders might miss, I think, along the way, just hearing this advice blankly.
Brian Bell (00:17:08):
So what is the specific pain point that AgentWeb exists to eliminate for founders?
Harsha Vankayalapati (00:17:13):
So the specific pain point I would say is that founders probably shouldn’t invest their time in getting really, really good at managing the systems that exist for go-to-market. They should get good at understanding the messaging that works, the problem that they’re trying to solve, and the customers that they work with, or even prospects. A founder needs to understand that. But the mechanics of go-to-market are confusing today. They were confusing when I started, and they’re more confusing today than when I started. And I think they’re going to be even more confusing as time goes on. Right. Like I, you know, like I had mentioned, I messed around with a few business ideas when I was just starting off. One of them was e-commerce. Like I would just do like drop shipping, for example. Right. And back then, like meta ads was actually like decently easy to use. Like it was very straightforward. But today it’s very complicated and they probably had to make it more complicated to service all of the different ways that you might need to run ads. Right. And it’s only going to get more complicated is my, is my personal view. I, you know, that is how meta makes their money today, but it’s like super hard to get ahold of support and they care a lot about their really high ticket clients and they might not care about as care as much about what your, you know, your issues are. Right. And so. that’s just one example and then you have like every social media channel and then you have email outbound it just changes every single day right there’s like do you use like outlook do you use gmail do you do you use like smtp or any of these other email like client providers right and like how do you like rotate domains and I mean, each of these different marketing channels has their own sort of game you have to play and it’s just going to get more complicated. Right. And I think that, you know, founders could spend the time and the energy like, you know, I’ve seen like, you know, four hours a day is like what you should be doing for go to market. And you could you could learn it. I think most founders are able to. But like that also takes time away from building and talking to customers and working on product. And I think that if the split could get made where founders do what they want to do, which is building product, talking to users, iterating, et cetera, whereas go-to-market can get outsourced, which is what founders typically are doing today too. They are outsourcing it, whether that be to you know, tools or like agencies, right? Because they wouldn’t exist if it wasn’t a problem. And so I think that for like our specific use case, agencies are really the competitor I think we’re going after, right? Because I think my experience has been that they don’t understand like the founder journey as much. They’re really built for like the traditional type of SMB, but like startups are just very different in how they operate.
Brian Bell (00:19:47):
Yeah. So why now? Why do you believe marketing and GTM are especially ripe for autonomous agents?
Harsha Vankayalapati (00:19:53):
So like I had mentioned, I think CAC is going to increase as time goes on. And we’re kind of like, I mean, Sam Altman had said this, like we’re in like the entering the fast fashion era of SaaS, right? I mean, Lovable is doing super well, like getting, you know, prosumers building pretty good apps as time goes on. And they’re only probably going to get better, right? Like there’s other vibe coding tools. Vibe coding tools out there like base 40, you could use cursor directly if you wanted to. And so there’s going to be a large influx of products, right? Some are going to be mediocre, some are going to be really great, some are going to be really bad, right? And so like sifting through all of that, I think the attention economy is going to become where the actual battle is going to need to happen. And I can tell that this is the case because like Lovable has mentioned in the past that they want to help founders actually go to market with their products, right? But that’s a whole other ballgame. And they announced that like a year ago at this point, or at least like you know, close to a year ago and they haven’t made way on it yet because it’s super difficult. Like it’s its own startup, right? Like we are in this space and we haven’t solved even like, I want to say 10% of the problem because there’s so many different ways that go-to-market can go, right? And so finding the right-
Brian Bell (00:21:07):
Yeah, I’d love to hear what’s the 10% you’ve solved and what’s the other like 90% that you’re looking forward to solving?
Harsha Vankayalapati (00:21:12):
Yeah, so the 10% we’ve solved is, well, one is like we’ve solved the nurture piece, right? So we can help brands really get in front of their clients on a regular basis. We do that through like SEO, AEO, right? So we can help them, for example, place on Google search. or even these LLMs, right? And so there’s interesting ways that they differ as well as ways that they synergize. We can also develop posts that actually sound like the user, right? Like that was a whole problem that, I mean, LLMs just have a very generic way of sounding. And something that was very important to me is like, if I’m making posts on LinkedIn, which is like how we’ve been able to drive a lot of traffic ourselves, it should sound like I wrote it. or at least make it in such a way that i can remix it and add my own own ideas to it right and so we’ve built it with that in mind as well we’ve gotten a lot better at the image side too i think that like ad images when we started were just so easily able to be told that it was ai but like as the models improve you know we’re noticing that it’s actually getting very very close to being being like photorealistic or at least you know not There’s no difference really between doing that or running it yourself via like Photoshop or something like that, which is like a great, a great place to be. I’d say that the parts that are hard on the content creation side is like video has been a big challenge for us. You know, you could ship a video with the product, but like it doesn’t pass our bar even, which is like, you know, we want to be almost higher than our customers because like we don’t want to ship a product that’s bad and like, have people basically have a bad experience because like it is very much, it is very much like a results oriented product, right? Like if you’re not getting the right results with your, with your ad sets, for example, then you’re not going to drive revenue and then you’re going to be unhappy with what we’ve developed. Right. But then beyond that, there’s like so much more. I mean, there’s like, Hold outbound running those campaigns. There’s going on the different social media channels too, right? So like exploring on like what you can do on X and TikTok and Instagram, not Instagram, but like LinkedIn specifically on their ads platform, right? I mean, there’s just so many different arenas that you can play in. Each one has its own different battle.
Brian Bell (00:23:19):
Yeah. How do you think about autonomous GTM versus traditional marketing automation? Because a lot of these automated tools, I mean, I was a Marketo certified expert, you know, almost 15 years ago now, right? Doing all this stuff, you know, sending emails and sending ads and lead scoring and following up. But it was very manual to program it.
Harsha Vankayalapati (00:23:38):
In the previous startup, like I used, for example, HubSpot, which was like, you know, this more traditional marketing system,
Brian Bell (00:23:44):
right? They’re pricing way up. Speaking of CAC, wow. You start looking at some, you know, CRM with some basic sequences, like $1,500 a month per user. I’m like, what?
Harsha Vankayalapati (00:23:54):
yeah so expensive and and i think like hubspot is so fascinating to me because like they weren’t built like ai natively right and i mean to be clear it’s not clear today if you need to build ai natively like we’re taking the bet that you need to right but maybe hubspot you know they have some ai features but they weren’t built with that in mind right and so one of the challenges that i’ve seen with hubspot and this is just my own experience right is I came into HubSpot as someone who actually was familiar with the CRM. I worked every single day with Dynamics CRM at Microsoft, right?
Brian Bell (00:24:25):
Notoriously difficult CRM to work with, by the way.
Harsha Vankayalapati (00:24:28):
Right. And so I actually knew how to like navigate my way around a CRM on a technical level from using Dynamics, right? And like, They have power automate and that lets them sort of, you know, create, you know, BI documents, that kind of thing. Right. And so I had a lot of experience even on the CRM side. And even I found it difficult to like hop on, use HubSpot, actually set everything up and not for lack of ability, but probably for lack of time, you know, because I think like there was just better things for me to be doing as a founder, like going and actually just executing on outbound or like reach talking to potential users, that kind of thing.
Brian Bell (00:25:01):
I think back when I ran demand gen teams, I had like a 10, 12 person team. We’d have like one person, one or two people just doing the email marketing automation stuff. That was their whole job.
Harsha Vankayalapati (00:25:12):
Yeah.
Brian Bell (00:25:12):
And then you have one or two people doing the ads. And then you have one or two people doing the design and the copy and so forth. And it’s not easy to run demand gen with last gen tools.
Harsha Vankayalapati (00:25:23):
Yeah, it’s very difficult. I mean, it’s a full team’s worth of job, right? Like you said. And I think a founder doing that is even crazier, right? Because they have to be the whole team and they’re probably not doing everything they need to do.
Brian Bell (00:25:36):
And they’re not an expert in all those tools and how to do it and how to execute and
Harsha Vankayalapati (00:25:41):
Exactly. Yeah. And I think like the first step for what we’re building is that it needs to be, I mean, the agent should actually help you with that execution, right? It should become like a partner that actually tells you like, here’s what’s good, here’s what’s not good. And that actually will end up being probably our moat at the end of the day, because not just myself, but even the team has like almost 20 years of go-to-market experience between us, right? Our CPO, Fang Fang, she worked at Meta’s like ads measurement team, right? So she led that team and she worked at LinkedIn and she actually led their B2B SaaS outbound team there. We had that experience on the enterprise side. And Matt, our CMO, actually ran paid ads for his own business that he was able to sell. But they were running like 1.5 million in ARR on just running that D2C business. And then he went on to work at Preflect AI, which was an AI software platform. And he ran the go-to-market there as well. And he grew that from four to 10, right? So we’ve got a team of people that actually understand go-to-market even better than me. And we’re making sure that the agent just has all of that information so that if and when it does need to actually you know, co-pilot with somebody, whether that’s on, you know, HubSpot or even if we build out our CRM more like on our own CRM, right? That actually will do the process of onboarding them, setting everything up and actually being informative on like how to actually become a better marketer using AI, which is, I think, the best use case, right? I think that like humans should do what they’re really good at and then AI should do what it’s really good at.
Brian Bell (00:27:12):
Yeah, what I love about building MarTech, selling it and investing in it is that there’s this feedback loop of dogfooding your own product to sell your own product, right? So you guys are probably in your product every day using it to generate demand for your own product. Tell us about that.
Harsha Vankayalapati (00:27:26):
Yeah, I mean, it’s a great feedback loop, right? Because I think one of the best ways to drive our business forward is to like prove that we can drive our own go to market. So if I’m making LinkedIn posts that go viral or do super well, it actually is proof that the AI is working. And then now I can then tell this to founders and say, look, if you wanted to do the same thing, it’s actually a playbook that’s built into the tool and it’ll help you do that. And so founders can see the results live, look at the tool that I used to build it, look at the playbook and the exact way that I did it, and then just basically copy it and fine tune it for their own use case right and so one of the best ways for us to grow is honestly just to be better at go to market for ourselves that’s just on the you know founder branding side right like if we if we’re placing super high on seo same sort of thing right like we can show that hey we were able to do this for ourselves like here’s how we would do it for your business and then you know run the playbook there and then You know, lastly, on the paid ad side, it’s the same kind of thing. It’s a very great feedback loop between like, if we make our go to market better, we can get better clients. If we get better clients, we actually get better wins with them. And then we can have better case studies and then get better clients yet again, right? So it’s actually, I mean, it works extremely well in that model.
Brian Bell (00:28:42):
Yeah, amazing. How do you decide which parts of marketing should be fully autonomous versus having a human in the loop?
Harsha Vankayalapati (00:28:48):
Yeah, I think, I mean, we had to figure that out just by experience, right? Like, you know, we started this thing, like, how much can we put on AI? We probably put a bit too much when we just started, right? Because like, we were trying to get it to do like, video scripts, like writing the scripts for the ads, like, you know, even like analysis of the ads at the end, right? And so like, once we went through that loop, I think we figured out like, okay, today AI can do this, humans can do this. Like, obviously it’s a moving target, right? As the models improve, like AI can take up more and humans can do less, right? And so we wanna like toe the line between like, How do we build for what can be done today versus what can be done tomorrow, or at least what we anticipate can be done tomorrow? And so the fine line that we’ve drawn there is like humans are probably going to be best at the taste part, right? So like figuring out this image looks really good, right? And so like that’s an example of an image or like a post, like this is a really strong post, right? Like this would probably do well. And so humans are really good at that analysis part, right? And so that can be done probably in a shorter period of time than it takes to actually make the content or make the...
Brian Bell (00:29:55):
Kind of the taste, right? They kind of look at it and go, yeah, that’s tasteful.
Harsha Vankayalapati (00:29:59):
Yeah, exactly. Exactly. And so with that in mind, right? So like, how can we enable AI to do the heavy lifting of content gen or even like, you know, generating the headlines for the ads or generating... the email copy, right? Like how can we make sure that AI gets all the tools it needs to actually do the work? And then humans are at the end just approving and rejecting and then adding the changes that are needed. And that process we found to be a loop that actually is highly scalable, right? Because at the end of the day, like, I think a lot of people in AI are saying that humans will probably move towards more QA. I mean, that’s what we’ve even seen on the on the coding side, right? Because if I’m running cursor and I’m having it basically build a product for me, like I do, I run it at the end and I’m like, okay, this doesn’t look quite right. Like go ahead and change this. Like I’m rarely if ever going in to the code itself and actually making the updates. And so I think the same kind of thing is there too. Like maybe on the image side, like I go and put it in Figma, maybe like add some text or remove some text or something. But it’s doing like 90% of the work there. Right. So it’s,
Brian Bell (00:31:03):
it’s all kind of laughing because every QA engineer I’ve ever known wanted to get out of QA, you know, and go into regular full stack.
Harsha Vankayalapati (00:31:10):
I think people are like really like, I think split on the future of AI. Right. Because I think like half of people are like, okay, great. Like we can go back to doing jobs that are much more QA, but like people that are actually QA engineers, like, actually, I don’t know if I like this job that much. And it’s the same kind of thing with like other jobs too. Like, I think I saw a post that was like, you know, do you really want to go into like vocational jobs? Because they are kind of tough and like, they’re really hard to do. And like, if AI gets really good at all the laptop jobs and that those are the only types of jobs that’ll be left. Right.
Brian Bell (00:31:38):
So it’s, it’s like creator versus editor, right? Some people really like editing. Like that’s me. Like a lot of times when I write, I’ll kind of talk to the AI for five or 10 minutes about what I want to say. It drafts the article that I get to edit it. And that’s kind of my writing process now with AI. But some people are like, no, no, no, I’d rather write it all out and create it from scratch and the editing part they don’t like. Yeah. So I think you’re kind of describing that dichotomy or that push pull between editing and creation.
Harsha Vankayalapati (00:32:05):
Yeah, that’s, that’s actually so I’ve never thought of that split, but it’s so fascinating. I’m looking at my own workflow, like I find myself doing both, like at times, like sometimes I’ll have an idea that I’m like, okay, like, hey, I take the seed and just like plant a tree with it and see kind of how it grows. And then I go in and maybe like edit it how I want. And like, sometimes like, I’m just like, you know what, like, you know what to do, like, I’ll go and maybe like, make some edits at the end. So I think like, it’s, it’s interesting to have those two processes kind of how people operate, right? And building it in a way that actually lets you do both, because that’s actually kind of implicitly how we’ve done it. Like parts of it is like you can have the images get made from scratch and you go and like snip around and edit it. And the other side is like, oh, actually, like you can prompt it in and then actually generate the image exactly how you want it to.
Brian Bell (00:32:52):
And then, you know, do some revisions on that as long as it fits your brand. So let’s talk a little bit about building with AI agents. You know, they’re powerful but brittle. What’s been kind of the hardest reliability challenges you’ve faced as you start to scale?
Harsha Vankayalapati (00:33:05):
Yeah, that’s a good question. AI agents, like... The reliability challenges I think are largely because of the model shifting a lot, right? So I think the goalposts are moving super fast in a way that we haven’t seen in tech before. I mean, breaking changes get put into like open source products like all the time, which like just, it actually freezes you in place. And so like they might be making changes ahead of you, but like you can’t even make the update because it might break your code base, right? Like we’ve had that experience with some of the internal toolings that we’ve used. And so I think like the fact that it’s a move, like a, such a fast moving industry is the biggest challenge, at least from a traditional developer perspective, right? Like that’s why I think it’s fascinating to watch like the younger generation go at it because they don’t have this concept of like, Hey, things need to be stable in terms of like a stable ground to build on top of, like they actually will be totally okay. Rolling with the punches and like letting people, you know, the models just change and then like things break, but that’s fine because like they’ll just update their process and they don’t have like a marriage to the processes, right? And I think that that I think is largely where those reliability issues will come from. But still, it’s also a benefit too, right? Because like once the changes do get made and we adapt to them quick, we do see just better outputs overall, right? Like on the image side, it was night and day. Like once Nano Banana came out, for example, like the images just shot up. Like I think before that we were like, like we probably have to like maybe help people like prompt it and then you know maybe they have to go somewhere else to build them but like once that that change got made it’s like okay now we can actually do it and it’s only been getting better
Brian Bell (00:34:44):
there right so i think i’m sure something will come out for video too like four or five or whatever and it’ll all of a sudden be like oh this is perfect like this is nano banana level quality and that could happen probably in the next 12 months yeah
Harsha Vankayalapati (00:34:55):
exactly and maybe even sooner to be honest like it just moves so quick
Brian Bell (00:34:59):
Yeah, feels like every week, you know, there’s a new benchmark saturated. You see that 5.2 saturated, the math benchmark, like it scored like 100%. It’s like 100% on the Math Olympiad. You know, math is going to be solved. You know, that’s what they talk about on the Moonshots podcast a lot. You know, math is solved. It’s going to be solved.
Harsha Vankayalapati (00:35:14):
I have my own opinion on that too, but the interesting thing about benchmarks is that like, I don’t think they actually like approximate usage that well, because like we had seen even before, like with GPT-5 when it got released that it was just crushing the benchmarks, right? Like math, science, whatever, it was just top of its class, but like you could see it struggle, I think, right? When you would ask it even basic like arithmetic questions. you know? And I think that I’m sure you could do the same thing with like 5.2, right? Like I’m sure there’s like questions that people will find that it just tumbles out.
Brian Bell (00:35:44):
And I think that like humanity’s last exam, I think is that benchmark. And it’s, I think it’s still at 30 or 40%. Yeah, exactly. Exactly.
Harsha Vankayalapati (00:35:53):
And I think, you know, like I wonder how much of how the models are trained is to actually make sure that they look good on the benchmarks, you know, versus like general usage. Like that’s something that I’ve seen in the, in the industry too.
Brian Bell (00:36:07):
Yeah, you kind of train it to the benchmark. Look, we saturated the benchmark. It’s really just like fitted everything in the benchmark into your model, right?
Harsha Vankayalapati (00:36:15):
exactly right and so yeah of course you get 100% because like you put the whole benchmark in right and so you know that actually will allow you to you know crush the benchmarks right but I think like you could argue that’s what students do when they take like a SAT or GMAT test or something too you’re just kind of fitting it into your neural net like all the possible variations of all the questions and is that really real intelligence right really interesting problem well I’d love to talk about your ideal customer for agent web today who’s like the ideal startup or company
Brian Bell (00:36:39):
Well I’d love to talk about your ideal customer for agent web today who’s like the ideal startup or company
Harsha Vankayalapati (00:36:45):
Yeah, so one thing that we found is we started going pretty wide at the beginning. We’re like, okay, everyone needs to go to market strategy, right? So why not just sort of see what does well? And I think the best clients that we’ve had have been ones that are on the B2B side, especially like B2B SaaS. The reason we found is because they’re a lot more forgiving on their margins, right? So like B2B SaaS, for example, expected 80% margin plus, right? And so they have a lot more room to spend on CAC right and so that allows them to actually take some risks with even like hiring agencies as long as they bring in new clients i think the challenges for the b2b sas side are not so much on cost of customer acquisition as it is on actually proving if your product matters to the customers you’re going after and so it’s actually a worthy use of funds to like spend on actually discovering that through whether you’re running ads or using an agency or anything like that and so they’re very forgiving like with you know the initial sort of warm-up period and like we actually can get them really big wins right because like you know 10 leads can go very far for a b2b sas client right versus for example like a consumer product good client like 10 customers might only be like you know a thousand bucks if you’re you know average order value is like a hundred bucks right and so that might not even cover the fee of like what we would charge to do that work you know and so i think that the math works out better i’d say for like b2b sas clients and and also like bigger acvs yeah exactly bigger acvs like longer long tail right like you know the ltvs are much higher right so it just ends up working out better i also think folks who are much more focused on like brand can work well with us too. So it can also be B2B SaaS, but we’ve seen that even like on the deep tech side, because like in the process of training out our agents, it’s gotten really good at even understanding deeply technical concepts and writing content around it, right? So we had a client that was building like a manufacturing robotics company. And so we were writing blog posts that were deeply technical that I wouldn’t even understand if I read it, right? But it actually passed their bar that they’ve actually put on their site, right? And so that means that the agent’s gotten really good at understanding really deeply technical knowledge as long as you put in the information up front. And they can actually remix that and create content using the right parameters.
Brian Bell (00:39:08):
When did you know you were onto something? You’re like, okay, we have something here.
Harsha Vankayalapati (00:39:11):
Yeah, I think... I think the comparison between how this journey has gone relative to the previous one, I think like we tried a bunch of ideas in the first startup after we kind of left the incident management space and each of them had their own struggles. It was like a struggle to sort of justify the problem or even like get people on calls and like actually understand and get people describing the problem in a way that we could even build towards, right? This was a case where like I felt such a strong pull, like I think maybe the exact moment was like when I was in San Francisco to actually just, you know, meet with other founders and potential investors as well. I would just explain my idea, like not even trying to pitch. I would just say, here’s what I’m working on. Like I’m helping founders with their go-to-market. And people would like opt in, like, oh, actually, can we take a call? Yeah, that’s a big problem for us.
Brian Bell (00:39:57):
Yeah.
Harsha Vankayalapati (00:40:00):
yeah and so i’m like well yeah happy to do so right and we’ve ended up you know closing a few of those clients and i think that that really shows to me that at least like directionally the idea is the right the right way to go right whether or not the tool itself is a perfect tool is like probably not true but like at least you know the moonshot vision the north star is correct and i think that that means that we’ll grow in that right direction as long as we keep talking to people and getting their feedback so
Brian Bell (00:40:26):
So how does it work, you know, for founders out there listening who might be interested, like what’s onboarding look like? What does pricing look like? What kind of CAC have you experienced on the platform? All those.
Harsha Vankayalapati (00:40:36):
Yeah, it’s a great question, right? So we’ve got really, I guess, two different paths, right? So the first easiest path is just onboarding onto the product itself. And so that, you know, what you would do with that is you just sign up with our signup link that’s on our website. It’ll take you to a Stripe link that it’s only for, I think right now is like maybe $199 a month, maybe even less. But basically you would just log on, use our agent and deliver results that way. We can even help coach on how to make the agent work better. Or we have the option if that’s like, you don’t want to add another tool to your stack, which I wouldn’t personally, like I’d rather have someone else take care of it, is actually to do it as a managed service, right? And so we actually will pilot the agent for you and make sure that the agent is actually operating, getting you the results that you want. And so how we do that process is a little bit more involved. So we actually start with like a free go-to-market audit, which is myself and my co-founders will come in and actually look at what you’ve done with your current go-to-market, what’s worked and what’s not worked. And then we’ll actually develop a strategy that you can even take home and not even use us for. Or if again, you want someone else to handle it, like we can come up with a way to sort of split it up in a way that, you know, maybe the agent handles some or like we would handle some and then deliver your price at that point. And so the way that we typically do that is it’s either, you know, you can either do brand building, lead gen, or you can do both at the same time. For the brand billing side, it’s two and a half thousand a month, which is actually highly competitive with an agency. Like if you were to treat that with, you know, Look at an agency on the market. The lowest priced ones I’ve seen are like $5,000 a month. And those are actually kind of mediocre. The ones that are probably going to be good are like $10,000 a month. And so we can save all that money for our clients because we’re just using agents in the background. We don’t have to pay any marketers at the bottom level to really do this work. And the same price is actually there on the lead gen side. It’s $2,500 a month for lead gen work. And so to do both of them, we’re actually doing that as a bundle at $4,000 a month. And so then we would just handle both of them for you. We’d make sure that you’re getting lead gen, which is like hold outbound and paid ads. Or we would do brand building. And that would be like SEO posts, newsletters, founder posts, right? Making sure that you’re getting good founder posts out there as well. And so wrap that all up as $4,000 a month. And we think that that’s been like a price point that a lot of folks have been really excited about.
Brian Bell (00:42:56):
Yeah, and then what, how long until founders know it’s like working? Like what does success look like, you know, 30, 60, 90 days?
Harsha Vankayalapati (00:43:02):
We’ve seen success internally in like as little as the first ad run if we’re going to do lead gen for folks, right? Like it’s actually super clear when it’s working. And it’s almost like a media, right? And so I’ve seen that. I’ve had that experience. It’s super exciting to watch, honestly, as a person who’s making it happen for someone else. Because run ads, we did this for a client where we ran only $500 worth of ads and we got them 168 leads just from that spend. And I was like, wow, this is like PMF if I’ve ever seen it.
Brian Bell (00:43:30):
Yeah, the cost per lead of three bucks is off the charts good.
Harsha Vankayalapati (00:43:33):
Crazy. It’s crazy. And so I was like, okay, well, like that’s, that’s super awesome. Right. And we’ve had clients where it’s not gone that well. Right. And so I think that like, you know, within the first 30 days, honestly, we’ve been able to see really great results and like, we’re so okay to like you know just give the money back honestly after the first 30 days if it doesn’t work because we’ve just seen it work so quickly and if we do think there’s a chance like we’ll say hey maybe it might take another month on the lead gen side like we’ll make some tweaks here and like updates and actually result and get some strong results for you in the next month but we’re not like really super interested in like hey we’ll keep you on board just to sort of pay our bills if it’s not working for you right like we’d rather work with folks where we’re also delivering value on our end. And so those are the ones that I think we’re the most excited to work with too. And they’re the ones that are most excited to work with us. And so we offer that self-service plan really just as a way to sort of downsell people. For example, don’t see immediate success. They can just work with the agent directly, which we’ve been using to actually run their go-to-market anyways. And so they’ll at least walk away with the agent being able to do the work that we did over the course of those two months. That’s very cool.
Brian Bell (00:44:43):
So what are you excited about over the next, you know, 12 to 24 months?
Harsha Vankayalapati (00:44:46):
So I would say like more integrations with other platforms. I think that we’re certainly building, building very much our own platform, our own version of things. Like for example, like a CRM or building like an AI native CRM, like outbound, same kind of thing. But we’re also increasingly finding that founders have their own setups too, right? So they might use something like a HubSpot or they might use something like a Marketo. And so like if we can build agents that actually help onboard onto those platforms, like that would be super exciting to build. And I’m also super excited to like have our agents actually be a lot more autonomous for certain workloads too. Because I think like parts of what is preventing us from moving, taking those steps forward is, you know, the LLMs are not quite where we need them to be yet. But like over the course of the next 12 to 24 months, like they’re just going to get more better. more powerful right and so they’ll understand things better they’ll understand instructions better they’ll take less tokens to get the results that you want and so that’s going to result in just like a large uptick in the amount of workloads
Brian Bell (00:45:49):
that we can do concurrently yeah what about the long term what does success look like over 10 years hmm
Harsha Vankayalapati (00:45:54):
Yeah, it’s a really good question. I think, you know, in the 10-year time horizon, if AgentWeb gets to the point where it’s, you know, handled the marketing side, right, I think we would then move on to covering more of the business, right? We would cover sales, customer success, and build out agents for those processes as well. I think that in an ideal world, this becomes basically a companion to something like Lovable, right? So you would use Lovable to build your product. And maybe over a 10-year horizon, maybe we even built something like that, like a vibe coding tool or a tool that helps you even launch a product.
Brian Bell (00:46:32):
Yeah, I think AI would get sufficiently advanced in the future that you could just use whatever ChatGPT 7.3 is and just say, hey, just go make this for me.
Harsha Vankayalapati (00:46:39):
Yeah, exactly. Right. And so you build out your tool, right? And then this would do the process of actually finding a market for your tool, you know? And so I don’t know, I don’t have an opinion yet on like the agent to agent market because like that’s a whole other ballgame. And like if it ends up taking off more, like we’ll certainly build towards it. And that might require its own different realm of what marketing needs to look like. However, like in the case of, you know, you’re marketing to people, they buy your tool and they use it. that equation would be, you know, we would want to handle the piece of like, hey, how do we get the product to market and like go across channel, right? So, you know, marketing, sales, customer success, help you actually, you know, grow your business very, you know, almost like, I guess like vibe, vibe grow your business kind of thing. Vibe growth.
Brian Bell (00:47:23):
I like that. You got vibe code and you got vibe growth. Yeah. Yeah, exactly. I love that. Well, let’s wrap up with some rapid fire. What’s a strong opinion you hold about marketing that most people disagree with? I would say like,
Harsha Vankayalapati (00:47:37):
yeah, I mean, paid ads is actually a totally realistic way to grow your business. And it would be, I think PLG is actually not the right way for a lot of startups to grow. I think it’s actually better to do marketing led or I guess founder led.
Brian Bell (00:47:53):
Yeah. Or generate the lead, do the demo, try to close the AC, you know, the contract kind of thing. Yeah. I would, I’d probably agree with that, especially at the beginning. I think YC would say that too. So what’s the hardest decision you’ve ever had to make as a CEO?
Harsha Vankayalapati (00:48:04):
I would say like turning down clients or like actually like firing customers that we’re working with just because like we’re not actually getting them the right results. Especially like if it’s like a friend we’re working with, right? And we’re like, okay, man, like it’s not working. Hey, sorry. Yeah, it’s not going to work out. Whatever, but it’s not working. And like sometimes they really need it too because it is like marketing can help a business that’s not doing well sometimes. It’s tough to see that happen.
Brian Bell (00:48:30):
Yeah. What founder behavior do you think will disappear because of AI agents?
Harsha Vankayalapati (00:48:36):
I think that the behavior, like I actually think 996 might disappear. The reason for that is because like today people are doing 996 because like, well, one is like fashionable, I guess, to sit in your house all day and sort of vibe code a behemoth of an app that maybe people will use, maybe they won’t. I think that like as AI agents improve, like I have some friends that are even building towards this, right? Which is like, I have a friend who’s building an app that actually lets you like voice code your way through a product and actually like ship out these tools. And you can be on the move and do that. And I think that that’s like, if products like that take off, I’m super excited for like that future world, because that means that, you know, you don’t have to be sitting at home and just on front of your computer to actually deliver really good results. Like you can be out and about like social, go to the gym or whatever. Right. Like I think that’s the, that’s the promised future I’d say that AI agents have that like, it actually will be a lot less work and a lot less computer work. I think people will just become healthier.
Brian Bell (00:49:38):
Yeah. I think, I think working at 996 is completely unsustainable. Don’t do that. You know, Like maybe a nine, seven, you know, five. And then, you know, four hours on Saturday or something. Yeah. That’s kind of what I work. I start work about eight, kind of finish at six or seven. Maybe I throw another hour in in the evening sometimes. I’m working 50, 55 hour weeks. And I feel like that’s right at the edge of like sustainable. Yeah. And then, you know, once you start getting to that 60, 70, and I did, I worked on Wall Street. I’ve worked in lots of startups. It just starts to get like, just your brain gets foggy and you start making mistakes and your college starts deteriorating and...
Harsha Vankayalapati (00:50:14):
I think in the SF world, it’s really fashionable today to do it. And I’ve met plenty of founders that do it today. And I think that you just start to see the life leave them after a few weeks. It’s like, hey man, this might actually not even be that good for your startup if you’re doing that much work.
Brian Bell (00:50:30):
Yeah, probably working too hard. You got to work at a sustainable level. I mean, you could do short bursts of that. Like, hey, we’re trying to make this deadline. We’re trying to onboard this big customer. At some point, you’re like, okay, I just need to work a 40 or 50 hour week to kind of recharge. Yeah, definitely. What’s one thing you’re actively unlearning either from your previous startup or Microsoft?
Harsha Vankayalapati (00:50:49):
I think in the previous startup, I was a lot more timid in terms of even doing marketing or just even talking with the team. Just because I think my personality types, I do want to make sure everyone’s happy at the end of the day. But sometimes you do have to... Well, in the case of LinkedIn posts and that kind of thing, if you make a fool out of yourself, it actually might be good marketing. So... That’s a whole other thing. But in the case of like talking with the team, it’s like making sure that like the truth gets said, at least as you see it. And then obviously you can have a discussion around that. And I think that ultimately, like, you know, having those really strong, like, you know, heartfelt discussions between like the core founding team is always going to be super important. I think like even Steve Jobs talked about this. apple got super big but they would still have like you know three hours a week meetings where they would just talk about the whole business and make sure that like everything is like really really you know well done and like everyone knows what they’re doing that kind of thing and i think that like making sure that we do that at a good enough pace is going to be super important what ai capability has
Brian Bell (00:51:53):
most changed your roadmap i guess we talked about nano banana but what else i’m
Harsha Vankayalapati (00:51:57):
trying to think if there’s any there’s like i think like all of them have always like rejiggered kind of the trajectory of the company, right? Like when we started, it was like all about like sort of browser agents. And then that actually, that whole sort of like wave of browser agent tooling actually is like what really helped us start off agent web as like building out the initial sort of agent that would go out and do marketing work. Then like, you know, image APIs came out and then they got a lot better and then allowed us to actually unlock for us a new like wave of content that we can develop, right? And so I think like the next big one will be probably video. Like once that gets really good, I think that that’s gonna actually change the paradigm of how marketing even looks. And I think, like, Sora, in my opinion, Sora 2 got very close, right? Because, like, I made a post, like, where I was, like, shaking Sam Altman’s hand, and it just, like, looked so realistic that I had people telling me, like, oh, you met Sam Altman? That’s crazy.
Brian Bell (00:52:54):
I did go to YC twice, so, yeah. Chances are you meet him, like, at least once.
Harsha Vankayalapati (00:52:59):
Yeah. Yeah. Yeah. And I think that like, you know, that, that, that, like that whole process was, was super interesting to watch. And I think that like, as video improves, I think it’s just going to unlock a new wave of business too. And I think that like the last change, I think that like we’ve been observing. And so we’re, we’re, we’re building kind of around it without like directly kind of interfacing with is like, honestly, Meta’s AI advance, right. Advances. Like, I think they’re going to start making it so that the ads that are generated through their platform are going to perform super well. Like they actually, I think, did a bit of an arbitrage where they let anyone put in AI ads on Meta to sort of see what worked and didn’t work without having to spend their own money. And then now it’s they’ve done all that training for free, basically, on like what types of ads work for people. And I think that, you know, we’re going to start to see them be a big player in the space as well.
Brian Bell (00:53:49):
Last question. If you had to start over again on ancient web, what would you do differently in the first 30 days?
Harsha Vankayalapati (00:53:54):
I think in the first 30 days, I would focus a lot. I would have focused a lot more on how can we get in front of users quicker and like focus less on launching and like making sure that, you know, we make a big kind of hurrah about what we’re doing. Because I think like the product evolved a lot from then even. Like we started off, you know, actually started off as a generic agent, right? So like back then, like, you know, Manus was super popular and we were kind of building a tool that was a generic agent that like, maybe was competitive with Manus, but like kind of had its own flavors. And I think that like we launched that, got some feedback and then like sort of reiterated. And then from there, I think the decision got made that like, hey, like let’s just look at what our team is strong at, right? Like we’ve got 20 plus years of go-to-market experience on the team. Like we could probably build something really sharp in the go-to-market world. And it’s a hugely... Like a large problem for startups. Right. I think that we focused on like the selling to like selling the shovel sort of method. We’re like, actually, like maybe we sell it to marketing agencies. We don’t have to worry about it. And that actually is actually more work, I think, than selling to startups directly and like actually managing it for them. I think we went through the cycle of like, I think this is the typical thing. I think PG might’ve given this advice is like, Hey, you start with, you know, if you try to sell it to like a, sell a product to a legal firm or something, and they don’t want your product, but you really believe that that product’s going to work. And maybe you just start your own legal firm and then you compete with them. Harsha Vankayalapati (00:55:21 continued): and you prove that it works super well, right? And so like we did that cycle over the course of like two to three months. I think we could have just done that in the first 30 days. Like, hey, actually, let’s just prove that this tool is gonna work extremely well. Like we’ll just become the agency and do everything that we can to like be a marketing agency and then just start going after it, which is sort of what we’re doing today, right?
Brian Bell (00:55:41):
I see a lot of that around the industry, actually. Tech enabled agencies, right? We backed a few. You guys, Lexi, a law firm, doing tech enabled law, right? Instead of trying to sell to law firms, we’ll just become the law firm. And it’s all AI and agentic enabled. And so we can do everything five to 10x more efficiently and lower cost and higher outcomes. Yeah. It’s kind of a business model right now. I don’t know what to call it. It’s like agentic, verticalized, tech enabled services or something.
Harsha Vankayalapati (00:56:07):
Yeah, I mean, and I think like the forward deployed model too is also getting very in vogue from this as well, right? Because as we become more like agencies or like, I mean, the agencies become more tech enabled, like there’s more opportunity for people to actually like embed themselves in someone else’s team, right? So like you can, like in the case of like an AI enabled law firm, right? Like you can almost be like a general counsel for whoever you’re working with. super easily, like in a way that like was much more scalable than it ever was before. And like, we’re doing the same thing with marketing, right? Like we’ve become like an embedded marketer for our clients, right? Where they basically can tap us as a resource or tap our agent as a resource. And then we can generate the strong results for them.
Brian Bell (00:56:51):
Yeah. Well, Harsha, thanks so much for coming on. Learned a lot.
Harsha Vankayalapati (00:56:54):
Thanks again.
Brian Bell (00:56:54):
Thanks so much, Brian.
