#32: What are AI agents capable of doing?
Robert Scoble and Hieu Le discuss what AI agents can do now. Agents write code at roughly 80% correctness with humans reviewing. Multi-agent research teams complete typical tasks in about 15 minutes. One concrete example: Kippy OCRs receipts and invoices, then emails you a spreadsheet.
Questions this video answers
About 80% of the cases are correct.
Around 15 minutes (expected as a long-running task result).
$6.6 billion (mentioned by speakers).
About three times faster in the past three months.
You said that the AIs are creating rooms where it can do business processes in the room to do calling of customers and things like that.
Messaging, chatting with customer. Right now, the real time API is still very expensive, right? Yeah, it came out yesterday. You know, it's very, very expensive and is one way among many. Any communication ways? Yeah, right. Ways to communicate with customers, for example, chatting, messaging. So how about, you know, if a customer go to our website and so we need. We on the website we have a live chat widget. So. But employees are using this platform for internal communication. How about using exact internal tool to communicate with customer so they don't have to go to another subway to communicate with customers.
So if a customer sends you an email or calls your phone line. Right. What is the tool?
It's happening in this one.
On our side, it shows up in this tool.
Exactly. And they stand here, they stay in the system to communicate with the outside world.
Very cool.
Yeah, Everything is one system. Yeah. I think that's the beauty of this concept.
Yeah. So you said that AI is overnight. Right now your engineers are sleeping and you can spin up a whole bunch of tasks for the AI to code.
Yes.
And how good is it at that for your purposes? Right.
It's really good. Is getting to the level of scary good? Yeah, scary good.
And do your human workers have to check it in the morning when they arrive to work?
That's exactly. That's exactly.
Okay, so they see, okay, the AI wrote me a bunch of code for a feature request that I put in and they have to look over it, make sure it works. Make sure the code is correct.
Exactly. That's a. That. Exactly.
How often is it incorrect?
I think 80% of the case is correct.
Is correct. Wow.
Yes.
So no additional work needed. Your AI engineer overnight wrote you some code and it's just fine.
So we have a concept before AI, or even now, even now with a lot of AI. So what is the software development cycle? It all start with human from planning, designing and coding and testing. Right. So someone analyze a request and fire off a feature request. And Human, usually some junior developers start coding and commit and ask a superior to review the code. Then there's a lot of communication back and forth. Hey, you do this well. And this, you need to change. Change direction and so on. Right. So there's a lot of feedback along the development of the code. Yeah, Right. We can mimic that with AI. How about not starting the code with human will with AI? So AI can analyze the request, start coding, committing the system. Human can Come in and say this is good, this is not good. Can you try this? So they we find a way for human to provide instructions, further instructions, further, you know, discussions with AI to find a better way. And then AI can take that further inputs and then start, you know, recoding, start making changes. It's completely similar to the way human work but with the lightning speed.
Yeah.
You know a computer. Yeah.
You can have 10,000 agents all working at the same time. Right.
You can see on the screen here.
Yeah.
These are all AI agents I mentioned to you, right?
Yeah.
10Pm there, which is now, I think two hour and a half ago.
Yeah.
So three hours ago. So you can see all there is happening here.
Your AIs are doing a lot of work.
Yeah, well, all of them AI agents. Right. You mentioned a lot, you, you heard people mention a lot about, you know, agentic AI. Yeah, AI agents in customer service in many areas. We have AI agent everywhere, in particular in the code base. So AI agent is like, you know, human develop.
And how good a human developer depending. Because you said last year it was like a junior developer and this year you said it's like a senior developer. Right.
Things have flipped. Last year junior developer was having a great time because they had an AI coaching them, teaching them, you know, this is how you do. Right. They can, they can ask AI anything so that they can code faster. And instead of relying only relying on human superior to guide them where to go, they can ask AI. That's what they had last year. This year the AI can work directly in the code base. So AI has become a direct competitor.
To a junior developer.
To a junior developer. Right. So a superior now has two choices, relying on Juno developers. Hey, you have to ask AI if you don't know, you have to ask AI code and then ask me to review another option has a right to the supervisor. Ask AI to code.
Yeah.
Right.
So, and what's doing the coding under what? What's your architecture underneath what's your what, what model are you using? All chat, GPT or.
Yes, we obviously have cloud. Yeah. We don't have a privilege, you know, of having access to the talents and also the capital and you know, to, to develop our model.
Yeah.
And we don't have, and we don't need to do that now.
Most people don't. Do you have a. Sam Altman got what, $6 billion yesterday, right?
6.6.
Yeah. How many people have $6.6 billion in their wallet?
You know, so I actually introduced, I'm indebted to this guy Pon Gotter, I'm calling out, you know, Ader.
Yeah.
This is, this is among the best, if not the best AI code agent.
Okay. Is this your agent that you made or is this Paul? Paul.
So this is Paul making this agent and then we use it. I think the way we use it is very unique. That's the secret sauce. So everybody, this is becoming very popular among the developer community. But I think the way we use it allows us to do amazing thing. I haven't seen other people doing like this.
And what's your, what, how are you using it? That's different.
Yeah. So it's starting with the, with the development software development cycle concept. I actually tend to, to bet on AI.
Yeah.
Not only able to do work, but also able to make a lot of smart decision making. So we actually, the secret sauce starts with that. So we, we ask, how about. So right now the, the cycle of software development is a few months. A cycle. Right? Yeah.
An idea for a new feature. It takes a couple months to get it.
A few weeks. A few months.
Right.
How can we spin it? Increase the speed five times, even 10 times. And then I can see that it can actually increase the speed of that cycle a hundred times. Oh, what if we want to spin this? Right?
Yeah.
The other day, you know, I, I think YC Combinator mentioned that, you know, the, the only mode for startup is speed.
Yeah.
That's the only mode you have.
Yeah.
Right. How can you, you increase the.
If you have a customer and you keep giving them new features, they stick with you, right? Yeah, yeah. But they won't if the other person's doing 10 times more features, you know.
Exactly.
Or helping the custom. Doing something very specific for a customer.
Exactly. But you don't want to turn around, you know, invest a lot of resources and turn around after a quarter.
Yeah.
You want to turn around two days in two days. Yeah. Right.
Yeah. Customer asks on Monday, you want to give them a fix by Wednesday or Thursday. Right.
That means that you actually have to spin the cycle by some X. Right. So if you want to do that, you actually need to redesign the, the architect, the development architect. So we start with, with that, with, with that, you know, way of thinking, that vision. And we think that. How about if we start, we start with AI rather than with human. Human is slow.
Yeah.
Human is smart, but human is slow. How about if we start everything with AI so then we develop a concept we call AI Course first.
AI Codes first.
Yes. Not human. Cost first. Right.
I love it. I recommend people using AI first for Everything. Buying a house, ask AI. You're buying a car, ask AI. Your wife is on the floor dying, ask AI. It'll help you save her life.
So. So I started to mention this concept on X on Twitter a few months back, like two months back. So AI goes first rather than human goes first. So we started that. We thought it was great, but actually we can even improve it. How about AI Course first and then AI cost next.
Refactors.
It refactor and then continue the discussion. So AI cost first. It fire up, you know, it commit the code, turn the code. You know, with CI cd, continuous development, continuous integration tool in Git can turn into a build. Right. A release. And then instead of relying on so.
Soon, I'm going to get features on my iPhone or whatever, on apps that were completely AI built.
Definitely right now we can.
A human might have looked at and said, okay, that looks good.
Steve Task talking about, you know, you can code in English. That is happening right now in our team list. Yeah, we see a lot of data.
And your team doesn't even speak English. You're in Vietnam.
Yeah, they're in Vietnam. But our team is quite a bit special.
Yeah.
English is actually official language for communication. In our team, you see a lot of people make a lot of mistakes, you know, in the grammars, in, you know, spellings and so on. It doesn't matter. But we try to be on par with, you know, international teams, so.
And now we're getting really powerful translation tools.
You can see in our code base, everything is actually in English, right? Yeah, yeah, yeah.
Interesting. So this is all AI doing all.
This work for the past, for things happening in the last three hours, everything was AI.
Yeah. And then the human comes in and goes, you wrote this code. You wrote this code. We need to look at it, make sure it's validated. Right?
Exactly. And then they can add, hey, this is good. This is not good. Try this. So human coders over time have become a navigator.
Yeah.
Instead of a typer. Yeah.
A navigator instead of a typer. So they're clicking on results.
Yeah. And tell them, hey, this is good. Keep doing that. And this is not good. Try this.
Yeah.
So, yeah, I like to say to the team that over time, you know, you guys become a navigator instead of a driver, instead of your typer.
And.
We are seeing this.
Hundred times faster.
Hundred times faster is possible.
That's crazy.
For the past three months, we see three times faster already.
Yeah.
Just for the past three months.
Yeah, it's crazy.
It's really crazy. And it's scary.
Yeah. And is it? Does the AI keep up to date on the news like you said? OpenAI just released a real time API. Does it read the news reports like, oh, there's a new thing I can do. I can add that to my app.
That's part of the. Well, that's part of.
That's what I want.
That's part of the reason why we started this conversation, actually.
Yeah. There's 5,500 companies on my X list and I don't even have all the companies. Right.
We can actually do that later on. I can show you. Maybe I can show you right now. Part of it is, but I can show you right now. Let's say.
Right now you.
You probably have a team, but let's assume that you are on your own doing all kind of stuff.
Me and Irina and Grant, three people.
A lot of research. You need to do a lot of background research and so on. Right. Summarizing and so on. How about if you can have a team of AI agents doing a lot of stuff for you?
I'm into that.
One of the team is a team doing research for you.
So. See that? I wish I could hook that up to my ex list because then I could say, hey, go and research these 5,500 companies and go and find out how much revenues they have or how you know or whatever. Right, Whatever you need to know.
I'm showing you part of this right now. Okay, let's say if you have a request. Right. I'm not sure how well we do.
Yeah.
You can shoot an email your request.
To this research that.
RealTimeX co. Yeah. And what do you need? Tell me. I can type.
Can you find me companies on the S&P 500 that have P2E ratio of less than 20 but still are growing fast?
Okay, wait, five companies on S, P on S or.
Or NASDAQ or whatever market you want.
On the market.
On market.
On market.
On stock market or whatever you call it.
Stock market. Okay. Yeah.
And that have a PDE ratio of less than 20 episodes.
P. Yeah.
P slash E ratio or whatever.
Yeah.
Of less than 20.
Okay.
Yeah.
And I can say that we can even like, you know, all. Let's say audience. What is it? Research. What is the audience of this research?
Business managers.
Business manager.
Yeah, yeah. CEO, cio, CTO and executives.
We can say language, right. If we want. You know, it can the report come back. Can be in any language. A language so Spanish, French.
So let's say that it could even write a report with emoticons I did that one time. Chat GPT will even speak back to you if you're, if your kid is using a lot of emoticons, you can copy them and put them in a chat GPT and it'll tell you what the kid's saying in emoticon language.
Right. So imagine we, we fire up this request to the team. English okay.
Yeah.
Without AI, it's pretty good for the team to come back in let's say a few days.
Yeah.
With maybe five pages, a five page report or ten page report. Right. Let's see.
I just said okay, you just sent it.
I just sent it.
All right.
Yeah. And of course you know this is a long running task so usually we can expect a work done with after 15 minutes.
15.
15, yeah, yeah. 1, 5.
Interesting. Then you can refactor and say I need more. Right. Send it back and it'll come back.
Depending on how good it is. Right. I mean we will see. But let's see.
So what's going on right now that we sent this? This kicked up a bunch of agents that's going over to the S and P. GPT researcher. GPT researcher.
A team of researchers doing a bunch of stuff. First thing they, they do is they pick up the request in email. That is actually the email UX I mentioned the U. And part of the reason why we actually having this conversation. Right. So first agent pick up the email, analyzing the request and turn it into a beautiful prompt, a beautiful proposal, beautiful request. Because you know what, I just typed this. So abstract. Yeah, right. Even with typos and so on. So the first agent will improve the request and send the request to another team. Another team would say hey, this, we need to go search. So they implement tools, get back a lot of materials, raw materials, and pass it on to another agent doing data analysis. At the same time another agent is developing the outline. Research outline. Right. Another agent. This is multi agent framework. So another agent analyzing the data and turn it into an Android app. First graph.
This is nice because it lets normal people use agents without knowing anything about agents. Right. We didn't know anything about agents there. We just said, hey, can you find us all the companies on the S and P that have a PDE of 20?
Yeah. Right behind the scene, you know the it. The Asians work very, very in a very similar way. Human working. Yeah.
This is something I said on stage that to think agentic you have to realize you have a new employee, whether they're a junior or an intern or what. It doesn't matter. You have an intern, let's call it.
Sure. Assistance.
Right. And you, you have to understand how to talk to them a little bit, but not much. You just.
Exactly. Right, exactly.
And the advantage of agents is they can do all their work all at the same time. Right. So if you have a.
Sure.
A list of 5,000 names and you need addresses for them, they can do that all at the same time. A human has to go one by one by one.
Exactly.
Right.
And then you can, you cannot easily scale your interns.
Yeah.
But you can pretty easily scale your AI agents. I think that is the scary future for human.
Yeah, yeah.
Right.
Yeah. It's, it's changing my life already. It's like, yeah, what do I do with these things? You know, that's the real hard thing. But you've given me some answers that, you know.
Yeah. You now can have a lot of assistance around you.
And we saw a spreadsheet a couple of weeks ago that uses agents to fill in the spreadsheet and stuff like that. There's a lot of different approaches to this.
Exactly, exactly. I, for, for that kind of spreadsheet AI, I think it is, it's getting a lot of attention.
Yeah.
But I, given we work with Asian already and this is not an outsider talking. Yeah. I am quite skeptical about, you know, the real usefulness of this kind of thing. Right. Because usually, you know, we have a lot of feedback and not always we need speed. Right. We need quality as well. So quality versus speed versus. And also we need to factor in.
The cost, which is why you have several different agents doing several different things to try to make sure the quality is high.
Exactly. Remember, OpenAI has different pricing for long running, low response APIs they call batch API. And the cost is way cheaper. And I think people tend to neglect, tend to ignore that part. We asked, you know, small team.
Yeah.
Mostly, you know, try to sell, try to survive with our own resources in Vietnam. How can we compete with, you know, resource rich startups in here? So we have to optimize everything, in particular the cost.
And, and this is why your agents run overnight.
Yeah.
Because they can run slow and use cheap resources.
Exactly. We don't use the right model for the right thing.
Yeah.
Even now, you know, 01, 01 preview, 01 mini. Most of the tasks we don't need 01.
Yeah.
So why do you need to use, why do you use 01 if you don't need it?
Yeah. Use a cheaper model, smaller model.
Use the right model for the right thing.
Are you using any models locally? We do you do?
Okay.
And are you using Llama or Mistral.
Or a Lot Gemini? Deep code, Deep Seq. So the idea is that you would be able to switch models very easily and even automatically. That's the secret sauce.
Yeah.
So.
So your AI can pick the model it's going to use because it knows something about the model. This might be this kind of thing. Right. Like looking UP S&P 500. One model might do it really well and cheaply and another model does another task better.
We call model routing.
Interesting. Model routing.
Yeah. Because otherwise, you know, if you just rely on one API key for everything, then you pay a lot of money for expensive models without needing it.
Yeah. For business people who aren't in this world yet, they're still doing things the old way. How do they have to change? Let's say they download your tool. How do they have to change culturally or definitely tell us a little bit about how you get in the new AI way of working instead of the old human way of working.
You know, what's the most difficult thing for business to do? That is change.
Yeah.
Establishing a new habit is so difficult. Right. Therefore, you know, 90% of the digital transformation projects fail.
Yeah.
Because it's so hard to ask people to, to change.
Yeah.
Right. So how do you deal with that? And we're struggling, you know, having customers, you know, embracing changes. Right. Even the business owner had that need. But they can not make the employees change and therefore they fail. So one way of doing that is actually how about how do we deal with that? And I think the best way to deal with that is not to ask people to make changes. Even they don't need to use this tool. We can go back to the old way. If you need something, send your request to this email. Behind the scene, our team will do and the team is there. So we tend to invest, we tend to, you know, invest in, in this ux. By doing that, you can just keep doing the way you like if you want to. If you, instead of, you know, using Slack, you want to use email. A lot of, you know, businesses are using email. Fine, you can just do that. But copy this email. Right. Like you know, the researchetonx co. So we receive that request and then, you know, behind the scene, we do it for you. So it's like, you know, service providing economy. But the AI agents are the providers. Interesting, right, Interesting.
What else do we need to know about working with your tool in particular, but agents in general?
In general, I think even, I think we can start even without the tool. I think it's the mindset. AI first mindset. And people talk about. Start to. People start to talk about AI First. What do they mean, force yourself to.
Use it for everything. For everything. You're eating at a restaurant, ask AI what you should eat.
Yeah.
It's not that you're not. You're gonna listen to it, but by forcing yourself to do it for everything, then you get into that mentality of how good is it at certain things?
Right, right.
And it's pretty good at telling you what to eat. I took a picture of a menu, of a logo on a menu in a restaurant in Prague. So not here in Silicon Valley, far away.
Sure.
It told me what to eat. It knew what was on the menu because the menu is on the website. Right. And the AI could read the website and tell me what to eat. And it read all the Yelp reviews and all the other reviews. Right. And all the blogs and whatever else was out there about this restaurant. It knew the restaurant pretty well. You know, the locals all said, yeah, this is exactly what I would eat here.
So it's so natural for you.
Yeah.
But it's so difficult for almost every.
Most people aren't like that.
Most people. And in particular, in an organization where you have a diversity of people with different mindsets.
Yeah.
So how do you. So how do you, you know, help them establish a new mindset? People. My friends ask me because, you know, I'm AI first. I'm AI obsessed. So people, you know, start to come to me asking for advice, you know, and to share with the team. And I felt that one of the way to help people realize this is to ask this question, what is AI to them? This is a very basic question, but I think most people don't think about that. They don't answer it. Right.
Yeah.
What is AI to them? Most people I heard say that AI is a tool. To me, it's more than that, it's an intelligence. To me, AI is.
And that could mean a lot of things.
Yes. And this is key, I think. To me, this is key. This question, that basic, that simple, that seemingly stupid question is actually key.
Yeah.
What is AI to you? To me, AI is the extremely smart consultant, which is very accessible. I don't want to use the word cheap. It indeed very affordable. But it's very different from very smart. Human consultant is that this consultant, this AI consultant, not only can give you advice, but can do stuff for you.
Oh, yeah.
No human consultant will do that for you. Will do things for you. They give them advice, they Take money and they go. And they are very expensive. Right. So if we treat AI like this, we will see the full potential, the full values.
Yeah.
Without AI Assume that, you know, there's a human with the capabilities of AI right now.
And the problem is everybody's going to have a different answer. Right. If a photographer sees AI as a photographer or an editor, a photo editor or something like that.
Exactly. So let's say if you don't have AI as you have right now, but last year you have, you say that, hey, your CEO can come come to you and say that I just met this person is so intelligent. We need $500,000 to pay for him to stay in the team for a week. Would you say yes or no to pay? I think most of the time people will say yes, because otherwise if you don't say that, yes, that consultant will go to the competitor. Yeah. Right. So let's say you pay half a million for this consultant to stay for a week. Yeah. You have to try to make use of his time.
Yeah.
All the time. Right.
You'd probably hire them to set up SAP so you can do inventory or something.
Exactly. Set everything right. Try to make, make use of every second of their time. Yeah. So that you can, you know, retrieve, you know, all the cost that you pay for this expensive consultant. We have to treat a high like that. We take for granted that, you know.
Because it's so cheap.
It's so cheap and we take for granted.
Instead of 500,000, it's a dollar, maybe 20 for all the APIs.
That's what I'd advise, you know, my friends, my business friends, that, you know, if you treat AI like that, you will appreciate the value.
Yeah.
And then you can start, you know, talking to your, your team and start embracing AI into your everyday work. Otherwise, you know, if you say that, hey, so it's affordable, you can just get some answer back. Then you don't see the value. You don't. You cannot force the team to.
So treat it like an expensive consultant.
Exactly.
That's a great tip.
It is, Yeah. A consultant, Right. Very expensive. Assume that you have to pay millions to get this consultant. Otherwise, you know, other people will grab him or her.
Yeah. It's like having Sam Mottman sit, sit with you. You know, he's expensive.
Yeah, definitely.
Right?
Yeah.
The AI knows more about AI than Sam Lahman does. Yeah, I bet. I bet if I ask AI some questions that Sam Altman doesn't know for sure, even though he runs OpenAI, that AI knows a lot of things Sam Altman doesn't know.
Definitely.
He is, he knows everything in the Smithsonian. I don't know. I mean, I've been to Smithsonian, you know, five times in my life. I don't know anything about anything in the Smithsonian. But the AI knew everything. Everything.
That's the whole point. And people tend to miss that because, you know, it's so affordable. Yeah. So easy. So people, people don't see the, you know, the deep values in that.
Yeah.
So treat AI as a consultant. Very similarly, very expensive. Consultant.
Yeah. Or a Lawyer you're paying 1500 an hour to or something.
Exactly. You have to.
By the way, it does your law too, Right. Do you have lot of features in here? You know, check out contract, check out build me an NDA. You know, there's all sorts of legal contracts that, you know, write me a contract for, you know, selling this company a thousand widgets, you know, something like that.
Right. Most of the time AI can do 80% of the work. Right?
Yeah.
Instead of. And for you, instead of, you know, spending hours doing 100%, why don't you just do it 20%?
Yeah.
That's always, I mean, that's the concept that I actually used to navigate our development into products.
Yeah. Yeah. So for writers, they're good editors, definitely. Hey, take, take my book and read it and tell me what I did wrong or tell me how to fix my proofreader. It's way beyond proofreading. Right. It'll correct your writing and it'll say, oh, you stomach use a lot of run on sentences or whatever. Here, fix it. Yes. It's crazy.
Yeah. So, yeah, I think we have to come up with, you know, some, some ways of thinking and putting into simple sentences for, for people to really see, you know, the value of AI. And I think that's the only way, you know, to tell people to embrace AI People start talking about AI first. Everything is AI. But how do they translate that? Business owners, they wish to have that, but it's so difficult to translate that into every.
It's hard. My wife doesn't use A.I. right. And I keep preaching, hey, why don't use A.I. for everything? Ah, she doesn't get that.
Right.
Most people don't. Yeah, most people don't want to add that extra step on to like even asking it at a restaurant, what should they eat? They just want to look at the menu and order something.
Right.
They don't want to ask AI first to see what AI would say before they look at the menu and ask them what to eat.
Yeah. And I think that's actually in a team. That's the danger of. So in a team, I see that there are three groups of people. The one, the pioneer. Yeah. They just see they can envision, you know, what the AI will look like. Yeah. And they're not very good people. The bottom, you know, group, they couldn't see the, they couldn't be visionary.
Yeah.
But they listen, right. They trust, you know, okay. You know, I trust my leadership and they say that very reason. Reasonable. And I trust that. And there's a group, you know, pretty good people and those are the most difficult. I think those are facing the biggest challenge because they're good enough to trust themselves at the same time to be skeptical about, you know, what is happening. Yeah, those are the resistors. Yes.
All the time there's a resistance. Oh, there's too many hallucinations. All right. I didn't get a good enough answer. One time I asked AI My answers are better. Right?
Yeah. So those are, you know, pretty good, but not good enough to be visionary, to see, you know, what is coming. Right.
Yeah.
Yes, it can hallucinate. But what, what is the, the gap between GPT4 and GPT 3.5? This. Right. And the, and the development cycle is weeks.
Yeah.
Not year anymore. If we talk, you know, if we take into account the entire community, not just OpenAI, but meta and Google and so on. Right. So we see a lot of change.
A lot of updates, a lot of smart people.
Exactly.
A lot of money, a lot of effort. Right. And these are smart people. I follow 25,000 people on x.com who are in AI. A lot of them have PhDs. You know, these are not, these are not people you meet in the street. These are really high end people.
And I think that is the actually with the difference between this web three, you know, the web three, the bitcoin and blockchain and this AI right now we have, I think, an extreme concentration of talents.
Yeah.
Into this domain. People no longer talk about blockchain.
Well, there are, I follow about people, but it's not the same.
It's not the same, I would say.
And when you look at the bios of people in, in each community, I mean my go through my AI list, it's like PhD, Carnegie Mellon, Stanford University, University of Illinois, University of Shanghai.
Right.
These are people who are very, very educated and educated at math and science at a level that very few people are. Right. And my friend can write math equations on a whiteboard for how an autonomous car or how a car Would go through an intersection that's not a normal person. I can't do that. Maybe you can do that. I can't do that.
There is a extreme concentration of talents.
Yeah.
And putting, you know, focus on this AI Nowadays, we didn't have that with every other technology.
Yeah.
Even blockchain. Right. Yeah.
And it's the smartest people in every industry. There's people doing drug discovery. I just saw that, you know, agriculture is being done with AI. Mining is being done. Being done. It's just every industry is getting changed by. By this.
And.
And it's the smartest people in that industry who are using the AI. Right. Because who. Yeah. If you're smart, you're looking around. Oh my. You know, that's a good tool or a good thing to use for running our business.
Yeah.
So anyway, so where do we find you on the web?
So I have Twitter. I'm not very active, but, you know, understand why I use Twitter, you know. Yeah, yeah.
And what about your business? Where. Where do we find. Where. Where do we try your tool, your service out?
Yeah. So best thing for me, best thing to try is actually we. So as I mentioned to you, this is the tenth month, you know, being in the US Trying to find US market fit. Right. Yeah. First thing, first product that we develop for the US market is when I came here, I really realized that, you know, people have to deal with tax return and, you know, managing invoices and so on, you know, not only for businesses, but also for individuals.
So find something wrong at Costco, you need to take it back.
Yeah. So we developed some, you know, product we call Kippy. And if you have an invoice instead of, you know, gathering the invoices in a box, and then at the end of the year, you have to scramble around a few days and trying to organize it. Now you can just shoot. There are three ways that you can shoot the invoice to this. Keep it. And instantly within a minute, you get back a Google spreadsheet where you have the E invoice store for you on the cloud. And we do OCR under the hood, you know, and categorize the expenses for you. So I think this is the best way to try the product that we design for the US market.
That's cool. That's cool. I hate expense reports.
I know.
So you just put. Take a picture of the expense, report the expense and it'll make me an expense report.
Yeah, if you have a. If you have a receipt, you can try.
All right.
Yeah. One of the way either you send, you see, keep at. Keep it@us. That is email UX.
Yeah.
Or even, you know, you don't have. The beauty of this is that you don't have to install anything.
Yeah, yeah.
Keep your own way. Give your own way. Right. If you, if you're happy with email, use email. If you're happy with sms, you have sms. If you are on messenger, send to Messenger. And we are opening up other channels, for example, WhatsApp and so on. So I believe that could be an interesting UX for people to use AI.
Very cool. Thank you so much.
Thank you.
More from Robert
45: Sandy Carter Unveils the Future of AI: Myths, Agents, and Creativity | SXSW Insights
