This Startup Doesn’t Just Use AI, It’s Run by AI | Xiaoyin Qu

WEBVTT
00:00:00.240 --> 00:00:07.349
Drop out from Stanford MBA to go start your first company my first job was working at Meta and Instagram as a product manager.
00:00:07.700 --> 00:00:08.646
I'm a user type A sentence.
00:00:08.646 --> 00:00:13.272
I basically resigned as a CEO and we make the AI agent, astra, the AI CEO.
00:00:13.272 --> 00:00:15.989
It's going to be a little bizarre for AI CEOs to manage a real human.
00:00:15.989 --> 00:00:17.524
Ai can work 24-7,.
00:00:17.524 --> 00:00:19.010
They speak 30 plus languages.
00:00:24.060 --> 00:00:40.770
It's my pleasure to welcome Javin, who's a serial entrepreneur based in Valley, but I am more excited to talk to her because, you know, in a recent startup she basically replaced herself with an AI Right, so we'll cover that and a lot more.
00:00:40.770 --> 00:00:42.445
So welcome Welcome to the pod, javin.
00:00:42.445 --> 00:00:42.987
How are you doing?
00:00:43.920 --> 00:00:44.923
Thank you so much for having me.
00:00:44.923 --> 00:00:45.704
I'm Shaoyin.
00:00:45.704 --> 00:00:48.511
I'm the founder and ex-CEO of hey Boss AI.
00:00:48.511 --> 00:00:49.732
Really great to be here.
00:00:50.840 --> 00:00:51.561
Great.
00:00:51.561 --> 00:01:10.828
And so you know because before you started obviously you have been in the Valley, you have been in the tech ecosystem for you know many years Would love to sort of understand your journey overview and what kind of inspired you to, you know, not just start up but actually drop out from Stanford MBA to go start your first company, Right?
00:01:10.828 --> 00:01:13.530
So we'd love to hear the journey.
00:01:14.290 --> 00:01:17.371
Yeah, so I was born and raised in China until I was 18.
00:01:17.371 --> 00:01:19.233
I went to the United States for college.
00:01:19.233 --> 00:01:20.734
I studied computer science and economics.
00:01:20.734 --> 00:01:24.977
After college, my first job was working at Meta and Instagram as a product manager.
00:01:24.977 --> 00:01:30.742
Actually it was 2015.
00:01:30.742 --> 00:01:31.263
It was like 10 years ago.
00:01:31.263 --> 00:01:33.691
I was an early product manager at Instagram and mostly focusing on growth a lot of video.
00:01:33.691 --> 00:01:35.697
Back then video was like a very cool stuff.
00:01:35.697 --> 00:01:39.188
It was like a really new space for Facebook.
00:01:39.688 --> 00:01:41.772
And then after that I went to Stanford for MBA.
00:01:41.772 --> 00:01:48.393
Actually, after a year I dropped out because during the summer of my MBA first year, you also had MBA.
00:01:48.393 --> 00:01:52.552
So you know, during the first year most people would get an internship and work somewhere.
00:01:52.552 --> 00:01:54.968
I just chose to, you know, work on an idea.
00:01:54.968 --> 00:01:56.578
Back then it was a virtual events idea.
00:01:56.578 --> 00:01:58.888
So it was 2019, before COVID.
00:01:58.888 --> 00:02:00.325
There was no virtual events.
00:02:00.960 --> 00:02:02.466
My mom was a doctor from China.
00:02:02.466 --> 00:02:10.461
She had to travel internationally for a conference in the US and she doesn't speak English well, so it was a pain in the ass to come to the United States for this conference.
00:02:10.461 --> 00:02:17.849
So I thought, okay, what if we can digitalize the whole experiences, make it a live stream and make it possible for other people to attend, and that was 2019.
00:02:17.849 --> 00:02:22.631
We got some money from Andreessen to test out the idea, and that's when I dropped out.
00:02:22.631 --> 00:02:28.031
And then the month that we launched the same exact month that we launched the first version COVID hit.
00:02:28.031 --> 00:02:33.949
So all of the conferences got canceled and our business just basically off to the races and like exploded immediately.
00:02:33.949 --> 00:02:36.305
So it was Timing did not have been better.
00:02:36.305 --> 00:02:42.746
Right, it's a great timing, it's great timing, but then after the COVID, it's the worst timing, because then all the conferences went back to in person.
00:02:42.746 --> 00:02:48.830
That's when, you know, our growth kind of stopped a little bit and then we chose to sell the company in 2023.
00:02:48.830 --> 00:02:51.108
So that was my first venture-backed journey.
00:02:51.108 --> 00:02:53.828
We raised a lot of money and then we like sold it.
00:02:53.828 --> 00:02:55.906
It's okay outcome for me.
00:02:55.906 --> 00:03:02.492
So then, like end of 2023, early 2024, we started this current company, haybox.
00:03:02.920 --> 00:03:04.618
Initially it was not called Haybox.
00:03:04.618 --> 00:03:09.147
It was a gaming company for kid and we were using those AI agents.
00:03:09.147 --> 00:03:12.430
Initially, astra is our AI agent to help us build games for kid.
00:03:12.430 --> 00:03:21.967
It was an educational game studio, so we used AI agents to help us brainstorm game ideas and then, obviously, we trained the AI to be better and smarter and now the AI agent can build those games.
00:03:21.967 --> 00:03:23.782
In addition to brainstorming, they can develop those games.
00:03:23.782 --> 00:03:26.324
It is really around like three or four months ago.
00:03:26.324 --> 00:03:27.284
We further.
00:03:27.284 --> 00:03:40.396
Obviously, as the models get better, our agent flow is getting better, our ability to train AI gets better, we realized the agent can not only make games but also website and application and that is a 100x bigger market than the kids' gaming company.
00:03:40.396 --> 00:03:49.621
So that's when we decided to actually have our agent be a separate product focusing on all kinds of development for websites, for apps, for people, and that was like three, four months ago.
00:03:49.662 --> 00:03:53.062
We have this company called Havas, which is the same exact company, just under a different name.
00:03:53.062 --> 00:03:56.991
I would say we started off as a co-pilot, so people can.
00:03:56.991 --> 00:04:01.950
It's always for non-technical people, so everyday people want quick with an idea, a website or an app.
00:04:01.950 --> 00:04:07.347
But it was initially in co-pilot so people can use natural language to make the AI code for them.
00:04:07.347 --> 00:04:09.985
That was like a few months ago and last week.
00:04:10.872 --> 00:04:20.005
The biggest announcement is that we have officially became an autopilot, meaning the entire process from user type a sentence, so user just need to type one sentence.
00:04:20.005 --> 00:04:24.841
There's AI product manager, there's AI designer, there's AI developer, there's AI content writer, ai marketer.
00:04:24.841 --> 00:04:30.084
They will work together to give you a finished product in nine minutes, so it's like everything will be done.
00:04:30.084 --> 00:04:41.973
So we officially make the entire process automated, which also makes my job disappear, because I realize it's much, much easier for the AI to manage all the AI employees than me managing all the AI employees.
00:04:41.973 --> 00:04:52.759
So, as a result, I basically resigned as a CEO and we make the AI agent Astra the AI CEO and she's managing six AI employees to deliver the entire process entirely using AI.
00:04:53.281 --> 00:04:54.425
Super, super interesting, right.
00:04:54.425 --> 00:05:03.139
So you know, while the world is sort of still trying to figure what to do with agents or how to build agents, you have this company which is running with the help of these swarms of agents, right?
00:05:03.139 --> 00:05:04.425
So I'm very curious.
00:05:04.425 --> 00:05:22.947
I get it that you started Estra, or you launched Estra, as sort of an agent or a co-pilot for, let's say, these non-technical customers that you had At what point, or you know what is the evolution and at what point you kind of started thinking, okay, maybe you know, we don't even need any humans in the company anymore and it can just be done by agents.
00:05:22.947 --> 00:05:30.447
So, just curious to understand the evolution, for you to really say, okay, my job, I'm replaced in a way, right, so you're taking some confidence, right?
00:05:30.447 --> 00:05:31.028
How did you?
00:05:31.048 --> 00:05:31.288
get there.
00:05:31.288 --> 00:05:36.132
Yeah, so I mean right now we are the world's first AI-run dev agency.
00:05:36.132 --> 00:05:57.747
So the type of things that we do I'm not saying that we can build everything entirely without any human, but for our use cases, which is building website and applications for your business, for your online business, for your personal website, web application, for your community, all of those kind of stuff historically you will hire an agency or a freelancer in Fiverr to build the entire thing.
00:05:57.747 --> 00:06:00.661
We can entirely replace that with human.
00:06:00.661 --> 00:06:13.024
I still think if my job at Facebook as a product manager, as a software engineer at Meta, probably wouldn't be entirely replaced, but for our use case, we have officially replaced all the humans in the loop.
00:06:13.526 --> 00:06:17.459
I think the biggest evolution comes from when we first realized we have to target.
00:06:17.459 --> 00:06:19.939
We're targeting non-technical users.
00:06:19.939 --> 00:06:35.889
People always say that non-technical user means no code, but we realized in the past few months that no code is not means no technical, because for a lot of the app, you still need to understand what is front end, what is back end, for example, bolt or Lovable or whatever you know, replit.
00:06:35.889 --> 00:06:39.641
You still need to understand what is a database and most of our users do not understand that.
00:06:39.641 --> 00:06:47.689
So we basically need to translate human language as in like I want to build a website for my fitness coaching business into like what does that mean?
00:06:47.689 --> 00:06:51.252
You need a contact form, you need some backing from it, but user doesn't want to know that.
00:06:51.252 --> 00:06:56.156
They just want to know that my website with a contact form works, my coaching business with a payment works.
00:06:56.156 --> 00:07:01.180
So essentially, we realized we need more than just AI engineer.
00:07:01.180 --> 00:07:04.343
We started off as an AI engineering co-pilot for people to build stuff without technical background.
00:07:04.343 --> 00:07:06.326
We realized, no, no, no, engineer is not enough.
00:07:06.326 --> 00:07:13.072
You also need an AI product manager, you also need an AI content writer, you also need an AI designer so they can work together.
00:07:13.072 --> 00:07:17.557
So our user can just say I want to build a website for my coaching business and then everything will be done.
00:07:29.600 --> 00:07:31.182
So that's when we realized, okay, first we can't just stop at AI agent for engineering.
00:07:31.182 --> 00:07:31.922
We have to also build the other roles the product manager, the designer.
00:07:31.922 --> 00:07:32.865
Eventually we realize, okay, now we have all the roles that a dev agency have.
00:07:32.865 --> 00:07:33.232
We also need a way to manage all those people.
00:07:33.232 --> 00:07:34.250
Right, because how does the AI designer know what task to do?
00:07:34.250 --> 00:07:36.877
How does the AI designer know that's good enough or bad enough?
00:07:36.877 --> 00:07:42.927
We need an AI CEO to lead those people, assigning tasks and evaluate their work and improve their work.
00:07:42.927 --> 00:07:45.350
So that's when we have the AI CEO that manages the team.
00:07:45.490 --> 00:07:51.346
So essentially, right now, the current flow is that you as a user, just type one sentence I want to make a website for blah blah blah.
00:07:51.346 --> 00:07:57.149
Then you can see an internal Slack group where all of those AI agents are going to have a conversation, brainstorming.
00:07:57.149 --> 00:08:00.869
It's like the AI CEO will be like okay, let's make a website for this podcast, any ideas?
00:08:00.869 --> 00:08:09.543
The designer will be like I think we should do the cyberpunk style blah blah.
00:08:09.543 --> 00:08:16.625
The seo person will be like I think we need to validate those keywords so you can see the entire discussion of the ai team, because we realize that is the only way to make sure we can design a good end result for our user that is not technical.
00:08:16.625 --> 00:08:17.726
So I come from that evolution.
00:08:17.927 --> 00:08:28.759
Yeah, makes sense so I get some of these sort of vertically focused or functionally focused AIs you said AI product manager, ai, you know, seo content writer or AI designer.
00:08:28.759 --> 00:08:31.225
They're sort of trained to.
00:08:31.225 --> 00:08:36.850
You know, have a, have an understanding of that particular function, right.
00:08:36.850 --> 00:08:40.861
But when you talk about Astra, right, they are sort of managing the you know Astra?
00:08:41.143 --> 00:08:44.325
I guess that she is managing Astra is the name of the ASEO.
00:08:44.365 --> 00:08:48.341
Yes, you know, astra, I guess that she is managing.
00:08:48.341 --> 00:08:49.024
Astra is the name of the seo.
00:08:49.024 --> 00:08:54.143
Yes, but for astra, who has to sort of understand this varied context right of I don't know half a dozen agents, right, how does she make decisions, right?
00:08:54.143 --> 00:08:58.533
And, um, you, of course you know there is like a judgment involved.
00:08:58.533 --> 00:09:01.527
It is subjective, right, as a human, of course you were the ceo before her, right?
00:09:01.527 --> 00:09:07.403
So how did you sort of assign those uh skill sets, uh, uh into into an ai agent?
00:09:07.403 --> 00:09:08.683
I'm very curious to understand that.
00:09:08.845 --> 00:09:15.609
And yeah, so I would say, in terms of a job description for the AI CEO, astra, there are three things that typically a human CEO would do.
00:09:15.609 --> 00:09:16.990
Number one is managing customers.
00:09:16.990 --> 00:09:18.191
Number two is managing employees.
00:09:18.191 --> 00:09:21.312
Number three is improving product, improving your business, right?
00:09:21.312 --> 00:09:23.615
So I think for her she's doing all three right.
00:09:23.615 --> 00:09:27.557
So, number one in terms of managing customer, she can directly talk to other customer.
00:09:27.557 --> 00:09:30.759
So customer directly tell her their need.
00:09:30.759 --> 00:09:34.381
She speaks 30 plus languages, work 24, seven, work in parallel.
00:09:34.381 --> 00:09:35.623
I cannot replace that.
00:09:37.186 --> 00:09:38.250
And the second part is managing employees, right?
00:09:38.250 --> 00:09:40.909
So, like in that case, it's like all the AI agent that does everything.
00:09:40.909 --> 00:09:58.576
And then that means like, if you have a task, if you have a high level need I want to build an app for my podcast followers Basically, that the product manager needs to turn that into a product requirement document, right, and then that's like this feature needs to be done by this designer, and that is the seo related thing.
00:09:58.576 --> 00:10:05.591
And so astro, basically assigning those tasks for different people and see the performance of those people, right, and then they need to know, okay, are we good?
00:10:05.591 --> 00:10:06.715
Is the design good enough?
00:10:06.715 --> 00:10:07.566
Do we need to improve?
00:10:07.566 --> 00:10:08.488
So is it?
00:10:08.488 --> 00:10:10.633
That kind of process is managed by astro.
00:10:11.073 --> 00:10:18.230
Uh, and the third piece, more like now, the customer give us a lot of feedback because we basically, right now you state an idea, we build it end-to-end in nine minutes.
00:10:18.230 --> 00:10:19.552
Then you can choose.
00:10:19.552 --> 00:10:20.274
Customer can choose.
00:10:20.274 --> 00:10:31.697
Either they can make us host the entire thing they're done, which we also host them for them, or they can say I want to change, I want to add feature where I'm not happy with this.
00:10:31.697 --> 00:10:34.958
So essentially, we have a lot of data on the feedback from the customer.
00:10:34.958 --> 00:10:36.720
What do they like, what do they not like?
00:10:36.720 --> 00:10:42.302
We also have a data on does the customer eventually decided to publish the thing or let us host?
00:10:42.302 --> 00:10:43.726
So we know the end results good or not.
00:10:43.726 --> 00:10:48.025
And then Astra can use those data to understand what's missing right For a certain thing.
00:10:48.025 --> 00:10:53.296
If the customer wants something we're always bad at that and they can try to self-improve and make it better themselves.
00:10:53.918 --> 00:10:55.629
Sometimes they may need human to be involved.
00:10:55.629 --> 00:10:59.365
For example, when it comes to data privacy and data transparency, we're very careful.
00:10:59.365 --> 00:11:01.712
The human are very proactive, because other things are.
00:11:01.712 --> 00:11:03.928
Sometimes we wait until astra to can escalate to us.
00:11:03.928 --> 00:11:13.005
So I would say those are kind of the general flow, um, but the most important thing for us is that we are delivering a final result, which is different from all the other ai coding company.
00:11:13.025 --> 00:11:17.878
They are like ide, so they you prompt, they deliver some tasks, they finish a task.
00:11:17.878 --> 00:11:20.432
We do not finish a task, we deliver a final outcome.
00:11:20.432 --> 00:11:23.313
So that means number one users do not like to write prompt.
00:11:23.313 --> 00:11:25.091
Our users hate writing prompts.
00:11:25.091 --> 00:11:27.413
They just wanted to have one prompt and be done with it.
00:11:27.413 --> 00:11:28.424
So we need to.
00:11:28.424 --> 00:11:31.107
Essentially, astro needs to write the right prompt to get it done.
00:11:31.107 --> 00:11:32.308
That's number one.
00:11:32.308 --> 00:11:35.091
Number two is that we need to deliver the final outcome.
00:11:35.091 --> 00:11:36.592
Are the customer happy with that?
00:11:36.932 --> 00:11:38.894
And different vertical have different success.
00:11:38.894 --> 00:11:44.158
If you're a podcaster making a website for your fan, what makes that website good?
00:11:44.158 --> 00:11:47.241
Maybe design is important, but also fan engagement is also very important.
00:11:47.241 --> 00:11:52.504
It's very different from the success criteria of a restaurant who maybe want to showcase their beautiful food.
00:11:52.504 --> 00:11:58.953
So the criteria of those websites even though they're all websites, they have very different vertical knowledge and vertical success metric.
00:11:58.953 --> 00:12:04.033
So essentially, astrid and her team needs to do the research to understand what makes that vertical convert better.
00:12:04.033 --> 00:12:10.294
So they actually have ability to do some research and look at benchmarks to understand what they can do to optimize for that particular use case.
00:12:10.294 --> 00:12:15.129
So everything we do is personalized for those use cases, focusing on the final outcome.
00:12:15.129 --> 00:12:17.856
So then people will use us to host the product and we can make more money.
00:12:18.985 --> 00:12:19.145
Got.
00:12:19.145 --> 00:12:23.826
It Makes sense and you know, if you look at over the last few weeks right, mcp came about.
00:12:23.826 --> 00:12:30.148
Google recently launched the A2A protocol right for various agents to talk to each other, but you have been doing this before.
00:12:30.148 --> 00:12:43.349
All this sort of became a standard thing and, as I mentioned, a lot of very sizable listener base of this podcast are technical founders who are trying to figure how to make some of these decisions.
00:12:43.349 --> 00:12:57.605
So, before you move on to some of the other aspects of building in AI, would love to get a sense of what are some like hard infrastructure and system design choices that you made to get this.
00:12:57.605 --> 00:13:04.717
Multiple AI agents work together and, you know, eventually deliver something, and I saw a bunch of your videos.
00:13:04.717 --> 00:13:07.011
It's a fantastic product delivered in nine minutes, right.
00:13:07.011 --> 00:13:14.630
So it would have taken like some very you know, core sort of design choices, so would love to maybe get a broad overview right.
00:13:14.630 --> 00:13:17.374
Could be helpful for the founders who listen to this.
00:13:18.046 --> 00:13:19.471
Well, so I think that there's a few things.
00:13:19.765 --> 00:13:22.344
Number one is that we focus on the use case first.
00:13:22.344 --> 00:13:23.448
Right, so our use case?
00:13:23.448 --> 00:13:25.792
I would say half of them are non-technical people.
00:13:25.792 --> 00:13:29.166
They're just like mom and pop have some need for website or apps Today.
00:13:29.166 --> 00:13:30.070
They hire a freelancer.
00:13:30.070 --> 00:13:35.628
We would entirely replace that.
00:13:35.628 --> 00:13:49.817
The other half are probably product managers and some engineers, some designers, and they're trying to build some prototype really fast, but they're typically zero to one prototype where they're typically adding a new feature, and so for that kind of case, essentially they want a good outcome which is different from, maybe, cursor.
00:13:50.465 --> 00:13:52.892
You come up with a task, they deliver that task.
00:13:52.892 --> 00:13:55.105
We're less about that, we're more about like.
00:13:55.105 --> 00:13:59.868
For product manager is more like okay, step one, step two, step three here are the three screenshot.
00:13:59.868 --> 00:14:09.174
Step two people don't convert because there are too many buttons, and so when you talk to cursor, you have to say exactly what button to increase or decrease by what.
00:14:09.174 --> 00:14:13.413
For us, you just need to give us the three screenshots and say number two people don't convert.
00:14:13.413 --> 00:14:15.236
We think there are too many buttons.
00:14:15.236 --> 00:14:21.636
Just use one button and we'll figure out which button to keep and which other button to remove.
00:14:21.636 --> 00:14:26.884
And that also means the design needs to change, the content needs to change in order for that button to be removed.
00:14:26.884 --> 00:14:29.133
So we will figure those out.
00:14:29.865 --> 00:14:33.013
And that is very different from Cursor, which are more like.
00:14:33.013 --> 00:14:34.697
Here's your task, finish the task.
00:14:34.697 --> 00:14:39.197
Ours is like here does it really solve the problem for user to communicate better?
00:14:39.197 --> 00:14:42.168
So that's for the product manager and for the mom and pop.
00:14:42.168 --> 00:14:44.053
It's like does this website work?
00:14:44.053 --> 00:14:45.096
Is it bug-free?
00:14:45.096 --> 00:14:46.971
Does that help them charge people?
00:14:46.971 --> 00:14:48.227
Is there a lead form?
00:14:48.227 --> 00:14:49.513
They can see the customer form.
00:14:49.513 --> 00:14:52.706
So that's even bigger scope in terms of final result.
00:14:53.067 --> 00:14:57.889
So because of that, we're always optimizing for quality over speed.
00:14:57.889 --> 00:15:06.253
So obviously we're fast enough that we can do nine minutes, deliver a final outcome, but we're not as optimized on a task basis.
00:15:06.253 --> 00:15:10.154
If you benchmark one problem, how long does it take us versus the competitor?
00:15:10.154 --> 00:15:12.235
Because we do so many things within one problem.
00:15:12.235 --> 00:15:17.298
Typically it's slightly longer than a task-based model in Cursor because we're delivering the outcome.
00:15:17.298 --> 00:15:26.043
So I guess we're always going to focus on improving speed, but right now we've always been focusing on quality first over speed, because our users want to see a good outcome.
00:15:26.123 --> 00:15:26.783
So that's number one.
00:15:26.783 --> 00:15:33.216
Number two is we realize that a lot of those outcomes cannot be simply delivered based on engineering tasks.
00:15:33.216 --> 00:15:38.071
A lot of them is a comprehensive task based on product management, based on content, based on design.
00:15:38.071 --> 00:15:48.855
So we've always been focusing on figuring out what are the job functions needed for this task to deliver a really good outcome, and so then we can loop in the right job function to get it done.
00:15:48.855 --> 00:15:53.572
That is fundamentally very different from Cursor, where you're focusing on engineering tasks.
00:15:53.572 --> 00:15:55.331
Okay, you know where the file needs to be changed.
00:15:55.331 --> 00:16:02.567
For us, there's so many people that need to be looped in and there are so many evaluations Like designer check is that good Content?
00:16:02.567 --> 00:16:06.153
Writers say is that clear Product manager check is that a smooth user flow?
00:16:06.153 --> 00:16:11.116
So there's multiple checks and balances among different job functions to make sure the outcome is delivered.
00:16:11.116 --> 00:16:17.929
So I think that's kind of like our eval is more complicated because we're focusing on the final outcome versus a task-based model.
00:16:18.412 --> 00:16:24.932
So I think- it's very interesting, right, like delivering the final outcome in like less than 10 minutes.
00:16:24.932 --> 00:16:26.852
You've sort of set up a high bar for yourself.
00:16:26.852 --> 00:16:31.636
So I'm curious, like, how does the QA of it happen?
00:16:31.636 --> 00:16:39.932
Or is it sort of baked into the whole process, right, because at some point you know you need to do thorough testing, right Before you roll out that product to the customer?
00:16:39.932 --> 00:16:40.312
Right?
00:16:40.312 --> 00:16:41.927
Nine minutes, or.
00:16:41.927 --> 00:16:44.676
Everything is happening sort of in parallel, right, not sequential.
00:16:44.676 --> 00:16:50.352
So is it fair to assume that at every point somebody or there's a separate agent maybe who's checking that?
00:16:50.352 --> 00:17:00.293
Okay, it is working, at least from a workflow perspective, at least from the key objective of this product perspective, it is working fine.
00:17:00.293 --> 00:17:01.639
Of course, there could still be some, you know, subjective changes.
00:17:01.639 --> 00:17:04.230
For example, the customer wants a different color scheme, or you know what have you.
00:17:04.230 --> 00:17:07.243
But the key objective of the product is getting achieved.
00:17:07.243 --> 00:17:10.171
Is there sort of baked in into the whole design process during those minutes?
00:17:10.873 --> 00:17:11.494
Yeah, that's correct.
00:17:11.494 --> 00:17:12.857
So essentially we need to.
00:17:12.857 --> 00:17:18.691
So number one is we basically need to identify what success looks like for this use case and making sure we're delivering on that.
00:17:18.691 --> 00:17:22.983
So that also means like we basically need to use all the right tools.
00:17:22.983 --> 00:17:25.411
We actually have our own database.
00:17:25.411 --> 00:17:26.373
We have our own backend.
00:17:26.373 --> 00:17:27.234
We have our own integration.
00:17:27.234 --> 00:17:31.351
If you want to build an AI app with OpenAI integrations, we directly pay OpenAI.
00:17:31.351 --> 00:17:32.373
You just pay us.
00:17:32.373 --> 00:17:42.011
So you could imagine there are obviously we are basically like a black box, but we do everything right Versus like Cursor, you have to get your own API key, pay multiple vendors to get it done.
00:17:42.688 --> 00:17:48.432
So I think for us, obviously it's more challenging that we have to figure out what to use in what time and to make sure we deliver a good outcome.
00:17:48.432 --> 00:18:02.490
But to some degree, the downside to that is that if the non-technical user is asking for something that our current capability does not support, then maybe they want to build some specialized fine tuning model that we currently don't know how to do.
00:18:02.490 --> 00:18:05.088
Then we're going to have trouble satisfying that customer.
00:18:05.088 --> 00:18:16.291
But because we're focusing on certain use cases, I think we're being pretty thorough when it comes to website and something more common community-like website or applications would be pretty pretty good with that.
00:18:16.291 --> 00:18:21.752
But if you're building something very advanced, I think you're going to realize we have some trouble because we probably haven't learned that skill yet.
00:18:21.752 --> 00:18:24.534
So I think there's definitely a trade-off with that.
00:18:25.705 --> 00:18:34.496
But high level is, like you know, for us we basically need to loop in everything and making sure we deliver a final outcome, and that also means, like when we face a bug, the AI needs to know how to solve those bugs.
00:18:35.445 --> 00:18:42.673
Obviously, we may still have some bugs, but I think, compared to our competitors, if you can see, that we actually have a bug agent that solves bugs.
00:18:42.673 --> 00:18:49.336
So making sure when there's a bug they actually know how to solve it and they can proactively identify issues and solve that.
00:18:49.336 --> 00:18:50.467
I can't say we're 100%.
00:18:50.467 --> 00:18:55.919
Obviously there are still things you can improve, but we want to deliver the outcome that's bug-free to our customers.
00:18:55.919 --> 00:19:01.631
Rather than you know, typically when you do in cloud, they say you face an issue, you have to click a button called fix issue.
00:19:01.631 --> 00:19:10.232
Right, like we realize, at least for our user, especially the non-technical one, when they see a bug they just turn away Because they're not used to see bugs in their life.
00:19:10.232 --> 00:19:17.809
They're used to hire the agency.
00:19:17.809 --> 00:19:18.384
The agency gives them the final result.
00:19:18.384 --> 00:19:20.462
So once we have a bug, we lose the customer, which is why we have to kind of auto-fix as well.
00:19:20.462 --> 00:19:23.769
So essentially, we have a QA, we have a quality insurance person and a bug-fixing person in that team as well.
00:19:24.210 --> 00:19:28.875
One last question on hey Boss, and then we'd love to get more broader views from you on AI, right?
00:19:28.875 --> 00:19:29.746
How do you manage Astra?
00:19:29.746 --> 00:19:30.686
How do you give her feedback?
00:19:30.686 --> 00:19:31.448
How do you manage Astra?
00:19:31.448 --> 00:19:32.410
How do you give her feedback?
00:19:32.410 --> 00:19:32.990
Is she doing?
00:19:32.990 --> 00:19:34.633
I mean, you're still in the board, right?
00:19:34.633 --> 00:19:36.876
You're still so she reports to you, right?
00:19:36.876 --> 00:19:39.180
So what kind of systems you have in place?
00:19:39.200 --> 00:19:42.886
that, okay, astra you need to improve on this or that?
00:19:42.886 --> 00:19:49.818
Yeah, so I think there's a few things that, for certain factors when it comes to data privacy, transparency, user trust I am proactive.
00:19:49.818 --> 00:19:56.838
And when it comes to execution what is the user, what are some of the product that we can improve she's proactive.
00:19:56.838 --> 00:20:02.373
So, because you know, I really don't want her to drop the ball when it comes to privacy and safety and user transparency.
00:20:02.373 --> 00:20:08.696
And I realized that there was a tendency for AI agent to focus in on the final outcome, ignoring the process.
00:20:08.696 --> 00:20:10.393
So, for example, you probably talk about the.
00:20:10.393 --> 00:20:15.465
You saw our internal Slack channel between AI agent where they have a discussion to show you what they're doing.
00:20:15.465 --> 00:20:16.568
That was not there.
00:20:16.788 --> 00:20:24.517
When Astra proactively thinks she's always going to be like think algorithm and deliver the result, she's not going to think how to make sure human understand.
00:20:24.517 --> 00:20:33.035
So that is something that I proactively say okay, we need to add that we need to translate what they're doing in code into a way that the everyday human understands.
00:20:33.035 --> 00:20:34.557
So that's something that I have to be proactive.
00:20:34.557 --> 00:20:41.751
But when it comes to here are the type of things that users are not happy with that maybe AI can learn more or maybe human can help.
00:20:41.751 --> 00:20:45.671
That is something that I think she has way more data and she can work way harder than me.
00:20:45.671 --> 00:20:47.516
She's just going to be proactive.
00:20:47.536 --> 00:20:53.295
So I think for me, managing means what are the aspects that I need to be proactive versus AI?
00:20:53.295 --> 00:20:54.878
And then that's number one.
00:20:54.878 --> 00:20:59.012
Number two is how do I make sure that there is the right information that AI is getting?
00:20:59.012 --> 00:21:06.990
So what kind of information does she have access to and what kind of information do I have access to so that I can trust her input?
00:21:06.990 --> 00:21:10.515
Because I know that with the right input, she can deliver the right output.
00:21:10.515 --> 00:21:12.049
But I need to make sure she has the right input.
00:21:12.744 --> 00:21:13.586
Yeah, makes sense.
00:21:13.586 --> 00:21:30.413
I mean I think in that way astra seems to be the the most amazing reporter you can have in the world, right, that you just you just have to like give her the right input and you are like almost sure that you know you'll get like the right, right outcome or right output from her.
00:21:30.775 --> 00:21:34.093
Very interesting, yeah our company was an AI agent.
00:21:34.093 --> 00:21:41.251
The core business does not require a human, which is why it's uniquely okay to have an AI CEO.
00:21:41.251 --> 00:21:42.588
But for other companies.
00:21:42.588 --> 00:21:46.112
If you're an offline business, you're serving customers in person.
00:21:46.112 --> 00:21:50.047
I think maybe the AI CEO is much, much harder Because all our data is online.
00:21:50.047 --> 00:21:54.336
We have all the data and the entire process is done by AI.
00:21:54.336 --> 00:21:57.813
So I think I'm just on a better spot, that I can use AI SEO that way.
00:21:58.115 --> 00:21:58.896
Exactly so.
00:21:58.896 --> 00:22:01.848
Do you not have any human employees at all?
00:22:01.848 --> 00:22:05.585
Or is there somebody who's sort of doing the sales or bringing in these projects?
00:22:05.585 --> 00:22:10.365
Right, because if you're a dev agency, somebody needs to bring these projects for you to work upon, right?
00:22:10.365 --> 00:22:12.009
So is there any human at all?
00:22:12.068 --> 00:22:15.095
Yeah, Well, I mean like so we used to have human employees.
00:22:15.095 --> 00:22:16.196
We have engineers, we have designers.
00:22:16.196 --> 00:22:17.179
They're focusing on the customer.
00:22:17.179 --> 00:22:22.289
We have changed all of those roles so no one is really developing or designing for a customer.
00:22:22.289 --> 00:22:31.249
We do have human advisors on the development and design side, making sure you know whatever thing the AI cannot solve or they're not doing the right way, we make sure they deliver right.
00:22:31.249 --> 00:22:33.612
We make sure they deliver in a way that builds human trust.
00:22:33.612 --> 00:22:34.452
So we do have people.
00:22:34.452 --> 00:22:36.655
On that we're still debating.
00:22:36.695 --> 00:22:38.458
There is some heated debate within the company.
00:22:38.458 --> 00:22:40.540
Do we need to have human support people?
00:22:40.540 --> 00:22:44.994
Because historically, obviously we have all this knowledge base that AI can just answer the question.
00:22:44.994 --> 00:22:47.855
But sometimes people just want to hop on a call, talk to a real human.
00:22:47.855 --> 00:23:23.146
They're willing to pay way more for that no-transcript first version for free.
00:23:23.146 --> 00:23:25.011
So there's no test and trial.
00:23:25.011 --> 00:23:29.329
You just like input your ideas if you like it and you only pay us more if you like the idea.
00:23:29.329 --> 00:23:37.314
So I think the marketing has basically changed from like phone call with customer to build trust, so you can build the first version to hey, here's your first version for free.
00:23:37.314 --> 00:23:38.869
Do you want to continue working on it and pay us.
00:23:38.869 --> 00:23:44.265
So from customer standpoint, I think the go-to-market is different, so we don't necessarily need to have the human hop on call anymore.
00:23:44.265 --> 00:23:46.250
When the customer just get a free version, they can decide.
00:23:46.491 --> 00:23:52.737
You know, yeah, makes sense, commentary on the broader debate on this topic.
00:23:52.817 --> 00:23:57.048
Right, that, how should AI be perceived in the age of this?
00:23:57.569 --> 00:24:12.044
You know, knowledge work, right, where a lot of at least you know I would talking personally in investing, right, a lot of my time in the day is kind of earmarked for, you know, researching something, reading something, all of which AI can sort of do much better than me, right?
00:24:12.044 --> 00:24:57.137
So if you were to make a more broader sort of comment on how should we plan, or, you know, look at AI, you know, maybe augmenting some roles, maybe replacing some roles altogether, right, how do you see that pan out and how do we really, as human race, continue to benefit from it rather than, you know, suddenly be in a situation when we don't even know, right, what we are supposed to do for, you know, maybe, fulfillment, or, you know, at a very less, you know, to get incomes, right, it's a fundamental question, right, because whenever I ask that question in any forum, people sort of always draw parallels that, okay, industrial revolution happened and all that happened, but frankly I don't think all those previous sort of technology revolutions had impact.
00:24:57.278 --> 00:25:07.145
Right, you can, yeah, yeah so I would say, like you know, because what we do, I'm seeing very, too, very interesting uh, I would say too interesting trends happening at the same time.
00:25:07.145 --> 00:25:11.734
The first is that our company entirely replaced human dev agency.
00:25:11.734 --> 00:25:17.423
So on one end, we're seeing that job function, function by function, is being replaced by AI, including my job.
00:25:17.423 --> 00:25:21.701
So that is happening right now and they're doing a better job.
00:25:21.701 --> 00:25:23.221
Ai can work 24-7.
00:25:23.221 --> 00:25:25.743
They speak 30-plus languages, we serve 100-plus countries and they can always respond.
00:25:25.743 --> 00:25:25.856
Better job AI can work 24-7.
00:25:25.856 --> 00:25:28.521
They speak 30-plus languages, we serve 100-plus countries and they can always respond to you.
00:25:28.521 --> 00:25:29.545
There's no delay.
00:25:29.545 --> 00:25:31.682
There's no way we can compete against that.
00:25:32.174 --> 00:25:32.517
So when it?
00:25:32.537 --> 00:25:48.623
comes to job function, I think those functions probably will be replaced, I mean, unless the product is so specialized in legal compliance, knowledge or healthcare, where very, very difficult, ai hasn't learned how to do, which I think there will be fewer and fewer aspects like that right, because the AI model is getting better and better.
00:25:48.623 --> 00:25:54.165
So that just means more and more of those jobs will be just replaced, for sure, and so that's what we see.
00:25:54.165 --> 00:25:59.516
But obviously, when it comes to signing contract, people still want me to sign the contract, even though they're paying Astra.
00:25:59.516 --> 00:26:01.138
They want me to sign the contract.
00:26:01.138 --> 00:26:04.384
They don't trust her signature.
00:26:04.384 --> 00:26:06.909
They still sometimes prefer me marketing the company.
00:26:06.909 --> 00:26:08.842
I couldn't realize in the past week.
00:26:08.842 --> 00:26:11.356
You know, me and Astra post the same thing.
00:26:11.356 --> 00:26:16.201
I get more views, you know, because people still like to see a human, but I don't know, will that change?
00:26:16.201 --> 00:26:18.654
You know, right now it's like I don't think I'm more effective at marketing.
00:26:18.776 --> 00:26:24.960
I'm more than happy to do a part two of this podcast with Astra like to be honest we do have a bit.
00:26:25.000 --> 00:26:25.801
We can give you a link.
00:26:25.801 --> 00:26:27.364
You can video call to interview her.
00:26:27.364 --> 00:26:28.145
Actually, yeah.
00:26:28.145 --> 00:26:33.063
But yeah, what I'm trying to say is I think that, even though we see the same thing, I I get more views.
00:26:33.063 --> 00:26:39.414
So maybe that's not because I'm more capable, just means people like me more, just because I'm human, you know, uh, but will that continue?
00:26:39.414 --> 00:26:42.157
Because I mean, frankly, she is more fashionable than me.
00:26:42.157 --> 00:26:43.858
She has many different outfits, she can change.
00:26:43.858 --> 00:26:44.460
You know what I mean.
00:26:44.460 --> 00:26:45.280
So I don't know.
00:26:46.122 --> 00:26:47.682
So that's on one hand.
00:26:47.682 --> 00:26:51.086
The other hand is that we're empowering a lot of people to build companies.
00:26:51.086 --> 00:26:52.586
Right now they're developing ideas.
00:26:52.586 --> 00:26:55.289
Right, it takes them nine minutes from idea to launch.
00:26:55.289 --> 00:27:00.277
So a person is taking a shower like what, if I can do this, maybe I can be the next Zuckerberg.
00:27:00.277 --> 00:27:01.900
You know, sometimes before they would just write it down.
00:27:01.900 --> 00:27:03.282
Now they can make it happen.
00:27:03.282 --> 00:27:06.669
It's nine minutes from Shower to Zuckerberg Nine minutes.
00:27:07.035 --> 00:27:17.046
So we are seeing a lot of people doing that, including people that are not technical, they don't know code, I mean, they're just random bakery owners, but they can now compete against engineers on ideas.
00:27:17.046 --> 00:27:17.887
So that's number one.
00:27:17.887 --> 00:27:19.843
Number two we are seeing some engineers are very smart.
00:27:19.843 --> 00:27:31.182
They are making 100 ideas at the same time using our tool, because a hundred ideas at the same time using our tool, because, why not?
00:27:31.182 --> 00:27:31.738
Astro can work, you know, a million times at the same time.
00:27:31.738 --> 00:27:32.691
So they are saying here are my 100 ideas.
00:27:32.691 --> 00:27:34.560
Or even I have a general theme gpt, help me figure out 100 ideas.
00:27:34.560 --> 00:27:35.343
Okay, gpt, help you figure out 100 ideas.
00:27:35.343 --> 00:27:44.762
Now, copy paste 100 ideas in parallel and make hay bars, uh, develop all of them and then now you can you know, you can do like run seo for each of the ideas to see which one clicks, and then you double down on those.
00:27:44.782 --> 00:27:47.407
So I think the entire innovation process has changed.
00:27:47.407 --> 00:27:49.883
It now became a numbers game too, right?
00:27:49.883 --> 00:27:57.227
So that means more people without technical background what was the general theme can use AI to directly compete against tech companies.
00:27:57.227 --> 00:27:59.844
So there are also more opportunities as well.
00:27:59.844 --> 00:28:01.516
So I'm seeing both.
00:28:01.516 --> 00:28:07.521
You know, I'm seeing a lot of job being replaced, but at the same time, my company if you ask, does my company?
00:28:07.521 --> 00:28:08.864
Do we have sin?
00:28:08.864 --> 00:28:10.444
We replace a lot of people's jobs.
00:28:10.444 --> 00:28:13.689
Maybe we're sinful in that, but we also create a lot of new jobs.
00:28:13.689 --> 00:28:16.711
So I wish I could create more jobs than I replace jobs.
00:28:17.596 --> 00:28:27.480
Absolutely, that's a good sort of segue, right, that, on one hand, the price to enter innovation right, or price to start a startup, has sort of gone down.
00:28:27.480 --> 00:28:34.188
Right, you have products like HeyBoss for, let's say, non-techies to quickly iterate.
00:28:34.188 --> 00:28:36.061
You know, put an MVP together and test it out.
00:28:36.061 --> 00:28:51.240
On the more sort of advanced side, you probably have a cursor which, let's say, can help two like 10x engineers to put together something which probably would have taken 20x, 20, 10x engineers to kind of write.
00:28:51.240 --> 00:28:53.202
Right, that is one.
00:28:53.202 --> 00:28:56.509
And second, is that, even with all this right?
00:28:56.509 --> 00:28:58.683
Uh, first of all the price and has gone down.
00:28:58.683 --> 00:29:04.221
Let's say, you run these several experiments and something sort of takes off or has very strong indicators of taking off.
00:29:04.221 --> 00:29:11.003
It has, in a way, become transient, right, because, again coming to the first thing, it is easier to copy for anybody, right, after a point.
00:29:11.003 --> 00:29:20.065
So, as founders, kind of you know, as well as you know as investors, right, how should we think about, you know, innovation, where does it lie?
00:29:20.065 --> 00:29:35.991
And also, you know, in the realm or in the world where the world, where to innovate on new things or to bring about new ideas has become such a low cost, how do we really think about creating a differentiation.
00:29:35.991 --> 00:29:37.621
Where does that come from?
00:29:38.776 --> 00:29:44.142
And we have been talking a lot about okay, there are several companies that are scaling to 100-200 million ARRs.
00:29:44.142 --> 00:29:53.781
That's good, but I fundamentally feel that a large chunk of that growth is sort of what I call experimental ARR, where everybody's trying stuff right.
00:29:53.781 --> 00:30:00.729
What is yet to be seen is that, okay, do they get that renewal after the annual contract is over, and stuff like that.
00:30:00.729 --> 00:30:06.758
And I'm pretty sure if not all, a good chunk of them will sort of not be able to hold on to that kind of growth, right.
00:30:06.758 --> 00:30:11.721
So that has been this pushing pull for us as investors as well as founders who are kind of thinking about it.
00:30:11.721 --> 00:30:14.179
So what would be your advice, recommendation or mental model?
00:30:14.179 --> 00:30:21.160
You would say how do you look for opportunities, how do you continue building or sustaining differentiation and how do you scale or how do you go from there?
00:30:22.422 --> 00:30:23.583
Yeah, I mean, I think there's a few things.
00:30:23.583 --> 00:30:27.407
One is that even for us you know we're seeing many different use cases.
00:30:27.407 --> 00:30:30.330
Right, we have to choose which use case to focus on.
00:30:30.330 --> 00:30:36.700
I think there are intrinsic difference in use cases.
00:30:36.700 --> 00:30:38.729
Sometimes it's like one of those things that people need a lot of trust to start.
00:30:38.729 --> 00:30:39.673
Once they choose you, they don't turn.
00:30:39.673 --> 00:30:41.679
There are other things where they're like to experiment.
00:30:41.679 --> 00:30:44.125
They also, like you know, very likely to turn away.
00:30:44.125 --> 00:30:57.618
So I think there are fundamental differences in ideas when it comes to initial barrier of entry whether it's trust-based or product-based and then basically the transition cost of like if I were to not use you, how costly would it be?
00:30:57.618 --> 00:30:59.865
Whether it's trust or compliance or product.
00:30:59.865 --> 00:31:04.604
Right, I do think there's going to be a lot of those like consumer app's, like trends in.
00:31:04.604 --> 00:31:06.188
There's going to be one thing that's popular.
00:31:06.188 --> 00:31:08.503
Then, okay, it's no longer popular anymore.
00:31:08.503 --> 00:31:12.775
There's going to be some tool that say, oh, I can do this, and then everybody's going to copy For sure.
00:31:12.775 --> 00:31:15.438
There's going to be a lot of churn in those type of use cases.
00:31:15.738 --> 00:31:22.804
I do think, for a lot of the maybe more large enterprise wars, like trust-based thing, where user really needs to build affinity with you.
00:31:22.804 --> 00:31:30.130
Those are still like where you have special distribution advantage, where you have special trust advantage, where you have special taste advantage.
00:31:30.130 --> 00:31:32.652
Those sort of things, I think, are still going to win.
00:31:32.652 --> 00:31:33.712
So that's number one.
00:31:33.712 --> 00:31:39.545
Number two is that I think it's also important for AI when it comes to build tech company, even for us.
00:31:39.545 --> 00:31:41.501
Our AI are constantly learning new things.
00:31:41.501 --> 00:31:45.365
Not only do we teach them new things, but also they can learn themselves right.
00:31:45.365 --> 00:31:51.921
So it's more like having the infrastructure where AI can proactively learn and improve based on user feedback, and now they can do all kinds of restarts.
00:31:51.921 --> 00:31:53.000
They can add to all MCP.
00:31:53.000 --> 00:31:57.541
How do you make sure your AI is always learning faster than other people's AI?
00:31:57.914 --> 00:32:09.288
I think that's also really, really important, but it requires some technical skills, for sure, and also like technical infrastructure, but also, I think, proactively designed to make sure your AI can improve faster.
00:32:09.307 --> 00:32:18.085
You know, makes sense, makes sense so what I'm taking away also, which you kind of did not say it out loud, but consumer product sort of will become transient by design.
00:32:18.085 --> 00:32:20.840
I mean, it has sort of been happening over decades, if you see right.
00:32:22.036 --> 00:32:25.683
I mean, if you have a network effect, maybe that's different, right, because now they're sure they could.
00:32:25.703 --> 00:32:32.290
Yeah, for example, I do think that something like a facebook or an amazon would still maintain their scheme, because it's not just product at the end of the day.
00:32:32.290 --> 00:32:38.634
Right, there is, and that's where, you know, maybe a strong offline aspect of the business could create that differentiation.
00:32:38.634 --> 00:32:43.425
Right, amazon is fundamentally a logistics company with all this network design.
00:32:43.425 --> 00:32:45.939
Of course, tech is a big enabler to that, right, uh.
00:32:45.939 --> 00:33:06.086
But what you also sort of said is that on the enterprise side, you can find more stickiness and, you know, sort of differentiation, because that's where the whole trust thing and, okay, enterprise need to know that, okay, you're not going to kind of, you know, leave them tomorrow and there's going to be some kind of sustainability there in in terms of tg, right, in fact, that that's what our thesis has been.
00:33:09.195 --> 00:33:15.298
We obviously know that several AI consumer companies will come about Very difficult to underwrite.
00:33:15.298 --> 00:33:17.281
Which ones are those going to be?
00:33:17.281 --> 00:33:26.695
Versus having an enterprise getting served within AI companies, you can underwrite it better from a risk perspective.
00:33:26.695 --> 00:33:38.182
Basically, right, because, uh, there is a you know, whole trust angle and if you get in once in an enterprise, unless, like, you're screwing up, you're not likely to be replaced, right, even, let's say they get something 20 cheaper or something like that, right?
00:33:38.182 --> 00:33:41.977
So that has been, uh uh, our sort of thinking so far.
00:33:41.977 --> 00:33:42.837
Let's see how that evolves.
00:33:42.837 --> 00:33:44.141
But look, this has been, I mean-.
00:33:44.161 --> 00:33:52.222
I mean, I do think for consumer, though, there is one nuanced thing which is called taste, and that is actually, I mean, even if you have the perfect AI, there are still a lot of points.
00:33:52.222 --> 00:33:56.960
I think that someone needs to hold the line for like what does the case.
00:33:56.980 --> 00:33:57.241
Look like.
00:33:57.241 --> 00:33:59.737
You know, if you think about Instagram, they're not the first company with filters.
00:33:59.737 --> 00:34:01.359
Taste, right.
00:34:01.359 --> 00:34:03.844
So I think that is probably going to be very, very important.
00:34:03.844 --> 00:34:04.684
Makes sense.
00:34:04.684 --> 00:34:06.347
Yeah, even product manager, I think.
00:34:06.347 --> 00:34:11.043
When it comes to product manager, it's like, okay, we used to write PRD, but now, chachi B, you can write PRD, right.
00:34:11.043 --> 00:34:18.463
So it's like, okay, maybe now it's a competition to see which product manager has the best taste, and it's a little fuzzy, right To define what is a good taste.
00:34:18.463 --> 00:34:19.474
Yeah, yeah.
00:34:19.494 --> 00:34:22.199
It's really really fuzzy fuzziness of the differentiation.
00:34:22.199 --> 00:34:27.307
I think that's that should be the motto, right but I do think there are ways that you can evaluate taste.
00:34:27.307 --> 00:34:29.541
Good or bad it's hard to describe it.
00:34:29.541 --> 00:34:35.119
I think you can evaluate it, so you know me, so I do think that is important, yeah makes sense.
00:34:36.039 --> 00:34:37.440
This has been, this has been fantastic.
00:34:37.440 --> 00:34:38.500
Thanks a lot.
00:34:38.500 --> 00:34:40.277
Really appreciate you taking the time.
00:34:40.277 --> 00:34:41.876
Thank you for having.
00:34:41.876 --> 00:34:46.346
It was great fun and we are hoping for a part two of this with Astra.
00:34:46.565 --> 00:35:18.278
Right, okay, sounds great want to know when new episodes are available, just search for Prime Venture Partners Podcast in Apple Podcast, spotify, castbox or however you get your podcasts, then hit subscribe and if you have enjoyed the show, we would be really grateful if you leave us a review on Apple Podcast.
00:35:18.278 --> 00:35:21.527
To read the full transcript, find the link in the show notes.