“AI will change ecommerce FOREVER”

Jared Ward: 0:06

what's going on guys? Welcome to part two of OPS Unfiltered. This is with Fabricio uncle Fabricio now uncle Fabricio. He's from Flieber. So Flieber is a modern forecasting decision making tool for for modern e-commerce merchants, or retail merchants, rather. So we're doing a part two because I think both of our companies we're we're really excited about the future of AI and supply chain. This is a conversation that we would honestly just have offline and we're taking it online. First off, reintroduce yourself. Who are you and what are some updates that have happened with Flieber. How are you embedding AI into the tool?

Fabricio Miranda: 0:41

Thanks for having me, jared and Brendon my name is Fabricio Miranda AI into the tool. Thanks for having me, jared and Brendon. My name is Fabricio Miranda. I'm the founder and CEO of Flieber and Flieber is a decision-making tool for inventory planning. If you think about Flieber pre-AI and post-AI the way I see things happening it evolved a lot. I was very I'm not going to say bearish, but I knew AI was a little bit of a bubble and it's kind of playing out now. A lot of people thought it was going to revolutionize from the night to day and, yeah, llms are great, but you're not changing luminous. Nothing has changed. Nothing has changed at Flieber and the companies that tried to introduce. Ai was mostly either cosmetics or productivity tools, simply because it's a great LLM tool. So for things like synthesizing or summarizing things or making you know chats, those things are cool. But for decision-making and data gathering, it's not there yet.

Jared Ward: 1:36

I think a good topic to start off with is, for example, the CEO of Klarna made a post. It was. It went very viral. The the ceo of clarna made a post.

Jared Ward: 1:47

It was it went very viral. It's like we are no longer using any system of record like salesforce in erp's done. It's all ai and everybody was kind of like, really like, is it so? It's like this broader topic of is ai going to dismantle system of records, systems of record rather, or is it going to sit on top of the? Is it somewhere in the middle, like, where does ai actually disrupt our industry?

Brendon Beebe: 2:10

yeah, how do you see that, Brendon? I like to think back to like a year ago, or maybe it was a year, when chad gpt first came out and there was this rush to build co-pilots and it was like this cool concept and I remember we had somebody that was working at luminous and their idea of how to introduce AI was like what if you could tell your phone, I'm transferring it from bin A to bin B? I was just like nobody wants to talk to their computer.

Brendon Beebe: 2:37

Nobody wants to replace a user experience with like text prompts and have to deal with like the potential errors of that. Like people still want a normal interface and the easy question is like oh, let's just add a chat bot, but no people got rid of command prompts 20 or 30 years ago.

Jared Ward: 2:54

Nobody wants like people can only view ai through what they've seen and chat gpt, which is just prompt based ai so it's like oh let's add. Let's just like in the reporting we'll just add a prompt and then it like returns. It's like oh, let's add. Let's just like in the reporting we'll just add a prompt and then it'll like return. So it's like yeah, it's just an imaginative, I just demoed a product.

Brendon Beebe: 3:09

It was cool, it was really awesome. I could ask it questions about my data, but it was still. It was hard, it was slow because we'd have to go query the data. It just wasn't straightforward.

Fabricio Miranda: 3:26

Straightforward and all I really wanted was just a dashboard and they made it this complex that I had to access my dashboard by asking ai the right question and and I think, more than that, in the case of Flieber, we also tried that and we have designs and we showed investors. Investors loved it because oh investors love it investors love it you tell talk about ai everybody's.

Fabricio Miranda: 3:40

Oh yeah, now you're in the right track, how you can raise it million. And it was kind of the same thing. It was basically a chat on the side where you would say, oh, how many units do I have of product A? But isn't it much simpler to just open a dashboard filtered by that product, because you can then see that product or that product with other products. Or filtered by A products or B products?

Brendon Beebe: 4:10

And you can do anything.

Fabricio Miranda: 4:10

And in ai can you imagine in a chat, so I'd like to actually collect.

Brendon Beebe: 4:12

Who thought it'd be a good idea?

Fabricio Miranda: 4:12

yeah, instead of clicking the button to view your products out of stock, you have to tell the ai what how many products are out of stock, and and it's gonna give you a list, yes, and with nothing else on the list, right?

Jared Ward: 4:21

so I feel like this is my opinion supply chain tools the disruption, AI. Disrupting supply chain tools is it's purely it's very specific decisions, very, very specific decisions, for example, like changing price based on certain contexts in your data across, like multiple channels For inventory purposes. I think obviously there's so many use cases around cleaning and mapping data and integrating to other tools. This is where I think.

Brendon Beebe: 4:52

AI, as an artificial intelligence, doesn't know the right answer, and that's where everybody jumps to forecasting. Or it's going to be able to figure out how much I should buy or what I should set the price to. Actually, there's machine learning algorithms that existed for a decade yeah that, do that.

Brendon Beebe: 5:08

Use that large language models are not going to give you the right answer, nor are they going to be able to tell you the right price to set it at. It's going to be able to do really boring things. That actually takes up the majority of your time. Yeah, cleaning your data like for a reason?

Jared Ward: 5:21

what? Like how many customers have you gotten where their data was dog shit and you couldn't do anything and like how can ai impact?

Fabricio Miranda: 5:28

yeah, I agree 100 with Brendon. But, that said, I I have kind of a hot take here because I lately I just made a post today about my view of uh revenue per year per employee. Yeah, um, and I had an objective of getting 250 000 per year per employee and now I'm thinking it's going to be more like $2.5 million tenfold per year per employee and in a few years I think it's going to be $25 million per year per employee. And that shows how bullish I am with AI. After we started doing a bunch of tests inside of Flieber. I don't think the technology is not even close to that, is not even close to that.

Fabricio Miranda: 6:01

But I think what I what changed in the in the past couple weeks was my vision of what, how, this, that these companies are going to interact with each other, right, so, um, talking to my co-founder, shahir, a few days ago, he is in the silicon valley, uh, and being, you know, totally, I think, drowned in this ai world.

Fabricio Miranda: 6:24

Everything is about ai at silicon valley at this time and we're talking about how, the company, what is the role of these companies in in the future, and we, I think the vision that all these companies are going to be assistants. Uh, I think it's a little bit co-pilot, but it's less, and co-pilot is more assistant. So think about it, think, think that today, if, if you want to have a like, you have a company and you need like content, for example, the only thing you can do is to hire someone that produces content or hire an agency that produces content, but at the end of the day, you're adding one more team member, directly or indirectly, because you have that need and you have a bunch of other needs in the company, right? So in the end of the day, you're assembling a team right around all the things. You have that need and you have a bunch of other needs in the company, right? So in the end of the day, you're assembling a team right Around all the things you have to do. That's the paradigm of the old world.

Brendon Beebe: 7:10

I mean, you're just essentially pitching the idea of agents being able to do actions for you Exactly.

Fabricio Miranda: 7:15

So what I'm saying it's not only that, it's connected agents. So imagine that you now have the AI content creator, which is your assistant, uh, and then you have another I don't know whatever you need to do like accounting and you're kind of onboarding this accounting agent. But this accounting agent will be connected to the, to the other agents, and all of the context needed for this accounting agent to perform is going to already be there. It's going to be immediately trained. You don't have to train a person and get them up to speed and deal with the days that they have to take care of their son and then leave early and stuff. So that kind of clicked in my head that AI assistants are going to be things that you're going to turn on and off and each time you turn on, it comes into the team and it's kind of aggregated into the team. Any time you turn off, it's going out away again. So that's the hot take I have. So I would love to talk to you guys about that.

Brendon Beebe: 8:16

You sound like who's the guy on All In podcast. I forget his name J-Cal you sound like J-Cal. J-cal J-Cal. You sound like J-Cal J-Cal gets overly optimistic with things.

Brendon Beebe: 8:25

And I think what he does and I think most people do, is you see this progress in technology, you see this huge jump and everybody's natural response is to say, like, well, if we continue at this rate, we're just going to be out. Yeah. But in reality what happens is that first 80% is a huge jump and that final 10% in gains, which is the most important 10%, to actually be productive and helpful maybe takes a decade, maybe it takes 20 years, maybe it takes 30 years, and I think we're at least somebody who uses AI constantly for programming. It has so far to go because the mistakes it makes is so stupid, like if you've ever worked with it on a constant basis.

Fabricio Miranda: 9:06

My team works a lot with AI for programming. Does your team use AI for coding?

Brendon Beebe: 9:10

Yeah, constantly, but it does a lot of great stuff.

Fabricio Miranda: 9:14

Yeah, but how much it saves in time.

Brendon Beebe: 9:16

Oh, it does. But the idea of autonomous agents is so far out.

Fabricio Miranda: 9:19

Oh, no, no, yeah, I'm not. I think you're going too far, also in my view of the short term. I'm not talking about autonomous agents. I'm talking about assistants. So decision support is not decision making. It's not agents that are going to make the decisions for you. It's a content creator. It's not a content creator that is going to just get whatever you need and produce whatever you want.

Fabricio Miranda: 9:41

That's not what I'm talking about. I'm talking about the potential of the connection of all these assistants. It's the same thing you have today with chat gpt, but instead of being a chat gpt content creator, it's an agent that is an assistant I'm sorry that is integrated with the other assistants and working in group, right. So, my, my, I think the hot take here is that there will be connectivity between these assistants that you onboarding to your team, right, and I even think there's going to be a company doing that connection with those assistants. So I think you're you also thought that I was thinking about, uh, uh, one of those worlds where no people do anything anymore, and it's not what I'm talking about.

Jared Ward: 10:24

I'm talking assistants, not agents at least in the next 10 years. The companies that are going to win or where ai where you're going to see ai embedded is it's in vertical SAAS companies, so like think like the healthcare space or think like I don't know, bridal shop erps, and then the ai agents that you're talking about are like these ai plugs on systems.

Brendon Beebe: 10:47

Let's not use the ai agents, but they are agents like the. It's the general.

Jared Ward: 10:52

The problem is like there's going to be there's. There's so many layers of context to make decisions and do things that it has to be vertical. Specifically, it has to be for like the restaurant or for the bridal. I agree a hundred and like a very specific part of it.

Fabricio Miranda: 11:09

Why isn't that the way to do it? Why isn't that going to happen? Because what I think is exactly that is that all those assistants are going to be extremely specific for a use case and you can turn them on and off, and if you change the use case or if you have different use cases, why would you make them very?

Fabricio Miranda: 11:25

specific Because of what he's saying, because each type of use case has specific context. So there are going to be cases where you can embed all contexts into the same model, but there are cases that you have to be very specific. Just think about image producing, for example. If you go and produce image based on general models, you're going to see certain results. If you get specific models, for example, you want to produce an image uh, like your, I don't know your kid you know disney character. If you have a ai model trained for producing images of people in a disney character is going to do a lot better job than a general ai that you do you'll give the same prompt well, I guess, when you, when you say train, are you talking about a different prompt or are you talking about different LLM?

Fabricio Miranda: 12:18

No, I'm talking about the training of the AI itself. So what do you feed the AI? And you work on the prompts of the AI for the app to produce, because the training is through prompts, right? So you're basically feeding information, seeing the results, perfecting your prompt feeding information, seeing the results, perfecting your prompt feeding information, seeing the results, perfecting your prompt up to the moment that you have a prompt. That is is going to encompass that, that, whatever you're deciding. So what I'm talking about is that prompts go different ways. They can start the same, but they go different ways based on the different use cases. That's, that's how I see it.

Jared Ward: 12:52

So I'm going to name a couple things like in, because when I think of supply chain, specifically for retail, I'm going to name a couple of like checkpoints and you guys tell me, you guys tell me which one you think is is most prime for disruption for ai. So like, for example, booking freight forwarders totally 100 disruptive?

Fabricio Miranda: 13:09

I don't think so I, I agree.

Jared Ward: 13:11

I, I specifically say it's like that's awesome replenishing from your 3PL to FBA.

Fabricio Miranda: 13:16

It's less easy than the first one, but I think it's disruptive also.

Jared Ward: 13:21

Price adjustments for all of your channels.

Fabricio Miranda: 13:25

So if I'm, that is a harder one, I think, because there's a lot of context outside of the data. For price adjustments If you're replenishing inventory, you can parameterize that. For price adjustments, if you're replenishing inventory, you can parameterize that. Yes, you have moments where you're outside of the parameters and those are going to be the exceptions, but the norm. You can kind of automate the pricing. There's so much around it. I don't think it's going to be disrupted by AI. Do you guys think, well, I'm only going to give my opinions, or are you guys going to give yours? Come on, I don't know. We're recording this, but nobody's gonna make you accountable. Brand random it's just a chat between friends in a moment that there's no decision about anything, no clear future.

Brendon Beebe: 14:07

So and just say what goes on in my brain is I come up with an answer, that I come up with a retort, and then I'm like, yeah, that's a good point, it's too broad. I think it comes down to the problem of, like booking freight forwarders is too broad. No, the term ai is too broad oh, okay, okay because we're essentially us. We're creating this like solution, that is like can do everything, and then we're saying what can it do or can't do?

Jared Ward: 14:41

so what? What I think is interesting is I the discussion that Brendon and I have a lot offline is like what is our vertical? Because again, it can't just like we serve a bunch of industries, but like what is our vertical if I were to say that luminous is building a vertical system of record.

Jared Ward: 14:59

Like what is that? And I would say, like it's. It's for modern omni-channel merchants. And where, where are we going to embed ai? Well, it's going to be like the specific use cases that we say see for omni-channel merchants, that you could replace with an AI. Some very specific things I would say is onboarding, developing your product catalog list and mapping your data from one system to another or from Luminous to another channel. That's almost where I'm most excited.

Brendon Beebe: 15:34

excited like these super boring things that happen in the background I think the greatest impact on ai are kind of way more boring to talk about. Um, I agree, I'm thinking back to four up, which was a vertical SAASs company. I'm thinking like what could we have done differently? And the things that become way more efficient have nothing to do with the software. Really. It comes down to training somebody on how a golf course works becomes way easier, the immediate.

Brendon Beebe: 16:05

I can critique and grade every interaction. When I hire a brand new customer support rep out of BYU and they're helping somebody, every interaction they have now BYU and they're helping somebody, every interaction they have. Now. Ai can give them pretty decent contextual clues on, like what they're doing good, what they're doing bad, what they could do differently, and giving them tools to learn, like the questions that were unaskable before to google. You can ask and you can get instant clues into what that means. It's like when, uh, somebody in Alabama has problems in the golf course and they ask it and you're like what in the world does that mean? Like they could do now, ask ai and get contextual clues and learn way quicker. And I think what that does is it it lets you have a lot less people.

Brendon Beebe: 16:48

You don't need as quite as many employees, not even because ai is doing everything, yeah it's just it teaches, it can help you, as you said, like you can be an assistant to like, give you way more information way quicker. Yeah to like level up.

Fabricio Miranda: 17:03

So you're saying the same thing. I think in the beginning, when I said the assistant, I call it the week, you could. Yeah, I think you. You thought I was talking about this disruptive, you know, dystopian, yeah and it's not at all what I'm saying.

Fabricio Miranda: 17:15

What I'm saying is exactly what you're saying. How do you, how are you able to create a company where you, every single person in the company is producing a lot more because it has a lot more information? And then you know what you needed three people to do, now need one, because that one person plus ai is a superpower. Right, we're talking about superpower in Flieber, and I see Flieber as that. I see Flieber as the company giving superpowers to the people operating their inventory, making their inventory decisions. You know, hopefully you know, one person will be able to do the job of everybody in the company, because the AI is helping that person, giving the notifications at the right times and giving the context that they need, and all of those things that are easily the boring things that you were talking about right.

Brendon Beebe: 18:02

Here's a bit of a pivot Talking about. Like. There was a company I talked to a few months ago. While everybody was doing like these co-pilots, that really didn't go anywhere. They started implementing AI and doing lots of small things, and so their platform is all about importing data into other systems and they did this really cool thing where they trained a local LLM to make quick modifications. So it was a super highly specialized language wallet that could run locally on their servers.

Brendon Beebe: 18:33

They didn't have to go out to open AI and it would do things like you would import your Excel spreadsheet and maybe you have a column of emails and before validating emails sucks, and maybe you have some, depending on where you export it from. It's like it's not complete, or maybe it has like at instead of the at symbol, like 100 miscellaneous problems, and so what it let you do is it'll just let you type fix these emails, and so, instead of even coming up with some sort of JSON rule to modify one to the other, it would literally run that super localized, super specialized, large language model on every single row and fix everything for you. And it would do that for categories like fix these tags into the right format, like super specialized things, where it was almost like an upgraded excel spreadsheet it was so sick, yep, um, and that reminded me of you, what you were talking about, the whole mapping process yeah, yeah, exactly.

Fabricio Miranda: 19:27

That's one of the fields that excite me a lot, and that's what jared said. Uh, how do you get all this data that comes in onboarding processes? Both our systems have a lot of data on onboarding. How do you get that data and you make sense of it in an easier way? Right, the way we do it today is ridiculous. We basically, you know, a team of people, sit and study that integration and we have our APIs, and then how do you connect each part of that api to our api and route that, that, that uh data to the right place and and maybe transform that data in the middle of the way? And it's ridiculous. That's not going to be done that way. That's, that's one of the things that I deeply think that ai oh is going to solve.

Brendon Beebe: 20:10

It's so sick, like that's one of the things I think we've been doing. So we have like us for our integrations. We have a standard interface that we can hire a contractor and they just have to build into. But you give that interface to a large language model. You give the API spec of whoever we're building into it outputs the code. For us. It's not reusable. We can't offer that to our customers and not be all.

Brendon Beebe: 20:33

But it turns the integration process that used to be, you know, maybe 20 to 40 hours, into two to three hours. It's fully tested and it's better.

Fabricio Miranda: 20:41

Now imagine if you don't need that other end, because the AI goes one step further and is able to understand the context of the data that is coming in. So basically, you're just plugging into an API, letting things come without any kind of context and, based on the context in your platform of other similar data that is loaded, you're going to understand what is each piece of that information right and based on also the….

Brendon Beebe: 21:06

See, this is where it gets, though, into like this… I think we tend to do this with, like, really big features. We dream and dream and dream and all of a sudden it becomes this God feature that can do everything.

Fabricio Miranda: 21:20

And it's like we never actually get there. Yeah, yeah, no, I don't think. I think that's a company in itself. I don't think it's something that you have to do or I have to do, you know, as Flieber, or I think something interesting that we're already seeing.

Jared Ward: 21:27

Here's a use case in the warehouse rate shopping. So rate shopping is is already available. With tools like you can optimize it for, like the lowest cost. I think what are you gonna see. Rate shopping is already available. With tools you can optimize it for the lowest cost. I think what you're going to see.

Brendon Beebe: 21:36

What is rate shopping?

Jared Ward: 21:37

Rate shopping is where, so, for example, if I'm shipping out my product, in this box Shipping labels. Yeah, if I'm printing out a shipping label in this box. So the old way of doing it was I would have to, I'd have my weights and dims and everything in, I'd have the packaging details and on Ship station would display fedex and ups and this and that, so rate shopping is essentially all it is is like you would. You would manually input automations in your settings. Yeah, and then it would.

Fabricio Miranda: 22:03

it's not ai, but it's just optimized for, like it's just a rule for like okay, if it's in this zone, then yeah, it's this type.

Jared Ward: 22:10

If this then that optimized for like. So I think that's where ai is probably going to hit the market first is like okay, okay, well, like real rate shopping optimized to save you cost.

Fabricio Miranda: 22:21

Yeah, the problem to save you cost is that there's a lot of context around. What save you cost means right? Because maybe you prefer the most expensive one on this shipment because it arrives earlier than the cheaper one, and maybe in the next shipment it's the other way around. You have enough inventory. So that's where it gets a little hard to understand.

Jared Ward: 22:40

well, and that's why I go right and that that's yeah, as soon as you add in another layer of context, which is why you have to look at specific use cases, but I think that's that's one people in the warehouse generally would probably say shop for the cheapest rate, like that. That's probably what they'll do. And yeah, and you're, you're bringing a point like yes, when you have to look at an additional layer of context and like, okay, it kind it kind of falls apart there.

Fabricio Miranda: 23:03

Another one that was like I don't even think it falls apart. It just needs that extra context, right. So it needs to be trained to receive that extra piece of context. It's like forecasting, for example. We do forecasting with best sales, with price variation, product catalog. But if you want to add weather, or if you want to add you know, taylor swift is doing a concert in your town and you're going to sell more in that during that time, those things are almost like add-ons to the same process that you need to create ways to train the ai to consider them right, but, as at the moment, that you do and create and give that context, it's simple solution here's a question.

Brendon Beebe: 23:40

So in that scenario of shipping, it's almost deterministic. If you had a perfect rule set created, the solution to what label you should pick is deterministic. You plug XYZ in and then you should get A right Mm-hmm. So is AI suited for making those decisions where AI is not actually?

Fabricio Miranda: 24:00

intelligent, not a decision-making, yeah.

Brendon Beebe: 24:02

Or is it ideal for creating the rule sets? So, is it so because, in my mind, you use ai. Ai is great at creating really complicated configurations. What used to take an implementation manager maybe a few weeks to set up, you could get an ai to set it up instantly.

Jared Ward: 24:20

Yeah that was. That was actually like if I were to close out the point. So instead of me having to go like, for example, go into ship station and write all those automations manually, when the whole fucking point is like, just get me the lowest cost shipping label and and obviously there's going to be some other rules around that but if I could just prompt an ai like, hey, set set up, set up automations, so that I it automatically picks those and then it writes the 27 automations that I have to do, I feel like there's so many use cases like that where it could just write you automations, yeah, that's almost resolved.

Fabricio Miranda: 24:55

Right it's doing that already.

Fabricio Miranda: 24:56

Exactly. Yeah, what I think is more interesting and that's what I was discussing with you guys in the beginning is like where is this evolving? Because that's completely unknown. Nobody knows. Even the big tech companies are going different routes. And is it going to disrupt Luminos? I don't think so. I think systems of record will always be needed, but maybe, as I said to you before, I don't think you need to have an inventory management module. I think the inventory management can have assistants doing that part and you are going to then feed from those assistants and be the system of record Flieber. You know inventory decisions. I don't think decisions are going to be made by AI Not anytime soon. As you know, ai doesn't think decisions are going to be made by ai not not anytime soon. As you know we know, ai doesn't make decisions. Well, ai is. Is what is the next best uh ladder based on, you know, factors and etc. We can go deeper here, but uh, it's a next best ladder system, right, it's not a decision making system.

Jared Ward: 25:58

My pie in the sky like ai ai idea for free commerce is first off. I my hot take is that fulfillment is slowly becoming a commodity. I mean it already kind of is.

Fabricio Miranda: 26:10

You're seeing aggregation in the 3pl space.

Jared Ward: 26:12

You're seeing what they're doing with awd. I think the future of is going to be like robotics, 3pls, and simply because they can do it for the cheapest and the most efficient. So what? What e-commerce companies turn into is like it's this one super powered human that is using ai to make decisions like what would normally take them literally 30 minutes to. To look at all of the contextual, data to make a decision.

Fabricio Miranda: 26:40

That is the point now you can have.

Jared Ward: 26:41

This is this. My prediction is like in 10 years, everybody will be using like a robotics 3pl and then you can have one person running like a 200 million dollar brand. Yeah, yeah because yeah because ai can just power them to to make decisions like so much faster. Yeah I.

Fabricio Miranda: 26:57

That's my hot take. I don't know if that's no, no, I I agree 100, I think. I think that's it. That's that's why we talk now. We were talking about, you know, flieber's website earlier today. No, using the superpower is is a great term, because that's what I think AI is going to do.

Jared Ward: 27:12

It's it's going to superpower us, right, you're going to, we're going to be able to make the decisions better because we have more data, more easily accessible data in better quality, and so there's a bunch of different things that ai is gonna is gonna play a role so like, for example, like shopping for freight forwarders, whenever I purchase something, oh my god, like the, the amount of data that I have to gather, like, okay, what is my total weights and dims for the shipments, and like, uh, and sometimes a good erp can provide some of that contextual data, then I have to grab this, put it in an Excel sheet, put it in a formula. Now I have to send it to this freight forwarder.

Fabricio Miranda: 27:49

And then I have to email this one, and then I have to email this one. That is where AI will change. Shop for me.

Jared Ward: 27:56

If that data just existed, then technically an AI could be trained to just pull all of that for you and probably just send out an email and ask for a quote.

Fabricio Miranda: 28:05

The other day I was in Pittsburgh visiting my son at university and we wanted to go out for dinner in a place that didn't have online reservation. And then I had read somewhere about the Google Assistant for reservations. Have you guys seen that? Yeah, so basically Google calls the restaurant and makes a reservation for you. Now imagine that the restaurant also has their assistant, so they're calling. So that's what I think you're saying here.

Jared Ward: 28:32

I see what you're saying.

Fabricio Miranda: 28:33

Each of these freight forwarders might have their own assistants that are the sales assistants, and you have the purchase assistant and your purchase assistant is sending a message to all the sales assistants saying hey, I have this shipment here, give me the quotes. And then, suddenly, the snap of a finger, you have all the quotes in front of you, not only from the freight forwarders that you know and you have a relationship with, but all the freight forwarders in the world.

Jared Ward: 29:01

Even more boring. Another use case is like our inbound ship and reconciliation. So a lot of times the the brand system of record and then the 3pl system of record. You have a receipt that happens and you need to reconcile that like I don't know so many boring things like that where I I feel like ai can can help flag things quickly or fix things.

Brendon Beebe: 29:25

Listening to people online talk about AI frustrates me more than anything, because I feel like it's so easy to pivot into this like it's a god feature that can do anything and now it solves a lot of problems. What if you reframe this conversation? I think the industry that's been impacted most by AI has probably been writers, clip bar and programmers yeah um, I'm most familiar with programmers.

Brendon Beebe: 29:51

Let's go there. If I think through it's already impacted it more than anything else. If we just assume that type of impact will carry through other industries, it'll just be slower. What has the impact been?

Fabricio Miranda: 30:02

no, but I think the impact was just because it's those industries are tailored to l right we were talking about pie in the sky.

Jared Ward: 30:08

Now you're getting to like okay, what is what is like actually? Logically, the next impact I think you already nailed it like there's writers, copywriting, content, coaching, like that. I mean, that's where I'm. I'm literally seeing that in the tool right now that we use ask elephant. I think the next immediate step is just like coaching, coaching our, our internal employees on how to be better employees, and probably use it.

Fabricio Miranda: 30:36

We're tutoring kids. I was talking to my kid the other day. I was going to a concert with him, uh, in new york, and we had like a one hour commute and we're talking about that. He's going to have an assistant tutor. It's going to be part of his team. He's going to have an assistant scheduler. It's going to be part of his team.

Fabricio Miranda: 30:53

Today's schedule for him is a mess because he has, you know, soccer, he has the band, he has school and he has a bunch of different activities. And he's 16 years old, school, and he has a bunch of different activities and he's 16 years old. I don't want to be, you know, telling him what to do he needs to figure out. So he's gonna have a scheduler and whenever he needs to, whenever he has any issue, the scheduler is gonna come up to him and say, hey, you know, someone just booked or you just double booked. Um, for this time, what, what should I do? Oh, cancel this one. And then his assistant is going to connect to the assistant off the other side and he's going to come up with a decision on what.

Fabricio Miranda: 31:28

What is the next best time for that, for that appointment? Well, in my opinion, that's not even future. That that should. That should be done like in a few months, right? If? If it's not being done already, I don't know. But you don't see that as short term? Brendon, tutoring and scheduling, all those things that are one step above from simple lm tutoring?

Brendon Beebe: 31:53

yes, because it's it's replacing, it's not even replacing, it's even it's aiding, and it self-interaction already happens like this. I have so many examples like this skis kid scheduler one that's more like it's deterministic, it's like a calendar, right. But I also don't think you need like this concept of different assistants. I also disagree with because I think that's just a symptom of a llm with a small context window and that, like the idea of having to like have different prompts for an llm is just a matter of fact of their context window is so short. You need a different context, and so that's something I don't think you'll have.

Fabricio Miranda: 32:33

You think it's going to be one big thing that you're going to use.

Brendon Beebe: 32:36

Yeah, I think ChatGPT will be able to do a lot of these small things. You won't have to have custom assistants. It's how does those assistants interface with the real world? That, I think, gets more interesting and where there's a lot of different play, because ChatGPT can't interface with everything.

Jared Ward: 32:53

Well, if you look at where AI is disrupting systems of record or sorry, rather, augmenting systems of record, it's in like, for example, I do a sales demo and then it coaches me beautifully, like I can select all of the things that all of my meetings and it coaches me beautifully, it tells me trends, all of the things that all of my meetings and it coaches me beautifully, it tells me trends. I think something, something with like systems of record, like luminous and inventory systems. Honestly, it could be as simple as coaching them on their decisions. Like they did a transfer order and like it looks like this thing went out of stock and how do you? Maybe, maybe it could coach you on being a better operator, like yeah, we're and I your thing about tutors.

Brendon Beebe: 33:29

I think that is the ability to learn, I think has been accelerated like 100x. That's got to be the most impactful, the ability to like. I remember I was talking with my brother yesterday actually about this. He's in college right now and we were talking about how he uses it. He doesn't use it to, he says it gets every question wrong in his uh, he's like in a 3000 level engineering class. It gets every question wrong, but it knows exactly how to teach and so you can bring up any problem and it can tell you how to get to the right answer. And all of a sudden you're not stuck.

Brendon Beebe: 34:03

I remember taking physics in college and I would just get stuck on a problem. I wouldn't even know what question to ask Google Right. I'd have to stuck on a problem. I wouldn't even know what question to ask Google right. I'd have to go to a lab, I'd have to sit down with a TA. You know, teach 10 people at a time how to do a specific problem. I'd get stuck on the next one. That now ChatGPT exists at the point where it can actually explain exactly how to go step by step. You can even give it to the answer, and then it actually does even better, and I think that's how does that apply to everything we're doing in our life? How do we become even better learners? This is what happened last year with edi, a concept I was completely foreign to, had no idea what edi was, didn't even know what questions to ask google, and a month later we had embedded edi because it was able to get me through all these things where I didn't even have the concept of knowing what to ask Google.

Jared Ward: 34:49

Yeah, I feel like that. That's actually probably the next.

Fabricio Miranda: 34:53

That's already there, it's already. It's not even max, it's. It's this Well the fuck but like who's embedded?

Jared Ward: 35:00

who's embedded like real, like supply chain or operations decision-making coaching into their tool? Like, for example, I'm trying to replenish fba, but like I don't know how to even I don't even know what to reference, like what context to reference to like get to a good decision. I feel like that's where ai could like it could coach you, it could like, yeah, well, first, the question I have is only first.

Fabricio Miranda: 35:22

You know, I would love to talk more about this idea of a huge model that solves everything. Um, because I I don't know if that's the way I see the future.

Fabricio Miranda: 35:34

I, I tend to see the future as yeah you know, as if you went back to the 1900s and and and talked about the things that we're doing today. Nobody would believe, because those things didn't even exist at the time. Right, right, uh, and I think it's a little bit like that. I think we are going to evolve to somewhere where we don't have this big God called open AI doing everything for everybody that everybody needs. I do think that we're going to find things that are going to now slice that into very specific new niche things, because that's how humans are. Someone is going to be very good at. Maybe we're going to have over-specialization, maybe it's going to be coding for people who work with flip-flops. I'm just being stupid here, but it's just. I'm talking about hyper-specialization.

Brendon Beebe: 36:25

Because that's how many companies started, like six months ago or a year ago, building off the open AI API thing, like, oh, this would be so cool, but not realizing opening. I would be able to do that in six months. And I think that's the question you have to ask, cause, like because we're still in the LLM world, right, but what can? What is the next level? That isn't something like every time. Somebody's like prompt engineering was so big a year ago and it was so dumb, like just being able to adjust a prompt. There's no talent there and it's more of a bug of the llm and not a feature.

Jared Ward: 37:00

Yeah, because that's how you interact with it. It's a exactly.

Fabricio Miranda: 37:03

Yeah, yeah, and I think there's a lot of the data side that we're not talking about, right? So access to specific types of data that are hyper specialized are specific for that type of product a problem? I mean because, if you are, it's just the same discussion about generalists versus technical people that have a specialized knowledge. I think chat gpt is going to be an amazing generalist model, but when it comes to very specific use cases, it will probably get wrong. The answer is because it has so much context about other cases that are not that one that maybe it's probably not going to be to be able to identify that you were talking about that specific use case.

Brendon Beebe: 37:47

I would have thought that maybe a year ago, and like the best peach, the best uh programmer would probably be a llm specific to coding, and so you have like co-pilot, but no chat. Gpt continues, or cloud 3.5 I think it's a little bit still, I think we're.

Fabricio Miranda: 38:04

You're using examples that are very llm driven, right so. So coding is not different than writing content, right? So it's basically what is the next best letter? Okay, so when you get out of that, what is the next best letter world? When you get more specific things, that you need more context, you need decision making capabilities at a certain level that ChatGPT today doesn't have, I think you get away a little bit away of this, this very bread and butter. Next best letter world? I think data is going to be play a huge role. Contextualized data is going to play a huge role, and and today you don't have that you kind of don't need.

Brendon Beebe: 38:44

Was that just a matter of having your own database of data that chat GPT plugs into? Maybe, I didn't simply Maybe that's.

Fabricio Miranda: 38:52

You know. Maybe two years from now we should do this podcast again about the same. But let's put on the schedule and see, because maybe you don't know, I don't know, we're just having you know nothing to say here, but this is where I go to going back to where is the world going?

Brendon Beebe: 39:07

So, going back to where is the world going? If I just look at the impact on programming, ai has destroyed the junior level tasks. Yeah, I have to question myself. Is it worth hiring a junior developer who's going to do what ChatGPT can do for me already? Yep. And if programming is the first world to be disrupted, writing the same thing, why would you hire a junior level writer? Yep. Going into e-commerce or forecasting? What junior level roles are there that are just busy?

Jared Ward: 39:37

work, yeah, See, like this is where my mind goes when I try to go get out of prompt-based AI like something that you have to type do this, and then AI does something. You have to type do this, and then AI does something Like an immediate thought is like okay, something in ops that happens all the time is data comes in, it's mismapped and you have to clean it. There's a junior level task that happens all the time, Immediately, like you could have AI map data, yeah, and that would be incredibly powerful.

Fabricio Miranda: 40:09

Oh, even customer success right. Why do you need junior people in customer success Right? The first level customer success can be completely disrupted by AI Right.

Brendon Beebe: 40:17

It's the most difficult task for being escalated to a next tier. What's the entry level task Like? What are you doing? Yeah?

Fabricio Miranda: 40:25

I think that's the point.

Jared Ward: 40:27

I guess we'll see what happens, yeah, and I think there's no right or wrong.

Fabricio Miranda: 40:30

I think and I'm completely fine with your world or your world, I think, here my idea is how do I get better at projecting what's going to happen with Flieber through discussion with intelligent people.

Jared Ward: 40:41

That's my I'm bullish on systems of record staying.

Fabricio Miranda: 40:47

There's a place for systems of record in in erp.

Jared Ward: 40:50

Yeah and in fact I think, like, for example, if you, if you think about net suite or salesforce, I actually think their best position to come up with very specific ai products uh that just sorry.

Brendon Beebe: 41:04

One more point that made me remind me, like this idea that system of records are going away, a vertical SAAS is going to be killed by AI for some reason. You would have to assume. It's easy to say it when you're a billion dollar company with a budget and you're going to hire engineering team, but can you honestly expect, like let's go to four up, that every million dollar golf course is going to build their own ERP? It's like absolutely not right now. Somebody going to build their own ERP? It's like absolutely not right. No, somebody has to build the tool set that's going to do all the work for them, or bring AI in a way that's applicable, and I think that's I agree with you where it has to be specialized. You can't have a generic chat GPT that does everything I was going to say. That's the.

Jared Ward: 41:43

But this is where the industry this is where the industry specifics of like, because that's where the industry-specific stuff, because the new system of record to emerge in the golf industry, for example, could be one that the default. It does all of the junior-level tasks. It counts your inventory I don't know what the junior-level tasks are. It manages the T-sheet or something.

Fabricio Miranda: 42:10

I feel like those are going to be the systems of record that that crush it. No, I think systems of record are here to say and and uh, you always need. I just question what is the role of the system of record? Because today there's kind of a confusion between being the system of record and being the source of truth, and those two things are separate. I think system of record is what connects to the different sources of truth, and the sources of truth are hyper-specialized.

Jared Ward: 42:33

They are going to be AI assistant based and system of record is the one that gets all the information, and it's also giving information to those tools I feel like specifically in the problem that we're trying to solve, because Flieber is more like it can connect to all of your different sales channels and all of your different 3PLs and it just makes it drafts. It's not necessarily the system of record. So I feel like what we'll find out in the next two years is either you will crush everybody or Luminous will be an awesome. No, I don't think I'm going will crush everybody, or or luminous, luminous will be an awesome.

Fabricio Miranda: 43:12

No, I I don't think I'm going to crush everybody. I just do that because I don't have access to data. I I don't think it's Flieber's role to be trying to access data.

Fabricio Miranda: 43:22

Yeah, system of record. It's not a role. I would much rather have you guys having all the data and me plugging to you guys and collecting all the data. That'll be my dream world. It's just that it doesn't exist because my customers don't don't use luminous or don't use net suite and just sell straight to the. There's kind of a a mix of things going on now because of how it's still incipient this, this whole industry, is. Yeah right, but man, you solved that problem. I hooked to you. I'm happy moving away from being a system of record.

Jared Ward: 43:54

Yeah, anyways, well, thanks for coming on, Fabricio. We'll chat about this in a year, and we'll see where everything shakes out. That's awesome. Yeah, Thanks for having me. Guys See ya