From Piano Prodigy to AI Founder: Building Marketing Agents That Replace Entire Sales Teams
A deep dive with Vivian Cai, 22-year-old founder of Streaml, exploring how she walked away from investment banking at Evercore to build AI agents that automate everything before the sales meeting, and why conviction matters more than experience
At 22, most people are figuring out their first job. Vivian Cai walked away from investment banking at Evercore, where a prestigious finance career and half a million dollar salary were waiting, to build something that sounds almost too ambitious: AI agents that replace entire sales and marketing teams.
But here's what makes her story different. She didn't burn out from six hours of daily piano practice as a kid. She found creativity in repetition. She didn't romanticize the grind of investment banking. She automated it. And when everyone at Evercore complained about tedious work at a hackathon, she was the only person who actually built a product to solve it.
Now she's building Streaml, an AI native marketing platform that handles everything before the meeting. Outbound prospecting, inbound content, social media management, GEO, SEO, all orchestrated by agents trained on viral posts and platform algorithms. Her pitch is simple: companies spending 500 to 600K on five or six SDRs will soon only need one.
The twist? She found her CTO using her own tool. And despite being just one and a half weeks into fundraising with investors already asking "where's your term sheet?" her answer about giving up has never wavered: no.
Key Takeaways: What a 22 Year Old Founder Gets That Most Don't
The Finance to Founder Path:
- Finance professionals see pain points clearly but often lack technical skills to build solutions
- Working in banking teaches 10% of what running a startup does
- Best combination: banker's market analysis plus strong technical co-founder
The Real VC Equation:
- VCs care about one thing: buy low valuation, exit high valuation
- You don't need profitability, you need validated market and scale potential
- Product market fit matters, but growth trajectory matters more
The Agent Economics:
- Companies spending 500 to 600K on five or six SDRs will need just one
- Humans stay in the loop for essential execution, agents handle everything before that
- Live data scraping beats static databases when capturing sales signals
Q: You practiced piano six hours a day as a kid. Most people would burn out. What made you fall in love with that kind of repetition?
Vivian: I had a picture where I'm literally coding on the piano. I'm practicing but I have my computer on the side where I write code.
I think I treated piano as a form of creativity where somehow after when I'm playing piano or after I play piano, I feel that I have a lot of creative ideas and I will sort of switch between playing piano and doing other things.
So it's kind of like a creative medium that I have. I don't treat it as a chore per se. It really is, when people ask me oh how did you stay consistent, I think it was mostly I didn't treat it as a job or a chore, but rather something that makes me more creative where I can, it's like a way for me to almost relax or switch to a different brain state per se.
Do you still play while you're working on Streaml?
Yeah, I am a piano major at my school. I'm like two months away from graduating, but I play for my juries and exams and stuff like that.
🔥 ChaiNet's Hot Take: Most people see repetition as drudgery. Vivian saw it as a creativity unlock. That mindset shift, treating practice as fuel rather than work, is what lets someone code while playing Chopin and see no contradiction. It's also what makes her uniquely positioned to understand which repetition humans need and which AI should handle.
Q: You had three paths ahead: celebrity pianist, investment banker making half a million a year, or startup founder. Why choose the hardest one?
Vivian: I remember so vividly that when I was working at Evercore, there was this hackathon, a firm wide hackathon, and everyone working there was like oh this needs to be solved and that needs to be solved.
But at the end of the day, out of the entire office, I was the only person that actually built a product to solve that issue.
I remember I built something similar to Streaml as well as a PowerPoint automation tool. So it would be, you know how when bankers come and get comments from VPs and MDs, it's like oh change that font or change this number to XYZ. So that PowerPoint tool is an extension where you can literally put in your prompt and then it will change it automatically for you.
So I mean the gist is that instead of being the person who is doing the repetitive work, I much enjoy using AI to streamline that repetitive work and make Evercore my client instead of working there.
🔥 ChaiNet's Hot Take: Everyone complains. One person builds. That's the founder gene in action. While an office full of bankers talked about problems at a hackathon, Vivian shipped two products. Her goal isn't just to escape the grind, it's to sell the solution back to the place she left. That's not just ambition, that's strategic revenge.
Q: When you walked away from a prestigious finance career to build Streaml, did people think you were crazy?
Vivian: A little bit. A lot of my friends in banking, I don't think they knew, but after they knew, they thought it was very bold, like a very bold move.
I think especially they would think that startup is super risky in comparison to banking or the traditional banking to private equity path.
And then I also talked to a lot of them, they were like I definitely want to pursue a startup in the future but I want to gather more experience or understand how a company operates through banking and through private equity, which I 100% agree with.
I think that for example when I'm pitching to investors, I took a lot of insights from how the biotech companies I worked with, how they pitch to investors. But I feel like this is only about 10% or even less of how a startup really operates.
So honestly, I learned so much more actually working on this startup than anything that I've done in the past. So that's why I think even though it is a little bit risky, it is something that I want to pursue and I could only grow by actually building it rather than gaining more experience from say banking or PE.
🔥 ChaiNet's Hot Take: The traditional path says get experience first, then take risks. Vivian's calling BS on that. Banking teaches you 10% of what you need for startups. The other 90% only comes from actually building. Waiting for "enough experience" is just a comfortable way to never start.
Q: With so much noise in AI right now, how do VCs even decide what's worth investing in?
Vivian: Honestly, I think for VCs, all they care about is valuation. If you think about their ultimate goal, it's to buy at a lower valuation and exit at a higher valuation.
So right now it really is, and what brings the valuation of a company high is the market, whether they can actually make money. And they don't even need to be actually profitable. OpenAI, they're not profitable to the slightest. But it really needs to solve a problem and there needs to be a validated market that people would use this product.
So you need basic product market fit but you also need that your product would scale and it cannot be just a revenue positive business, it needs to be a startup.
Fair. And growth. Yes.
🔥 ChaiNet's Hot Take: VCs aren't betting on profitability, they're betting on valuation arbitrage. OpenAI loses money but commands massive valuation because the market is validated and the scale is obvious. For founders, this means stop obsessing over unit economics in pitch decks and start proving market size and growth trajectory.
Q: What's been your toughest moment so far and how did you cope with it?
Vivian: I think there's definitely, from time to time, I can't recall the most tough moment but it's definitely multiple tough moments where I had to overcome.
For example, honestly right now because we're fundraising and we're about one and a half weeks in, but there are pressures from either our previous investors or people around me that are like, oh why don't you have a term sheet yet?
Yeah, I would say right now is a tough moment that I'm going through. But looking back at the past, at how I overcame these moments previously, is to just go for it.
Instead of indulging in the feeling of pressure, I would just reach out to more investors and talk to more customers to boost our ARR. Solving the problem itself is so much better than dwelling in the mindset of oh I'm in such a tough moment.
It's to just do it and solve the problem.
🔥 ChaiNet's Hot Take: Pressure is inevitable. Dwelling is optional. One and a half weeks into fundraising and people are already asking "where's your term sheet?" That's the startup reality check. The difference between founders who make it and those who don't isn't absence of pressure, it's refusing to indulge it. Action beats anxiety every time.
Q: Do finance professionals have an advantage when entering AI because of their data mindset?
Vivian: They would know the pain points for sure, but a lot of them don't have the technical ability to build it out.
I talked to friends of my age, analysts and summer analysts. I also talked to associates and VPs that left banking and wanted to start entrepreneurship. One of them from JP Morgan actually came out and ended up still doing investing, like search fund or VC, because they simply don't know how to build a product to solve that pain point.
So I would say that would be the biggest problem. But they definitely have an advantage in terms of analyzing the market and what pain points need to be solved.
So the best combination would be a banker with someone with really strong technical skill, which I am very grateful for. My CTO, he worked at PayPal and was an excellent engineer for a really long time.
🔥 ChaiNet's Hot Take: Finance minds see problems clearly. Engineers build solutions. Neither alone is enough. The magic happens when someone who understands markets deeply partners with someone who can architect systems. That's not just a good team, that's the optimal founding configuration for B2B AI companies.
Q: How did you find your CTO?
Vivian: My CTO, I actually used our own tool. I used Streaml.
Wow, that's amazing.
I was searching for engineers who have experience in LLM and agentic workflow, data scraping, data analysis, etc. And then I saw his GitHub project on GitHub and it was super relevant to what we were doing. So I reached out.
And then super coincidentally, our investor, our first round investor, actually also talked to him previously. So they connected us.
🔥 ChaiNet's Hot Take: Using your own product to find your CTO is the ultimate product validation. It's one thing to pitch your tool to customers. It's another to trust it for the most critical hire of your company. That's not just confidence, that's proof the product actually works.
Q: Was there any time when you thought about giving up?
Vivian: No. Always been banging on that. Okay, this is the aim.
🔥 ChaiNet's Hot Take: No elaboration. No qualifications. Just no. That level of conviction at 22, one and a half weeks into fundraising with no term sheet yet, is either delusional or founder material. Based on everything else, it's the latter.
Q: What's been the worst rejection you've gotten and how did you handle it?
Vivian: I wouldn't consider it a rejection, but there was a time where we struggled with selling the product. At the time we only had an outbound tool, so it's like finding, there are literally about 200 of the same products on the market right now. That's basically a data scraper that finds you a list of people and sends them an email or a LinkedIn message.
So when I approached investors, they were like, why should we invest in your stuff? When I talked to customers, they were like, I would not even pay 50 bucks for this product. I just would not because it's not going to do anything.
So that's a moment where I realized and we made a pretty big product switch, enhancement. So right now we can do anything that a marketing agency can do, including outbound, including inbound, branding, social media management, GEO and SEO.
Because what's the goal that we're trying to get to? It's scheduling as many customer calls as possible. And to reach that goal, there are multiple steps in between that gets you to that goal.
If all you do every day is send thousands of emails through different domains, it's just going to end up in people's spam boxes.
So that's why we spent, from our initial funding to now or two weeks ago, we spent building and training our inbound agent. So it writes the best social media content that's human-like and not AI sounding per se.
🔥 ChaiNet's Hot Take: The pivot wasn't from failure, it was from customer truth. When people won't pay 50 bucks for your product, you don't have a pricing problem, you have a value problem. Vivian didn't try to convince them the outbound tool was worth more. She rebuilt the entire product to solve the actual job to be done: booking meetings, not just sending emails.
Q: Explain how Streaml actually works. What's the secret sauce?
Vivian: So for the outbound, it's you finding the customer to get them to book a call. For the inbound, it's them finding you and booking a call directly.
For our inbound agent, the process looks like this. At the beginning of the week or the start of a campaign, you throw in all the images and videos that you want to post per week. And basically the agent, and then you also tell it your objective. Say I'm trying to book customer calls, I'm trying to book calls with investors.
And then it will schedule out the whole campaign, including how many posts to post, what to post, how many comments, how many likes, how many DMs to send, as well as when to send them, what time of the day to send. It will plan everything out, and then you review it and you can start the campaign.
How we train the agent is based on all of the viral posts over the past couple months or years on different platforms, including how to optimize to different platform algorithms.
Say LinkedIn, for example, you can only post one per day to get the best engagement. And then for Twitter it's like five to six times, multiple times. And for LinkedIn, you probably only can send about 20 connection requests would be optimal, but then Twitter, you have to follow as many people.
The agent knows what's the best plan based on the training data that we gave it. So that's the inbound part.
And then the outbound part is pretty standard to what others are doing, but a couple differentiators. When we are scraping the data, we do a live scrape across different websites on the internet.
So let's say if we're looking for restaurants, then our agent is probably going to go to Google Maps or Yelp or other food review sites. But if we're looking for founders, then we're probably going to go to Pitchbook, Crunchbase, LinkedIn, etc.
Whereas all the other outbound tools, they just bought a third party database.
So you're basically live, your agent is actually visiting those websites, visiting those places and then getting the data rather than getting it from like Apollo or a fixed database system.
That's interesting. Including someone who literally just posted two hours ago that oh I'm looking for a marketing agency or oh we're fundraising. And these will be the sales signals that we capture and pull into the leads pool as well.
🔥 ChaiNet's Hot Take: Static databases are yesterday's news by definition. Live scraping captures signals as they happen, two hours ago, not two months ago. When a founder tweets "we're fundraising" or a company posts "looking for a marketing agency," that's a hot lead. Dead databases can't compete with real time intent signals.
Q: If Streaml becomes the marketing AI agent everyone uses in five years, what happens to sales and marketing jobs?
Vivian: It will get decreased. So right now people might spend 500 to 600K hiring about five or six SDRs, considering that they're an SMB. In the future, they only need to hire one.
Would we even need 90% of current office jobs if these agents become perfect?
Honestly, I believe that there always needs to be some parts in the process where human is involved. I believe it's unrealistic or impossible that humans are completely out of the loop.
But everything that's before the human essential execution, we can use agents to automate.
So for example, right now what we are automating is everything before the meeting. All our clients need to do is literally sit on the meetings and attend the calls. Everything before that we handle with the agents.
So in the future it might be something different. Maybe we have a super AI that maybe it's not me talking right now, it's an AI talking. But further down the process, there probably still needs a human to be involved.
So our product will continue to revolutionize where we will streamline basically anything that can be streamlined or is tedious at that moment.
🔥 ChaiNet's Hot Take: The job isn't disappearing, it's compressing. Five or six SDRs become one. But that one person isn't doing busywork, they're doing the human essential part: the actual conversation. The agents handle prospecting, sequencing, content, optimization. Humans handle connection. That's not job destruction, that's job evolution.
Final Thoughts: Conviction Over Experience
Vivian's story challenges almost every conventional wisdom about starting companies.
You don't need years of experience. You need conviction and willingness to learn fast. Working on a startup teaches you 10x more than observing from inside a corporation.
You don't need to be technical to start an AI company. You need to understand the problem deeply and find a technical co-founder who can build. The banker plus engineer combination might be the optimal founding team for B2B AI.
You don't need profitability to raise money. You need validated market demand and obvious scale potential. VCs are buying valuation arbitrage, not quarterly earnings.
You don't need to avoid all competition. You need to solve the actual job to be done better than alternatives. Two hundred outbound tools exist because they're all solving the wrong problem. Streaml pivoted to solve the real one: booking meetings, not sending emails.
And you definitely don't need to give up when investors ask "where's your term sheet?" one and a half weeks into fundraising. You need to keep reaching out, keep talking to customers, keep solving the problem instead of dwelling on the pressure.
The future of sales and marketing isn't about eliminating humans. It's about eliminating everything tedious that happens before the human conversation. Companies will spend less on armies of SDRs and more on skilled closers who show up to meetings agents already booked with qualified prospects.
At 22, Vivian is building that future. And based on her track record of being the only person at Evercore who actually built solutions instead of just complaining, betting against her seems like a mistake.
Q: How can people connect with you and learn more about Streaml?
Vivian: I'm very active on LinkedIn and always happy to talk to founders, investors, or potential customers about what we're building at Streaml. Feel free to reach out if you're interested in AI marketing automation or if you're trying to book more meetings without hiring an army of SDRs.
Final words: The bridge from piano prodigy to AI founder isn't as strange as it sounds. Both require understanding repetition, knowing which practice makes you better and which just wastes time, recognizing patterns, and having the conviction to keep going when everyone else thinks you're taking the hard path. Vivian walked away from half a million dollars and a prestigious finance career because she saw a better way to solve repetitive problems. Now she's building agents that let entire teams focus on what actually matters: the conversations that close deals.
Related Shorts
Explore short-form videos that expand on the ideas covered in this blog.