The AI Memory Problem Nobody Is Solving (Until Now)

I'm I'm a big fan of of the AI and and Vibe coding in general just because of how fast we can get from that idea to to the demo. But >> can you name a couple of CRM that you like to work with? >> I have notes from 10 years ago that's no longer relevant within my vault. So I might at some point say, "Hey, what was I doing 10 years ago today? " Just to do some sort of comparison, right? That. So there may be some value in going back and looking at uh notes from from 10 years ago, but when I ask something about today, I don't want it to pull today from 10 years ago because it's not relevant. >> Are you seeing an AI opportunity with your app? Are you doing anything with AI in that app?

Good day and welcome to the show. I'm your host Pete Cooper and our topic is AI hype or help. Today's guest is Arthur Ketchel II and with his company a ketchel. Com he helps small to mediumsiz businesses transform with automations and AI. So let's get into it. Good day Arthur. How you doing? Arthur Ketchel II. Nice to meet you. Um >> nice to meet you Pete. >> Thanks for coming on. >> Thanks for inviting me. >> Yeah my pleasure having you. Arthur, tell us um what you do and um what your business is. >> Absolutely. So, I I'm Arthur Ketchel.

I'm um a software engineering leader that that's really been in the the business for 20 years and really my my goal that that entire time has has really been business automation which has really taken off with AI. Um the opportunities that AI has kind of granted within within the the automation space is astronomical. So the uh but the the key at the end of the day is is the the same. It's uh what what I do is build practical software that is used by by the business to operate and generate their u end goal. So it could be a a particular mission that that the company is trying to achieve or it could just be uh one task that that the the business is automating.

So yeah, I did a Google keywords um uh trend Google's trend keyword trend search a couple of weeks back and 900% increase in uh people looking for automations. So it seems to be that the the general public uh or at least business people are aware that AI is unlocking the automations. But you make the point that you've been doing it for 20 years.

So take us back to some of the automations that you've been working on before AI cuz that's I think that's interesting to understand that and of course just for context we all know about Henry Ford and how he automated the building of cars right so I think everyone's familiar with the benefits of you know mass production type automations what are some of the ones you worked on before AI came along >> when I I think that's what's interesting right because AI actually allows more more mass production. So, we're getting closer to the the Henry Ford analogy. Um, but the software development in general has always been a a practice of of automation.

So, it's I I've always said engineers in general, um, founders, software developers, we like being efficient, right? It's uh it's kind of a a key principle of who we are and our goal every day is to to do more with our time. And that's really what it automation and and what AI offers is that that ability to do more with with your time. So I I've automated things as long as I I can remember. Um, I I've got code running in in significantly large organizations all the way from startups to to Fortune50s.

So the within that that space I I've got payment automation, accounting automation, um I I've got in inbox automation, practical like productivity workflows, um knowledge workflows, capturing um notes and um task management. Um, so all of those kind of workflows that that you might think of within a business at one point or another I've automated it. A a significant kind of use case has always been the CRM. Almost every organization has a customer relation management system of some sort. Uh and the the automations that that go in that that sort of um cumer management platform are astronomical. So every organization has a some sort of workflow uh around customer relation automation.

>> So a simple automation I think everyone could relate to is in your emails you might want to forward a newsletter. I did this for my team. Say a marketing newsletter that I receive, I want my se my marketing team to to receive that. So I just set up a little rule that when the email sees that email, it passes it on. There's no need for AI in there. It's just basically searching a set of rules and if it meets those rules that the subject and the the sender's name matches then I automatically send it to uh one of my marketing team.

That's something that I think has been around for quite a long time and that's an automation very simple as it is uh which doesn't require AI and I think tell me a little bit more about how automations work in CRM before AI because I'm going to get we're going to get into AI. So how would an automation work in the CRM before AI? >> So I think you make a a good point there. Rules engines are not new and that's really what that traditional use case is.

Um and and there's really um a conversation going on right now around AI against the the difference between deterministic rules engines which is like what what you described where the input is analyzed and and the output is um going to be the the same every time based on the the same input will get the same output. So that those deterministic rule rule engines have definitely been uh key to to automation for years and they apply at all levels. So on a a CRM for example um you may have a last contacted rule wi within your rule engine of hey this is the last time I spoke to this person. I I don't want to to lose contact with them. So, I should follow up with them every 30, 60, 90 days.

Uh, and that rule when it triggers then can can trigger certain automations. So, that that those types of workflows and that type of automation has been around for a long time. We've all got those uh rules on our inboxes forwarding and and filtering. Um nowadays with with AI you take that to the next level where the the AI adds to the deterministic rule and really starts to make decisions around there that that is non-deterministic. So depending on on the model, depending on the um day of the week, uh the exact context provided to the the model and the the agent at that particular time, you may get a different output, >> right? >> And that's really where the automation is changing. >> Yeah.

What's what's a good example? I'm thinking of something like you generally talk to this person at that company, but somehow you determine that you need to talk to somebody else and you happen to have that information. Um, that's not a strict rule. It's something that has to be, I guess, a multiffactor kind of decision has to be entered into something more like what a human would do. So >> yes, >> I'm thinking >> yeah. So >> yeah, maybe you can give >> a good example there is uh what I have on on the support inboxes of some of my products.

The I actually have an automated AI agent that processes all of those incoming messages and it categorizes it into lowrisk, medium risk, high risk and depending on the uh category that that the message is identified as will determine whether the agent is allowed to respond to it directly and help the the particular user or whether it needs to to escalate to to me or another support representative. Um so th those sorts of judgment calls that that we would traditionally do as humans are now kind of being added to to the rule engine. So >> what's a risk in this context? >> Uh depending on on the the request or the query.

So it from a support context it it'd be something like um the the user's requesting a refund. So the the judgment call uh on that is not something that that the agent is yet able to to decide. So that might need escalated where something like the um the user needs help resetting their password is something that the agent can respond to with with the standard password reset workflows. >> Okay. Gotcha. Gotcha. So and the AI is part of that risk assessment. >> Yes. Yeah. So it's doing the the risk assessment which uh is kind of the enhancement on top of that deterministic workflow.

So that's the the non-deterministic aspect where um I I can tweak what what defines that um layer of hey this is risky or this is not risky by by tweaking the the model or the prompt. >> So that solves the problem for you. Uh as an entrepreneur, a soloreneur or a consultant that you can cut your workload down spending less time on emails and more time on the human critical tasks and I guess have more time at the beach even right >> that sounds good and there's there's still plenty of work to to be done and and I think that that's the the pro to me and software engineers were were one of the kind of first impacted uh from AI, right?

Um because AI can do some of what software engineers have done, but but it can't do everything. So, I I'm a big fan obviously uh of the the AI and the advancements that that it brings, but there's also a lot of fear around it. And I I think that fear is that that it's going to replace um people and people aren't going to to have anything to to to do anymore. And I I don't think that's true at all.

So I I think >> what happens is is our focus is on more important higher level work than the the repetitive task that >> uh it takes a lot of of our our time today and I think that is really the benefit >> and that's uh that's a process that's been happening since the industrial revolution really right we've seen you know what used to be um in the fields with a plow by hand has been replaced by a tractor. So, so you need someone to drive the tractor, you need someone to service the tractor, build the tractor. So, you have a whole industry being built with lots of people with jobs. Now, case building tractors was was originally done bespoke. They would build one at a time.

Thanks to mash production, you know, that's become very routine. So, those routine tasks are now being replaced by machines. So you need people to build those machines and then they have jobs and they service those machines. So we sort of keep evolving the machines to build the machines if you like. And in the same way in your CRM example back in the day there was no such thing as a CRM. There was a sales guy and he had his notes and that was it. >> The Rolodex, right? >> The Rolodex. Exactly. Yeah. If no one knows what Rolodex is, >> you mount all of your cards, business cards into a wheel. Yes. Yeah. I don't have a roll. The original >> I don't think I've ever had a Rolodex, honestly.

>> Yeah. Well, and I think that that's the key of it, right? I mean, it's it's a new level of abstraction. It's a new tool. And for me, that that's really what I've been describing AI to to people is that it's that new level of abstraction. You're adding another layer on on top of what's already existing. Uh and the the added benefits of that u and what you gain from that is you can step back a little bit. So you get that that higher level view and that's that [clears throat] abstraction is beneficial. >> Right?

So if we take the email example, uh I think everybody has an email account right and level of abstraction above that would be the CRM which has some link to its own email or some links to your email and the CRM does a whole lot of things by integrating contacts and history and so on. That's a level of abstraction because you can just look at that and say I know all about that client because everything's there. Another level of abstraction on top of that is what we're talking about right now with the automation. So the on top of the CRM and we we're talking about it this way but I think it helps because the high levels of abstraction naturally flow upwards.

Then there there's these AI tools that are going in there and doing even more. So to cut down the amount of I guess grind work that needs to be done and and take you up to a new level so you can do more with less, right? Which is I guess fundamentally where we want to go with this. And that's really what I've been focused on is how do we take that level of abstraction and enhance it even more.

My my product, the basalt nodes, and that's really what it's all about, is kind of taking that uh it abstraction layer and bringing it all together to to provide kind of the the second brain, the the central memory layer, the central layer of context that that allows both the the human and the AI to to interact. Act well together. >> So, basalt nodes, did I get that correct? >> Yes. Yes. >> Basalt is a type of rock, isn't it? >> It is. Yeah.

So the you may be familiar there there's a few uh >> products in this space but the uh it's a volcanic rock >> and the the idea kind of be behind it is that it really accepts any sort of um document type but it stores it in markdown which is really the the format that AI is uh really good at at utilizing. >> Mhm. >> So translating knowledge, translating documents into that markdown format and and into a a document storage vault uh really kind of allows a lot of of contextual benefits. H so um basalt nodes would sit um an interface with a a typical CRM. Can you name a couple of CRM that you like to work with?

>> So the the Basalt nodes itself um interfaces with a a number of different things not not just the CRM but it can interface with with the CRM. And and the idea there would be that that you would create an organization system that that ties your your notes to a a particular contact. Uh and within the the notes vault, then then you could kind of keep all of your related information to to your contact in your your notes about that that contact. And you can do attachments. And um the the idea there is that then your your AI agent uh can can really search that that context and and the the relationships of your your notes between one another.

So you can really understand how that relationship has come together. So yeah, >> the the tools themselves are um >> really kind of interchangeable. >> Um so it's it's not one one particular system, but but the the concept of having that that vault and that that second storage. So >> So is this an app that goes onto your phone or is it a portal through through a web interface? How do you interact with the salt node? >> Uh it it's both. So, it it's a web app um but it's also a progressive web app. So, you can install it on your phone. Um uh and that's really the the idea of it is you synchronize those notes uh across systems. So, it it's local storage lo local first.

So, the the notes themselves are stored on the device, >> but you can also synchronize them with with Dropbox or or Google Drive. Um in order to synchronize them across multiple devices. So the idea there is if you have an a AI agent, you can type a note into your phone and your AI agent on another device can can then have access to that context. >> Nice. >> Uh and and same with your email. You can forward an email to it uh and log that in your vault and your agent will will have access to that as that second brain or shared brain but between you and your agent. >> So just I had that workflow problem myself and I'm sure a lot of people do. I'm meeting tens of people every week.

I'd really like to know uh how I got in contact with that person. And the way I do it is I manually write it in to my notes in my contact. But then I might send them an email. They might have sent me a document. Um there might be half a dozen different ways through the different tools and apps where I'm now interacting with that person. I'd really love to be able to have a tool that just said, "Okay, Joe Smith, tell me all the interactions I've had with Joe because I'm just about to call him and there it all is. There's the proposal that was last week. There's the how I got referred to him, the last time I spoke to him and have all that right there.

So, like in 30 seconds, I know exactly how to talk to Joe Smith. " Is that what this does? So it it handles the the note and the organization aspect. So the the actual uh relationship to to the CRM um that that part has to kind of be tied to to the CRM itself, but it it handles the the actual note storage and the the vault and the connection. So, so what happens when you log into the app is you you would search um John Smith and all of your notes related to to John Smith would would come up.

Um whether it's a uh you've stored them all to together in the same folder or whether you you've made notes across 18 different projects where you've interacted with with John Smith and and you've got 18 different notes about him. When you type in his name, you're going to see across all of those notes and all of those relationships of exactly uh what th those notes are. And and the idea of the system is not only can you ask the AI about it and it pulls up all of those sources for you, but it also ties back into your your calendar. It ties back into your tasks. Um, so that it kind of helps you organize all of your your thought processes so that you can uh really identify.

Not only did I just talk to to John Smith, but I I also had this action item uh as well as 10 other action items from from all of my other conversations today. So now when when you go to the the daily pulse, it'll show you across all of your notes, here's your your action items for for today. >> You made the exception that that the I don't understand it. Is the CRM linked to the basalt nodes or not? >> Um that that's not a feature that exists yet, but it it can it is something that that could uh could be done.

Um so the currently it's the basalt notes is just the the note takingaking aspect and the the productivity aspects but that is one one use case that I I've seen it used uh pretty significantly. >> So does it tie to many any any of the other typical apps that we would use like the contacts or my emails or a folder in on my drive? Does it tie that in? It it ties in currently to your drive to to your Dropbox and your Google Drive directly. Um we're still working on on the the integrations with the the wider tools. Um it can't yet directly synchronize um with uh Outlook or or Google.

And part of that's because the approval processes that that you have to go through in order to to get uh synchronization with with those um third thirdparty systems. But um it is a feature that that will come shortly. >> Yeah. There's this concept of a super app. Have you heard of that where um when you buy a phone all of the things all of the apps and I I don't know how many I've got in my phone, maybe as many as a hundred uh are all integrated into one user interface. This I think is a would be a huge breakthrough. I think you're taking a step in that direction where doesn't matter what the application is.

If you just want to know about John Smith, um, bang, you just ask it about John Smith and then I guess in that agentic style, which is this question answer style. When was the last time I spoke to John Smith? Oh, okay. Tell me else about the context that's important that I need to know. I'm going to make a call. Bang. It starts pulling all that up and it'll do it from all the different apps straight into you. And then I think there's I believe that that's the way the phone services are going. I think they're going to go towards these super apps and then >> yeah I agree and I think the the idea of basalt is really to be that memory vault uh where that that information is stored.

So where where all of the that information is dumped to in in order for the the AI agent to retrieve it and get the appropriate context and get the information to answer that that question. So that's really where Basalt kind of sits in the middle. It's not trying to to be the super app yet because you've got companies and Google is a good example of that with Google Workspaces that are kind of already in that space where where they're trying to to integrate all their products, right? Um and they've got one in in every category it seems.

But um so it we're we're not necessarily trying to to compete with with Google, but what we are trying to do is is just create that that knowledge space for the the agents and and the the user to to share all of that that information. >> So where does someone get basalt nodes to the can we leave a link in the description so someone can have a look at it? >> Yes. Yeah, it's live at at basaltnodes. Com. All right. Salt nodes. Com. Thanks. Now, you said there's some other products that you do you want to talk about any of the others? >> Yeah. So, I I I've got a a number of products that that I'm working on um in particular.

So, I a couple other products and you may know this that I I'm a licensed real estate agent in the state of North and South Carolina. So I I also do uh quite a bit in in the real estate space. Um so I I've really focused on uh how can I help that that community as well but because it's one that that I'm active in. So, one of the key areas and uh that I've really seen over the years is the community management space.

And so with real estate in general, a lot of homes now nowadays come with homeowners associations and and community associations that uh are really uh it at at the moment kind of focused on the uh regulatory aspects, the the management aspects of running a a a neighborhood or a community uh and just the the management aspects and and really kind of what what I'm trying to to build with >> Do you mean like a sorry to interrupt, but do you mean like a a a house that's part of a larger estate >> type community? >> It's a homeowner homeowners association community. So, it's a a neighborhood of houses that are kind of part of the uh the same association group. >> All right, cool.

Not familiar with that. I mean, I own a home in in PA, but I just bought the house. Nobody said, "Do you want to join a community? " But there is that option. >> Yeah. So, in a lot of cases, uh, and here in in North Carolina, you're not given the option. Uh, the builder when they they build the community. >> Oh, okay. >> Yeah. The the whole >> So, they'll have a whole subdivision and they'll everyone's part of that community. >> Yeah. So the you're just kind of forced to to join it because the the covenants are running with the the land there. Um and and I think that there's pros and cons to that, right?

And and I think that that's really why I I started building the uh GablehHub product, but because there there's so many different kinds of associations and how they're managed and the the key benefit, the key point of building them to to begin with was property values, was uh the community relationships of working together with with your neighbor to to build something that that benefits you all. Right. And and they they've really turned into more of a um policing than than a a community. Um, >> yeah, I got a friend of mine who's bought into a large estate and he was talking about >> I can't remember what it was, maybe putting some trees up.

Everyone had to have a certain number of trees as screen trees and one of the >> one of the people didn't want to pay to put the trees in, but they were kind of forced to because there was the need to create some screening between. But there was all these sorts of little little things that you know, yes, you are do you have your own property, but you're part of a bigger community and everyone wants to live at a high standard. So these are the rules that we establish. So I guess it solves the problem. I was I'd think of the problem solves the problem of of a rogue if you like.

He just wants to be dragging the whole community down because they're dumping cars in their front yard and rubbish out the back or whatever, you know. It gives people a bit of control over that. >> Yeah. >> Well, and that's why they were created. So, but what I what I've been trying to to create there is is a an alternative community management product. So yeah, it it does some of the the basics around the the business aspects of community management um of collecting dues and um dealing with past due d uh homes and violations in case they're they're not following the rules and and all of that.

Uh but really the the goal is to move from just a community policing uh community business platform into to more of a a shared community benefit platform. So there there's a couple different aspects. >> Again, is this an app that has someone has on their phone? >> So it it >> and it's kind of specific to that. Does it have like you know Smithville community? It's white labelled for for them specifically. >> Yeah, exactly. So that that's really the the the goal. So there there's two sides to it. There's the the community um that your your home's involved in >> and then there's the wider community. So the the idea is that you're all already kind of in these smaller community circles.

And um so I I'm building a product there that's called unblocked that that's part of the u gable gablehub network and the idea there is a social good platform. So you're you're part of this community and it's really going to to be and this one is is not yet ready for for launch. I'm just building it still but it's uh it's going to be a social network for for social good.

So as part of that that community >> so I guess the pro the problem of not having an app like if you don't have an app then somehow the information needs to be coordinated with the maybe they have monthly meetings in the community maybe they send out newsletters I don't know whatever they do but if you have the app it's much easier to administer it's all in one place everyone can have access to it you can find out what's going on in the community uh you can vote on it if you like you can have all these sorts of stuff just right there on your app. >> Absolutely. And it acts as that. >> Yeah. It acts as >> it acts as CRM, too, >> right?

So, back back to the CRM of now now you've got everybody's contact information, you vetted them, but because you you know who they are, they're they're your neighbors. Um so, so it really is kind of bringing more of >> of that neighborly feel back to to to social networking. Very nice. That's a great app. And is there anything else out there like that? >> I not that I've seen. I I mean I I think a lot of what the social networks have become have kind of deviated from that because the the social graphs have kind of shifted and and I think the idea here is to to kind of shift back the the social graph to to really be more community-based. Are you seeing an AI opportunity with your app?

Are you doing anything with AI in that app? >> Um, definitely as far as the business management aspects. Uh, so with the community management, with the the violation management, with the the um those sort of aspects there, there's definitely um some AI related features there. Um, one of the things that that that app is is building for is uh communityrun organizations. So, a lot of these organizations have gotten to the point because we're all busy, right? Homeowners don't have time to to run them. So, they've outsourced the the management to property management companies. So a lot of um this particular product is focus is on bringing that back home.

So it's uh it's taking ownership back over from the property management companies and having the the homeowners uh run it as a a neighbor again and instead of just a third party company that that's just policing. >> All right. So um what's the name of your app? Is it community management? >> No. So that that one is gablehub. So gablehub. Com. Yep. >> Right. All right. And that's available too. >> It's available. It's still still in development. So that that one's in a early beta. >> Early beta. And for people that don't know what beta means, it means it's uh released to certain people under trial. Yep. >> Correct. Yeah. Yeah. So, it it's got its uh early user base. It's not without its bugs.

Um so, it's it's still in development. And um for anybody who's interested in testing it out and and willing to to accept some of the the initial growing pains with it, um then then it's available for for them. But with that understanding that there there might still be some some bugs that that need to be worked out. So >> So I'm going to transition a little bit Arthur to talk a little bit more about your experience with AI sort of outside of your own products for a second. Um I think you made a comment that vibe coding has its risks.

Now vibe coding if anyone doesn't know is you use an LLM and you say build me an app and do this and do that and you just keep back and forth until you get the result that you want. Um I believe it was you that said that there's some risks around that technique. Can you um expand on that? [snorts] >> So I I I think as I mentioned kind of earlier as as a software engineer who's been doing this for for 20 years, I I think the the benefit is that that vibe coding allows you to create prototypes very quickly. So it it allows you to to go from from idea to to working example very very quickly and that on its own is a a huge benefit.

I I think the challenge and what >> it's not a new rule but it's it's what people tend to forget is the hardest part of of any uh implementation is is the last 20%. Uh, and I think that's where the the challenges come in uh with Vibe Coding and and I think you're you're kind of seeing it pretty significantly in the AI industry right right now as well where where you're getting results that you did not quite understand um or you did not have the foundation to to scale or the security layer to to protect against >> um challenges that are going to come up. >> Um and that's just not the vibe coding itself.

It's the lack of the foundation, the the lack of the the uh planning layers that that really go into that that development process. So I I think >> when we built software for for the last 20 years, >> everything was built in layers. So you you build the the architecture layer and then then on top of that >> you you may add the the data layer and and then you add the API layer and then you you add the middle layer and then finally you get to the front end. And we're we're kind of starting that in the reverse now with with vibe coding where you're you're starting the front end. It's filling in some of the details in the middle, but it it's not getting all the way through.

And a lot of that's just because it wasn't designed all the way through. >> Right. So, I use the example, too, if you like this one as I was explaining this to a client of mine that yes, you've got a working prototype, but it's like when Elon Musk rolled out his Roadster. Yes, the thing will do 200 mph and it looks all shiny and he drives it onto the stage and isn't it great, but you can't buy it because it's just not produced. It's Yes, it looks like you could go and buy it and it does certain things that tick the box, but it isn't a manufactured product. And he still hasn't produced that. I think that he rolled that rose was it five or six years ago.

I think we're going to get it next year. >> That's right. And it was a demo and there's a lot of stuff underneath there that has been hacked together. >> Um, and and in that case, you know, you've got to produce the tooling, you've got to do uh regulations, you got to build this, and there's software that they hacked from somewhere else. It's just there a whole hodgepodge of things. But to the outside observer, oh, why can't I just buy it? And I have this experience with my clients. They go, okay, well, it's ready, isn't it? Uh, you just see the front end. You don't want to see what we have to do to get there. Absolutely. Well, and that's that's the fun part of of a demo, right?

So, when you're demoing something, no, nobody's going to show you the stuff that doesn't work, right? They're going to show you the the happy path, the the stuff that that works well. And that's really where that that 20% analogy comes in, right? Like the the happy path may be the 80%. But the the non-happy path, the where things go wrong, the the edge cases, the the challenging use cases, that's the 20%. And and that's the >> the part that takes the the most amount of time in order to to resolve. And that's really the the extra layer of work that as I was talking about with each of the layers that that is really missing from from the demo. So AI is really good at the demo aspect.

It's not yet good at at filling in all >> nicely said. Nicely said I think that's a a really good way to describe a problem which I often struggle to communicate.

But I think that uh and tell me if you agree with this when you're building something particularly let's call it an app you still have to do the front end work in other words understanding the workflow that you're trying to solve how someone's going to use it the value creation stuff and that AI can help there but predominantly that's still the same the way it used to be you got to research it you got to know the client know the customers do your testing then at the back end as you say you know the last 20% is still going to be heavy lifting right and it's going to be a lot of work but the bit in the middle has been heavily affected over the last few years. That's come from there to there.

You know, maybe reduced the amount of time, energy, effort, money as much as by a fifth of the cost that it used to be to get to that first demo. >> So, yes, we've shortened the >> the development life cycle, but only that bit in the middle, which is the demo part. The rest of the other two pieces are still >> big and hefty. Well, and and I actually kind of take it a a step further because I I think what it allows you to do is it it allows you to add more features to to add more capabilities faster. Um, but I I think where we have to be careful is the the designation of what what I call kind of spec driven development >> um versus vibe coding.

So it's it's taking every aspect of those layers, defining your requirements, defining your specifications and and building out those layers by the specification and instead of just prompt one I would like to see this u web app that that does XYZ prompt two add now ABC it's I this is the entire focus of the application uh that I'm trying to build. Here's the the end to to end design and idea that that makes it a fully functional app in one specification that can then kind of be broken down and and added to as the the application comes together.

Yeah, we uh we're big advocates of this the scrum the sprint cycle methodology which you know we run internally sprints running every week which is pretty quick but it seems to work for us and we often you know we're often saying okay feature what features are we developing and I guess this uh absolutely >> tends to lend itself to the whole vibe coding type of thing not so much to the writing the specs >> um and we can get ourselves into trouble with with not writing specs because you you just sort of responding in the moment rather than looking at the overall goals that you're aiming to try and achieve.

So, it's almost like you taking us back to that that waterfall type uh approach to software development. >> Well, I I don't think that's that's the idea actually. So, I I think all of my my software development teams have have run scrum agile um for years. Uh and I think the benefit of that is the the iterative development and the quick feedback loops. But what I I think AI has allowed us to do is it it's really taken that feedback loop and made it even faster. Right?

So the the idea of agile, the idea of scrum is iterative development for quick feedback because waterfall has failed traditionally because if you would spend the entire time up front to to spec out everything by the time you got it developed the specifications had changed, new requirements had come up and the end result was never what was intended >> because there wasn't feedback throughout the process. So I think now with AI and AIdriven development, you still are following the the scrum and the agile structure where you're doing quick iterative implementations. You're just kind of spending that time up front within each iteration to really define what your goal is. >> Yeah.

Instead of just prompt A, prompt B, prompt C to to get to your goal, you're going to get to the same goal at the end of the day. It's just spending a little bit more thought up front uh in order to really define what it is you're trying to achieve and provide that fully formed thought to to the AI instead of the the three kind of split up thoughts. >> Okay. So I mean the temptation is to overuse AI and we see this a lot don't we? We see people using >> using using it and not putting the thought into it. Um one of the things that we do it's a thing called a spike. I don't know whether you're familiar with that where we ask people to go and research an activity rather than build something.

And is that what you're talking about? Like just having that moment of maybe it's one sprint to do a spike before you roll out the feature. >> Absolutely. Yep. That's exactly the the approach. Um, so we we've done that for years in software development and agile, right? So, uh, you you create a a spike story in order to test out proof of concept to to create your your specification of, hey, this this is actually what I want to do. Um, and by the end of your spike, that's normally your your output of your your spike story is is your your defined plan of all right now now I've tested this.

This is what I I want to do and spec specri development especially with AI is just following that that exact same paradigm. >> Cool. Cool. All right. Well, we went deep on software development and I think that you and I have obviously a common interest there. I'm not sure much. I mean, I guess people need to know that, you know, vibe coding has its limits. Um, and professionals like you kind of know where those limits are. Um, and I think if anybody wants to, I mean, the takeaway I would give to anybody is, yeah, sure, go and vibe code all you like, but just reset your expectations when you talk to a professional about turning it into a real product.

Because just because you have an app that you've got working on your phone doesn't mean you can roll it out to you know thousands of customers and you know have all the security and the issues and the scalability stuff addressed. I will leave >> and that's where that architecture comes in right is really addressing the those aspects. So, and I think >> again I'm I'm a big fan of of the the AI and and vibe coding in general just because of how fast we can get from from that idea to to the demo but to take it to production to productionalize it is that that extra layer on on top of that. >> Yeah. Yeah. Not to be underestimated. All right. Cool. Cool.

So one trend in AI, one of my questions from here, what do you think the trends are in AI that are affecting you? So I I do think um you've probably seen with with AI models uh context has really been the one of the the key points that that the models have made a a point of the more context the model has the more it can reason. Right? So the >> the context is is think of it as a big note, right? It's a a massive amount of >> of space that that you can just provided information, provided a a very big prompt and and based on that it it knows what it needs to know in order to to do its its reasoning. And at the end of the day, what what is a model?

It's it's a statistical um generator where it it's trying to determine what what the next appropriate token is. >> Um so it it's looking at the the statistics of what you've provided it and trying to decide what what comes next and and at the end of the day that's really what the model's doing but be behind the scenes. So I I think for for trends you you've got a couple of key areas to consider is one is that context management. It's how do I give it enough but not too much so that it it doesn't uh run off and get confused or or hallucinate.

Um, and two, how do I make sure that it has the appropriate guard rails so that I can >> use the the benefits uh of Aentic AI so that I can give the the agent wi which is just a model with with tools and and capabilities, right? Um I I can give the agent the ability to to do something within a defined set of parameters. So I I think a lot of those sorts of of new tools are coming up and I think uh for me it's important because that's really what makes the difference be between the the success and the failure of of an agent.

And that's something that that I'm working on with with Basalt as well is whe when you perform a query, I don't want to return all 10, 000 notes you may have in your system. I only want the five notes that matter to that particular query that you need at that exact moment. So, it needs to search through the entire 10, 000 notes, identify what what those five most important and relevant ones are, and bring that that back into the context. So, I think those two trends are are huge right now. >> So, the growing ability for the LLMs to to have a very broad context, is that what you think one of the main trends is? >> Well, I I definitely think that that's been a trend.

So the the models have been trying to to grow their context and part of that's just because code bases are are so large it's important and the the use case of AIdriven development is definitely a massive one in the enterprise. So it the contexts have have been growing bigger in order to support larger code bases. But that's a pro and a con. And and I think that that's really the the thing to consider there is just because you can doesn't mean you probably should and every case. >> So it's it's not necessarily growing the context, but having the context ring fenced in the way that's that's most relevant to how you need to use it.

I'm thinking of anam example in my case where the LLM that I use seems to have a context of history of things that I've worked on and it'll even sometimes say oh based on the stuff you've been doing this is so it's building up a context from all of my uh previous searches but if I don't use that particular LLM I use something else it doesn't know anything about it and it wouldn't it be great if if I had an LLM which was knew all the things that I was doing and I guess was really in my context text. I'm starting a new initiative. I uh you know I'm talking about golf instead of something related to work, you know, and it would understand that.

So, it's going to have to expand its context, but also keep it relevant to to keep up with me um as I go through my day. [snorts] >> And that's what I'm trying to build with basalt. Exactly. It's it's how do you get the the relevant context? And and from that same aspect, I have notes from 10 years ago that's no longer relevant within my vault, right? So I might at some point say, "Hey, what was I doing 10 years ago today? " Just to do some sort of comparison, right? And see how far you've come and what projects you were considering and stuff like that. So there may be some value in going back and looking at uh notes from from 10 years ago.

But when I ask something about today, I don't want it to pull today from 10 years ago because it's not relevant to >> to that that particular question. So, I I think >> that is is really a a huge >> trend right right now because I I think that's going to make a a massive difference in how successful the the AI tools are is in in being able to choose and refine the correct context for for the the correct use case. >> Well said. I think better context is probably how I would summarize that. We're running out of time, Arthur, but this has been fascinating journey I've had with you. Do you have any uh any thoughts um before we we wrap up? Any questions for me or comments?

>> So, I I think in general I the the synopsis of what we've talked about, right, is is AI is is really a a a helpful tool there. There's a lot of hype that that's around AI and and that's just because everybody's so excited about it, but I I think the underlying capability of of what AI really provides adds a lot of value and and I think just being able to understand the the difference between demo and demo ready and production ready is really kind of the differentiator. But between what what is helpful and what is kind of just hype, right?

So >> I I think AI really can accelerate work and give us the opportunity to decide what better work we we can can focus on wii which as we were talking about in the the beginning is how revolutions and the industrial revolution and all of those have really benefited humanity. And I I think we're there with with with AI. So that that's where it's really exciting. >> So overall your answer to AI hype I help is is help. >> Absolutely. Yeah. Definitely help. Definitely think it's uh it it it's moving us towards the the next technological revolution. >> Yeah. Interesting. Fascinating. Thanks Arthur. Thanks so much for being on the show. Um we had a great journey with you.

Obviously, we'll get those um products of yours in the description so people can check them out. Um and thanks for being on. >> I appreciate it. Thanks so much, Pete. That was great. >> Thanks. >> Bye-bye. >> Bye. >> Well, that was great talking to Arthur. He was a proponent for Vibe coding. Interestingly, even though he's a 20 year software developer, he compares vibe coding to specdriven coding and sees that they are really fitting together even under the agile and scrum methodology just with some basic tweaks. We also talked about some of the big trends that are going forward and he used the simple word context and this is where it's context has been growing.

In other words, understanding more of what you need the AI to know in order to help you with the next step, whatever question it is you're asking for them. So, this growing trend of context is continuing to grow is one of the main trends that we focused on. Also, he has some products of his own which help companies with automations. And one of the big problems that I believe we all have if we're not using automations already is handling all the different apps and how they handle, you know, you have a contact information over here and email over here and bringing those things together so that you have a good understanding of how you're dealing with your sales prospects for example.

Well, if you want to know more about Arthur and his products, please follow the links in the description. If you want to know more about Skillian AI Labs or Skilian, please do the same. I hope that helps and thanks for watching.