Watch the interview on YouTube
Eman Khoubian spent about $1,000 on ads to get his first paying customer.
The plans were $2 and $5.
So when I asked about the relationship between customer lifetime value and customer acquisition cost, we weren't exactly discussing a machine you should pour more money into. He was losing money on that customer. He knew it. But he had built something, put it in front of another person, and gotten that person to pay for it.
That was the first goal.
By our August 2026 interview, Eman said Whisper AI had more than 6,000 paying subscribers and was doing about $2 million in annual recurring revenue, still bootstrapped. He had started without an engineering background and used AI tools to build the original product.
Obviously I wanted to know how that worked. I also wanted to know how much technical work was hiding inside the phrase “without a technical co-founder.” And I was especially interested in something he had told me before we recorded: the guy running an AI company was moving back toward humans for his content and brand work.
We went from a high-school after-prom party to debugging software, buying a $5,000 domain, questioning AI's judgment, and the bedtime stories he makes with his kids. The building was interesting. So were his decisions about what to let the tools do and what he still needed to do himself.
In this episode, you’ll learn:
- How two teenagers found a venue, sold tickets, and discovered an expense they hadn't planned for.
- Why the feature people paid for wasn't the feature Eman originally wanted to build.
- What getting an AI-built app into production actually required.
- How his costs, team, and expectations changed as Whisper AI grew.
- Why he wants to build around the transcription workflow instead of betting everything on raw text.
- What happened when roughly 200 automated blog posts failed to bring traffic.
- Why we spent so much time on judgment, failure, kids, and the difference between getting an answer and choosing what to do.
Table of Contents
- Two teenagers and an after-prom business
- The traditional route took a while
- The useful feature wasn’t his original idea
- One customer was the first goal
- Without a technical co-founder doesn’t mean without technical work
- The $5,000 domain and the cost of growing
- What happens when transcription becomes a commodity?
- The work after uploading the file
- Two hundred AI posts and no traffic
- Getting an answer is different from making a decision
- How do kids learn judgment if the tool does their thinking?
- Why spend your life building this?
- Bedtime party, failure, and the next small risk
- Questions to take back to your own business
Two teenagers and an after-prom business
Eman grew up around entrepreneurship. His grandfather had come to the United States and started investing in real estate without knowing the language, the laws, or the rules. He understood the basic opportunity: buy something and figure out what it could produce.
That was in the early 1980s, when interest rates were in double digits. So entrepreneurship was visible in Eman's family before it became his own career.
His first real experience running something came during senior year of high school. He and a friend decided to organize the after-prom party.
They found a ballroom dance studio. It had a big dance floor and mirrors, which made it a pretty useful place to put a bunch of teenagers who wanted to keep dancing. Somehow they convinced the owner to rent it to two 17-year-olds. His friend's dad signed the contract.
I wanted the numbers. Eman remembered the rent being around $1,000. They hired a DJ for about $400, brought in a security guard, and sold tickets.
He remembered prices around $20 or $30, with a deal like two tickets for $50. His graduating class was roughly a hundred people, so with guests they were looking at a couple hundred potential attendees.
They made a few thousand dollars, he said. Then they learned about cleanup.
When they went back, the owner wanted another $100 because the place was dirty. They hadn't really accounted for staying after the party and cleaning everything themselves. In that moment, paying the extra hundred sounded fine.
The part that caught my attention was how other students helped sell tickets. Someone would offer to bring a group of 10 if they could get in free. Eman and his friend were happy to make that deal.
There were different social groups, and somebody inside each group could reach people they couldn't reach as easily themselves.
A little influencer marketing, before either of us was calling it that.
Eman pointed out that this was before Facebook and before selling an event meant posting a link. They were selling tickets by hand. The incentive was simple enough that people understood it immediately: bring us paying attendees and we'll give you a ticket.
He and that friend are still friends. The party worked. But it didn't immediately turn into an events company.
The traditional route took a while
Eman did some party promotion in college, though he described that as more about getting to party for free than starting a serious business. After college, he took the traditional route.
I asked what he meant by traditional. Finance and accounting, in his case, at a large real estate company. He stayed for about six years and worked his way up.
His dad's advice had been different from his grandfather's example. The message was more along the lines of: get a job, start making money, get off my payroll.
He thought he wanted to work in real estate. Then he went to business school. He had an opportunity to join a friend's startup work but chose strategy consulting at a large firm instead.
Eman said he tried to bring an owner's mentality to those corporate jobs. What would the owner want to see? What would make this business better?
I told him I'd heard a lot of people say that. In his case, the business he'd eventually built gave the claim a little more weight.
Still, I wanted to understand the jump. How do you go from finance, real estate, and consulting to building a transcription company?
By the time he started experimenting with AI coding tools, he'd spent close to a decade investing in real estate. When the market changed in 2022 and 2023, he paused new investing. He still had a portfolio to manage, but he didn't have a full-time project giving him that same sense of purpose.
So he started playing with Replit.
The first idea was just a landing page. Something to try.
The useful feature wasn’t his original idea
I stopped him to ask what Replit was.
Eman described an AI coding environment where he could explain what he wanted in ordinary language and have the tool help build it. He liked having the coding, hosting, database, and other pieces in one place, instead of needing to understand how to connect a collection of separate services.
His description was of the tool as he was using it at the time. He also emphasized that it had changed a lot since his early experiments.
Without it, he said, he wouldn't have built Whisper AI.
After playing with a landing page, he saw a dictation company's fundraising announcement and wondered whether he could build something similar. His version would show a live transcription and translation.
If one person spoke English and another person read Spanish, the Spanish speaker could follow the translated words as they were being spoken. That was the use case he was trying to build.
Then the Replit agent added a different feature: transcription of prerecorded audio.
That part hadn't been Eman's original idea.
I asked whether it had become the main thing he was selling. It had.
People were finding the app because they had recordings, often in other languages, that they needed transcribed. So he shifted his attention away from perfecting a live meeting assistant and toward the problem those users were actually bringing him.
There are two kinds of input in that story. The coding tool suggested a feature. Customers showed him that the feature mattered.
He still had to decide which signal to follow.
Eman dated the start of the project to June 2025, about a year before we talked. It hadn't begun with a grand plan to compete in transcription. He was experimenting, saw people respond to something inside the experiment, and kept moving toward that.
One customer was the first goal
At this point, I had a question I hoped wouldn't come across as offensive.
I'd seen plenty of tools where you drag in an audio or video file and get a transcript. Why was his working? How was he differentiating himself in such a crowded market?
Eman's answer started with branding and his own determination. But he also backed up to how small the original goal had been.
Get one customer.
It was a side project. If someone signed up and paid for something he had launched, that would be a win. So he put money into ads and worked on them until he got the first paying user.
That was where the roughly $1,000 in ad spending met the $2 or $5 plans.
I asked about lifetime value and acquisition cost. We both knew those numbers didn't work as a lasting business model. He even joked that the customer probably canceled right away.
But the early spend bought him a real event to learn from: someone had paid.
I compared those first attempts at advertising to lighting money on fire and watching what happens while it burns. You hope to learn something before deciding where the next dollars go. I wasn't saying you should keep spending $1,000 to acquire a $5 subscriber.
Eman said tools like ChatGPT and Claude made it easier to work through ad setup and optimization. They didn't remove the need to experiment.
One customer became 10. Then he wanted to know whether he could get a hundred. He said he reached that first hundred within roughly two months, from June to August.
The goal kept moving because he kept getting evidence that another step might be possible.
That is a much more recognizable beginning to me than the polished founder version where somebody seems to have known the whole business in advance. He wasn't sitting there with 6,000 subscribers in his head. He was trying to get the first person to care enough to pay.
Without a technical co-founder doesn’t mean without technical work
Getting people to sign up created another job: make sure the thing they paid for actually worked.
Eman was working nights fixing bugs. He was very clear about this part. An AI tool can help you generate a landing page, but a functioning production app needs testing all the way through.
Audio transcription has real technical complexity. Things break. Users try things you didn't expect. A feature that appears to work in one situation may fail in another.
He was constantly fixing and improving the app.
Sometimes a user would report a broken feature, Eman would work through it in Replit, publish a fix, and reply within an hour saying the issue had been fixed. People liked that.
There was a useful connection between the tool and the customer relationship. AI helped him make changes quickly. Listening to the person with the problem helped him know which changes to make.
I pushed on the technical-founder idea. The way he described the work, it sounded like he had needed to become technical in at least some ways.
He agreed.
He had to learn what was going on because the tools he depended on could be wrong. Sometimes they sent him down rabbit holes. At that point, continuing to ask the same tool for the next step wasn't enough.
He had to pause and ask: what's the actual problem? What's the best way to fix it?
That requires a different kind of attention from watching an app appear after you type a prompt.
I think this distinction matters if you're listening to stories about people building software without knowing how to code. Eman didn't begin with an engineering background. He also didn't describe a company where nobody needed to understand the product, test it, or take responsibility when it failed.
He was doing a lot of that work himself.
Later, he hired an engineer. The ability to start without a technical co-founder hadn't made technical expertise unnecessary.
The $5,000 domain and the cost of growing
Eman originally launched under a different Whisper AI domain, which he described as a proof-of-concept purchase costing about $50. Then whisperai.com became available.
The price was $5,000.
For a side project that was already eating a lot of his time, that felt like a serious decision. He could spend 10 hours in front of the screen building and iterating, enjoy the work, and still wonder what he was doing with his time.
Should he be concentrating on something else?
His girlfriend kept encouraging him. Keep going. Keep pushing. You have something.
I asked how he was paying the bills during that period. His real estate investing was supporting him. The project didn't have to replace his income immediately.
That's a meaningful detail in a story like this. He was working hard, but he also had another source of support while he learned and built. The early goals were about customers, monthly recurring revenue, and getting the next feature to work, rather than needing the app to pay every household expense that week.
He bought the domain in October 2025.
Then traffic dropped.
For about two weeks, he said, it felt like the growth had stopped. He'd made what he hoped would be a big investment in the business, and the immediate result looked worse.
After that, he said traffic took off. Looking back, he considered buying the full domain one of his best decisions.
By December, another transcription company had approached him about acquiring what he was building, including the domain. Eman considered it but wanted to keep going. If someone was interested that early, he felt there was more he wanted to try.
I also wanted actual operating costs.
At the time of our interview, he put Replit agent usage and hosting at around $3,000 a month. Backend API costs varied with usage and were running around $10,000 a month. There were other smaller services and fees as well.
Those were pieces of his costs, not a complete profit-and-loss statement. Ads, people, and other expenses also belong in the picture.
As the roadmap grew, he brought on a full-time engineer. That had happened roughly two and a half months before we talked, and he called it one of his best decisions. A second engineer was planned. He had also brought in an SEO agency.
Overhead was increasing because he was choosing to invest in growth. He could have stepped back more, he said, but he wanted to put the resources back into the business.
The company was becoming something different from the solo experiment. His decisions about spending and hiring were changing with it.
What happens when transcription becomes a commodity?
I asked where he thought the transcription industry was going and what gave him optimism about it.
Eman didn't pretend raw transcription would remain scarce. He expected it to become a commodity.
There are plenty of ways to turn recorded speech into text, and he expected the larger AI platforms to keep improving at that. So the question for Whisper AI was what somebody could do beyond getting a block of words back.
Speaker identification was one example. Working with the contents of the recording was another. He wanted to build an ecosystem around speech to text, rather than make the entire business depend on offering something nobody else could transcribe.
I pushed on that too. Gemini was already transcribing our call. Why wouldn't somebody just use a general AI assistant?
We went back and forth about which tools could accept which kinds of audio at the time. Those capabilities change quickly, so I wouldn't use a conversation from August 2026 as a current comparison chart.
Eman's larger answer was that the world contains a lot of audio that doesn't originate inside a meeting with a built-in notetaker.
There are different languages, file sources, and reasons to use a recording. Some people want an easy consumer interface. Other people want to build their own app around the capability.
Whisper AI had launched an API for developers, which Eman saw as another way to serve that second group.
His optimism came from the work around the transcript. What does the user need identified? How do they want to interact with the recording? How does the result fit into what they do next?
That doesn't make the competitive question disappear. It gives the company a more specific problem to work on than “we also turn audio into text.”
The work after uploading the file
Eman gave me a new example of that workflow approach. He said they had launched a Google Drive integration the day before our interview.
Historically, a user had to log in, choose a file, select settings, start the job, and wait for the result. His goal was to remove the repeated setup.
Connect a Drive folder once. Choose how you want the files handled. As recordings land in that folder, the system picks them up and emails the transcript in the requested format, or multiple formats.
That's how he described the integration at the time. He considered it a novel step for the industry; we didn't independently establish that it was the first.
The useful part was easier to see than the first-to-market claim. If someone transcribes recordings regularly, the work around the transcription can become its own repeating chore.
The transcript itself might be available from several places. Saving someone from going through the same file-selection and setup process every time can still be valuable.
It also connected to a job I used to do.
Two hundred AI posts and no traffic
Years ago, when I was a content writer for a media company, I had to write two blog articles a day from podcast interviews.
At first, I listened and wrote while I listened. This was before transcription tools were part of everybody's normal workflow.
Then I found a tool that could transcribe an episode. Perfect.
I'd drop the hour-long podcast into it, go to the gym, and start making an outline from the transcript while I worked out. Then I'd come home and write the article.
I was letting the software do the grunt work and keeping the writing and thinking with me. Today I have ChatGPT help with work I was doing myself back then.
So I wanted to understand what Eman meant when he said he was moving toward human content and brand work.
He had tried a tool that automatically wrote blog posts around target keywords. Generate the content, improve the search rankings, set it up and let it run. That was the promise.
Eman said they ended up with around 200 blog posts and pages that brought no traffic. Some weren't being picked up at all.
I asked whether he thought Google was identifying them as AI, or whether the content just wasn't good.
Both, he said.
He thought the pages looked spammy and might even be hurting the site's rankings. That's his diagnosis of what happened to those pages. It doesn't establish that Google penalizes any piece of writing simply because AI helped make it.
Google's published guidance makes a different distinction: it focuses on helpfulness and quality, and treats automation used primarily to manipulate rankings as spam. Using AI by itself isn't a blanket violation.
Google’s guidance on AI-generated content
For Eman, the practical conclusion was to bring humans back into the writing.
I told him this was great news for copywriters.
He wasn't saying to throw away every AI tool. Use it to research a subject, pull information together, do analysis, or help think through an outline. But he wanted a human deciding how to approach the piece and doing the actual writing.
That was where he was putting more of his content investment.
The tension interested me because his company exists because of AI. He's happy to use it for transcription and technical work. He's also willing to look at a particular use of it and say: this isn't producing what I need.
He didn't owe every task in his company to the same tool just because the tool had helped him build the company.
Getting an answer is different from making a decision
Eman brought the same concern to data analysis. When Claude analyzed something for him, he still had to question the conclusion.
Had it understood the problem? Was its answer right? What was it overlooking?
I loved his point about judgment.
I'd recently heard an interview where Elon Musk made a sweeping prediction about AI being better than people at everything within a short period. I didn't remember the exact timeline, and I wasn't convinced.
Judgment was the part I wanted to keep pulling on.
People push back when I say I can tell I'm reading AI writing. They say the user hasn't prompted it well, or the tools simply need more time. Maybe some of that will change.
But I wondered whether part of what we notice is that a piece hasn't passed through the kind of judgment a human brings to it. Does the writer understand who they're talking to? Do they know which detail matters here? Can they recognize the thing that technically answers the question but misses the point?
Eman thought both things could be true. AI can keep improving at producing an answer. Choosing the answer that's right for a particular person is a different problem.
He used an example of asking AI to analyze three possible markets for a business. It might collect and analyze more information than either of us could do ourselves. You could make a scorecard and compare the options.
But somebody still has to decide which business they're actually going to build.
What do you want? What constraints are you living with? Which tradeoffs are you willing to accept? The analysis becomes input into that decision.
Eman didn't think the next major company would necessarily be a business where an entrepreneur had handed all the thinking to AI and walked away. He expected people to remain involved in making choices and building something with the output.
We weren't proving a permanent limit on what machines could ever do. We were talking about the responsibility that still sat with the person using them.
How do kids learn judgment if the tool does their thinking?
That raised a concern for Eman about the next generation.
He was 42 when we recorded. I was 37. We'd grown up before AI assistants, before social media became ordinary life, and before you could ask a chat window to complete so much of your schoolwork or job.
He'd been talking with a high-school teacher about students taking the answer from a tool at face value. It worried him.
That doesn't mean every younger person lacks critical thinking. It was a concern about what happens when a habit of accepting answers replaces the practice of asking questions.
How do you teach a kid to question a conclusion if the easier move is always to type “give me the answer”?
I wanted to go down this trail longer than we had time for.
My thought was that judgment is shaped by failure and success. You try something. It fails. You learn something. You try again with a different understanding.
You also develop intuition about the person you're speaking to. If you work with a particular customer long enough, you start noticing things about that customer that aren't captured by an abstract description of the market.
An AI can analyze accounts of other people's failures. That isn't the same thing as a person saying, “I tried this, it didn't work, and here's what I learned.”
I also caught the qualification while we talked. If a system tries things repeatedly and learns from the results, maybe it develops more of that capability. Eman expected the models to keep improving too.
His instruction for the present was to keep doing the thinking.
Ask the tool to analyze the information, then think about what you want to do with it. Don't automatically turn every consequential decision into “AI, tell me what I should do next.”
For a known form or a defined workflow, asking for the next step can be useful. For deciding what to build and why, Eman wanted the human to remain engaged.
He expected a gap to grow between people who use AI to improve their own performance and people who depend on it to supply the thinking they haven't practiced themselves.
That was a prediction, not a measured result we established. But I think the question underneath it is worth asking right now: which ability are you making stronger while you use the tool?
Why spend your life building this?
Near the end, I asked the question I hadn't gotten to yet: what is all this work for?
Eman had described nights of fixing bugs, hours in front of the screen, and a growing list of things to build. His girlfriend had been supportive. But why make this the thing he devoted so much of his working life to?
He said those conversations about purpose had started with her asking him that same kind of question.
He has three children. She has two. They're a blended family with five kids, and his family is central to his purpose.
With Whisper AI, he also wants to build something bigger than himself.
He described opening his laptop late at night and seeing a thousand people around the world using something he had built. That was an incredible feeling for him.
I pushed a little further. Why is that worthwhile enough to hang so much of your vocational life on it?
Eman talked about an internal drive to keep building and seeing what could happen next. He had reached seven figures in annual revenue. Could he eventually reach seven figures a month?
That was an ambition, not something he claimed he'd already done.
He felt he could see steps on that roadmap. Maybe he'd reach the goal and maybe he wouldn't, but he expected to learn and grow through the attempt.
The original goal had been one paying customer. By our interview, he said close to 400,000 people had signed up in total, with more than 6,000 paying subscribers. Those are different measures: registrations aren't the same as current paying users.
He also valued the messages from people who said the product saved them time. Even some people canceling would tell him the tool was great and that they intended to come back.
You can put all of that into a dashboard. It's also a person somewhere in the world telling you that something you made helped them.
I recognized the small beginning in his story. I made my first children's book for my little brother Tristan. My early goal was to get one other kid to think it was fun.
That eventually became a career touring as an author and then a company helping other people become children's authors.
You don't have to know the whole career when you're trying to reach the first person.
Bedtime party, failure, and the next small risk
Because Eman has kids, I ended with my children's-book question.
If he could make a book his kids loved, and they adopted one idea from it that could change the way they lived, what would he want that idea to be?
First he took a small tangent that I was happy to follow.
When all the kids are home, they do something called bedtime party. They have ChatGPT make a story that includes the children's names.
If sharing was hard that day, they can ask for a funny story about sharing. If somebody had been mean, they can ask for a story about being kind. The day's problem becomes part of the story, and the kids listen for the lesson.
This was the same person who wanted humans writing his company's content, using AI to make a personalized bedtime story. The job and the audience mattered to how he chose to use it.
For his own children's book, Eman would focus on taking risks.
It's okay to try something. If you fail, you can pick yourself up, learn from it, and keep going.
He thinks kids need practice seeing failure as something they can survive and use. Then, when success comes, he wants them to enjoy it and remain humble.
That fits the version of his business story we actually spent time talking about. A first app with bugs. An ad bill that didn't make economic sense yet. A domain purchase that initially seemed to hurt traffic. A willingness to bring people in where the tools weren't enough.
Those aren't instructions to take every risk you can find. They're examples of trying something, looking at what happened, and making the next decision with more information.
The kid doesn't need to start a transcription company. The kid needs room to try, notice, and try again.
Questions to take back to your own business
- What's your smallest real test? Eman started by trying to get one person to pay, then used the response to set the next goal.
- Which part of your product are people actually using? A feature the coding agent added became the use case he chose to build around.
- Where does “AI built it” still need a person to test and understand it? Shipping quickly doesn't remove responsibility for the result.
- Are you measuring the result or counting the output? Two hundred pages weren't useful to Eman if they brought no traffic.
- What happens before and after the thing you sell? His Drive integration addressed repeated setup around transcription, not only the text itself.
- Which decisions still need your judgment? Use analysis to inform the choice instead of letting a confident answer quietly make the choice for you.
- What are you teaching the people watching you try? Eman's kids will see how he handles failure and success, not just what his revenue reaches.
If something in Eman's story gives you permission to begin, I think that's useful. You can start learning before you have every skill the eventual company will need.
But pay attention to what he did after he began. He listened to users. He tested. He learned enough to question the tools. He hired expertise. He changed his mind about where AI was helping.
The next experiment can be small. The responsibility for learning from it is still yours.
Watch, connect, and read the transcript
Watch the full interview on YouTube
Read the original interview transcript in Google Drive
The edited YouTube video is 38:12; the meeting transcript ends at 38:16, so its timestamps aren't exact video seek positions. Company figures, expenses, team plans, and family ages above refer to the August 25, 2026 conversation.