The Agentic Insider

TAI - Internal Patents 1-4
===

Speaker 2: Welcome back to the Agentic Insider. I'm your host, Philip Swan. In this show, we explore cutting edge ideas and trends in AI and data, as well as hear from. Other industry thought leaders in the AI space. This podcast is brought to you by arid ai committed to safe and responsible AI innovation. Let's dig in.

riverside_phillip_swan_raw-video-cfr_the_agentic insider_0036: [00:01:00] Welcome back to this week's edition of the Agent Insider. We are doing another internal, focused, uh, episode with the great Peter. Larsen and the amazing Spencer Bentley. Uh, Peter is our Chief Technical Officer, and Spencer is our AI technical fellow. And I'm here with Alistair, and we are going to be digging in today regarding the, um, the four provisional patents that we have, uh, recently filed and the significance of and, and the importance to the industry of why we're doing there. So, Alistair, why did we even pursue patents to begin with?

riverside_alistair_lowe-norris_raw-video-cfr_the_agentic insider_0039: Well, I think one of the most important things here is, uh, you know, from a, from a company perspective, you know, competitive positioning matters. And so, uh, you know, we didn't know originally when we started how many patents we would need. Was it one, was it more? And the thing with the provisional patent is it allows you to get patent pending.

It allows you to be able to have, um, 12 months to file the full patterns while [00:02:00] also sticking a stake in the ground that says, you know, this is the filing date that we can backdate it to. Um. I think that the, the reason we wanted to par file these was because, you know, from a, from a corporate perspective, we wanna protect the, the work that we're doing.

We wanna make sure that, that there's, that we're able to sort of get in front of the, the problem that we're seeing out there and that that problem's quite broad. Okay. In terms of, there's a lot of work happening. Right now around people building agents, very large, you know, agents that seem to do everything and using A-A-G-P-T or a foundation model, something like that, to be able to create something that, that has a very large surface area, which is, makes them, you know, significantly more likely to, to fail because they want, they, they're capable of doing so much.

And the approach here was that we said, why don't we do something different? Why don't we build out a multi-agent system fabric? Why don't we build, you know, a set of entitlements that go with that. Why don't we embed the compliance pieces that we need in here, and ultimately then build a, [00:03:00] a solution factory that can build off at the top of this.

And I think for, for us here, as we worked through this, you know, the, the four of us, and. And Jandy and Emily as well. I think that we ended up with, with four patents that are pretty robust. Um, they put together some key things where we had some breakthroughs, especially around the, in the idea of this int intelligence solution factory that we haven't seen before.

Um, that was sort of the fourth patent that we wanted to get to. Um, and for us, I think, you know. We're talking about 75,000 odd words, you know, 48 drawings, some, almost 300 pages of, of technical content here on how this all lays out. And I think this, this gives us an opportunity at Irid to be able to, to really talk about the, the value that we can bring, but know that it's, it's grounded in the, in good technical principles.

riverside_phillip_swan_raw-video-cfr_the_agentic insider_0036: So. In the ultimate long run, Alistair, I mean, this is, this is great. Why we pursued it. What is the real significance? I mean, are we talking about, I mean, I've, as you know, I've been involved with several [00:04:00] startups. There's only been one that we've gone to this level of, of, of, uh, IP protection. Ma many, if not most startups don't even bother going down IP protection beyond copyrights. So why? Why this level of rigor at this stage of our developments and

riverside_alistair_lowe-norris_raw-video-cfr_the_agentic insider_0039: Yeah, I agree with you. I mean, it's, I haven't seen a startup with this level of IP protected rigor, you know, in, in my experience either. And I think, I think partly that's because we, you know, we had a series of breakthroughs that we wanted to be able to protect. And I think long term, it also helps with the valuation of a company.

Um, because the, the i, if you are able to protect the IP and stick a stake in the ground that says, Hey, you know, this is ours and this is why we're doing it this way. I think it gives, you know, investors and it gives. Customers, the confidence that what's being done here is something that they can rely on and depend on.

And I think that that makes the, you know, that makes our value proposition, uh, a little bit more valuable, a little bit more comprehensive too.

riverside_phillip_swan_raw-video-cfr_the_agentic insider_0036: I agree. I mean, look, my example [00:05:00] is that one company I was referencing, um, that we, that we not only filed our own patents, but we actually acquired a large, uh, portfolio of patents. Um. Some people may remember Hayes modems from, if we're old enough to remember back in those days, we bought all of the Hayes modem patterns.

We were in the in the telecommunication space, and being able to protect our customers and indemnify our customers in the highly litigious environments, which at the time was the beginnings of voiceover IP was absolutely crucial to our success. We were able to drive a valuation that and multiple on the company based on having those patents.

To me, this is part of what we're doing here, do we think?

riverside_alistair_lowe-norris_raw-video-cfr_the_agentic insider_0039: So we should get into, let's.

riverside_phillip_swan_raw-video-cfr_the_agentic insider_0036: Absolutely, absolutely. So star like Peter and Spencer of, if you could go through and Peter, let's talk, let's talk through some of the patents, you know, and what it enables and, and you know. [00:06:00] Giving our audience what you know, the purpose and the real world application example of each patent may be.

Is that something you can do, Peter?

riverside_peter_raw-video-cfr_the_agentic insider_0037: Absolutely. So I think as we were wrapping our minds around what, um, an enterprise customer would want from an agentic solution, we think that they want to have enterprise agents, uh, be, uh, safe and compliant, uh, and scalable and a bunch of other enterprise ities by default. You don't want to have to craft that into every agent that you create, right?

We, we know we are going into an agentic future, but we, we want to delegate that aspect of the agents to somebody else. That's why we created this, uh, multi-agent, uh, infrastructure that is able to communicate between agents, uh, and. Have many, many, many versions of agents to scale all the way down from a single person used to internet scale if need be. And while maintaining that [00:07:00] safety and security that, uh, has prevented many projects, uh, with AI to get off the ground. So we, we basically say here, here's the, uh, here are the table stakes for talking about, uh, agent ai. It has to be safe. I have to be able to, to trust, not worry about my agent doing something nasty to me.

riverside_alistair_lowe-norris_raw-video-cfr_the_agentic insider_0039: And so patent number one is really about this scalable multi-agent system or MAs fabric. So, so let's talk a little bit about that. So, so tell us a a little bit about the, the, principles behind this and what the patent's about.

riverside_peter_raw-video-cfr_the_agentic insider_0037: Yeah, so the, the, the. Because we wanted to end up in a place where we can delegate things to this mass infrastructure we, uh, needed to create a simple building block. So think of Lego as an example, right? A simple building block in, in our environment is a, uh, piece of code or something that is imbued with a code life, maybe from an ai, maybe not. It has a [00:08:00] life of receiving messages and sending messages. That is all it knows. There's a little bit of plumbing in terms of, of logging, uh, uh, what it's doing, and so somebody can monitor it. But fundamentally, from a functional point of view, it just processes messages. So e everything, it can be become an agent in that system as long as you can put a wrapper around it that receives messages and generates messages.

riverside_phillip_swan_raw-video-cfr_the_agentic insider_0036: So how do you, how do you scale in that environment, Peter?

riverside_peter_raw-video-cfr_the_agentic insider_0037: Yeah, so the, I, uh, the trick there is to, um, reduce the agent behavior, uh, and hide it behind a scaling mechanism. So you, you can, you can craft this in many ways, but one of the ways of doing that is Kubernetes, right? You put the agents together, you package them in containers, and, uh, this is a, uh, something what we call semantic overloading, where we have many agents in the same container.

And then whenever an agent sends a message, it may in fact. Course, another [00:09:00] copy of the recipient to be spun up in case the existing, uh, copies are already busy, uh, making a request on, uh, on a further customer, so to speak. So the environment scales, uh, transparently to any individual agent and their function, and it scales, uh, to the limits of what an Azure, uh, uh, GCP or AWS can offer.

riverside_phillip_swan_raw-video-cfr_the_agentic insider_0036: Which is, which is amazing. Which, but now you start getting into, okay, you were talking about individual containers. Now can agents talk to other agents in other containers?

riverside_peter_raw-video-cfr_the_agentic insider_0037: Yes, so, so they agents do not know where other agents live. So there is a, uh, uh, essentially a routing mechanism that allows agents to be, uh, transparently talk to other agents, uh, uh, that they have been told, uh, uh, good at, uh, solving a particular problem and they don't need to know where they are.

riverside_phillip_swan_raw-video-cfr_the_agentic insider_0036: So, and ideally those list, those containers could even be running another machines, correct.

riverside_peter_raw-video-cfr_the_agentic insider_0037: That that's exactly right. And, and we are talking to, uh, [00:10:00] people who care about devices, right? So all of a sudden, you can imagine, as long as you can get the security down that something is running on my Kindle and something is running on my cloud.

riverside_phillip_swan_raw-video-cfr_the_agentic insider_0036: Let's double click into that. 'cause security is something that's near and dear to all of our hearts. Right? And, and how we're handling that. I mean, you know, I, we're, we're doing the basic. You know, we're not only just doing the basic protections, we actually have some pretty comprehensive security features built into the product to do that. Um, you know, not just, you know, how we handle data and everything else. Do we want to think about how, you know, how, you know, what do we protect and, and, and think about this? And and part of, part of what we're talking about is a zero trust model in this, in this architecture.

riverside_peter_raw-video-cfr_the_agentic insider_0037: Th That's exactly right. And, and there, there is, uh, there, there are many elements from, uh, existing, uh, uh, notions of, of, of web services and, uh, and, and, and you know, SOA and stuff like that, that, that carry [00:11:00] over, right? If you go into a cloud like that, let's say Azure, right? You have, uh, infrastructure in Azure to, uh, identify. Things that are doing, uh, stuff on behalf of humans, right? So in, in, in my Azure and an agent will have an intra id, so we know who it is. It, it is cryptographically ensured that this agent is who it says it is. Once you have that, uh, level of trust established, you can attach things onto it, right? You can attach entitlements onto that, uh, identity, and then you can build on top of that a, a system where we don't trust each other, but, but there's certain things that are stay in variant.

riverside_alistair_lowe-norris_raw-video-cfr_the_agentic insider_0039: And, and I love that because we we're almost now moving into patent number two, which is around agentic entitlements. And so we've got this, we've got this model that's built on top of Kubernetes. It extends Kubernetes, but it allows it to be able to scale up and down with semantic scaling. And I think in addition, you know, Philip, to touch on to the point that you are making here, one of the architectural pieces that we have in this is the [00:12:00] concept of a postman.

Okay, the a to, to Peter's point earlier, the agents don't know where the other, where the, all the other agents are. But the postman is able to deliver messages. So messages go transiting through the postman who, which is again an, an infinitely scalable resource. And it transition transits the messages, you know, outside of the pod to other pods.

And I think at that point then, before a postman delivers a message, it validates that the entitlements that is coming with the message are those that the agent that it is delivering to are correct. So it's the same with any letter you are looking at it. Is this the, is this the right name on the address?

Is it the right, um, is it, is it, uh, the, the right address? Is it formatted correctly? And then specifically, is this address allowed to receive parcels or packages, or is it, uh, not, not allowed to receive things with a battery inside the moment? Extending the analogy here, but ultimately, you, you know, it's making a decision.

So it says, okay, what I'm delivering to, what are you willing to accept? What's in the message? What have I got? Do they match? If so, then we [00:13:00] can, we can move ahead and go with the delivery. Peter, do you know, do you wanna expand from there?

riverside_peter_raw-video-cfr_the_agentic insider_0037: Yeah, I mean, I, I, I think that the postman scenario can go very far, right? You can think of, uh, of ancient, uh, Sumerians, uh, uh, putting clay envelopes around, clay messages with, with, with, uh, uh, uh, little, uh, script on them, right? And, and, and the notion of, of. Being able to seal packages with packages inside of them goes very far.

Right? Uh, one of the things that we, uh, can do in the environment is to, to have a transparent, um, adding of stops on a route of messages to other value. Adding functionality so that you can have, uh. Both, uh, oversight, uh, from a standards perspective, but any cross-cutting concern can be inserted after the fact.

The system can be essentially insert the oversight or the, the value add across, uh, all the existing agents, uh, totally transparent to the functionality of the [00:14:00] individual agent.

riverside_alistair_lowe-norris_raw-video-cfr_the_agentic insider_0039: Absolutely. And, and let me just go on from there as well. So we're talking about, you know, standing up and establishing a MA fabric that runs all of these, these applications and these agentic systems. But it's not only gonna be the ones that we instantiate, there are gonna be other MAs systems out there.

Okay. So, so then there's the concept of the, of the boundary postman and MCP and A to a. Can you talk a little bit about that?

riverside_peter_raw-video-cfr_the_agentic insider_0037: Yeah. So in, in, in, uh, many ways, and, and, and, and I want to pull in Spencer here as well to, to, to talk about MCP perhaps because essentially the, a bunch of companies, Spencer right, have come up with individual answers to, uh, a Genentech behavior. So maybe you could expend on that.

riverside_spencer_bentley_raw-video-cfr_the_agentic insider_0038: With MCP. specifically, um, uh, uh, it's worth the, the elephant in the room is, is a security concern around MCP. Um, it's not specifically with the, uh, security of making the connection to MCP. You can do that, but once you have [00:15:00] made the connection, you've then got access to the model directly. And if you can insert improper commands, uh, you can potentially access anything that model has access to.

So for example, if, if a model has a a.

rag database that's a retrieval augmented, uh, database where the model is able to pull from company specific information, and that could be a customer database it could be anything. Um. If it's connected to the model, you have to make the assumption that this is now public domain and that was not known before.

And so a few, a few companies have come unstuck with that. um so one of the things we want, we try and do is build on top of MCP, to, to wrap it in a protective layer. So that's one of the approaches we do is, is we create a, a, an idea of [00:16:00] a bubble and things outside of the bubble. We'll require zero trust and a lot of scrutiny, and that those things within the bubble we can treat less so.

riverside_alistair_lowe-norris_raw-video-cfr_the_agentic insider_0039: So it's more of a sort of a force field around things that, that allows us to be able to protect, you know, things going in and out. And as Peter's point earlier, as we can insert almost an infinite number of checks, okay? And balances through this as we route messages, it allow. Us to ensure that the zero trust principles that Philip was, was espousing earlier can, can be used even if we are talking, you know, from AM to, to any, any other person's, ma.

riverside_phillip_swan_raw-video-cfr_the_agentic insider_0036: one thing, Spencer, you were mentioning mentioned in the past, and I love the analogy, is think of Mt. P as HTTP and think of your of Iridius built around your Iridius core. Built around that is giving you the the S in front of HTTP to give you the security. Is that a fair

analogy?

riverside_spencer_bentley_raw-video-cfr_the_agentic insider_0038: Yeah. Uh, it, it, it, it's not exact because, um, I don't think Mc [00:17:00] PMCP with, uh, OAuth for those who want to know the, um, more technical definition for it, uh, it, it is secure. It's, it's great. The That isn't the problem, which is what H-T-D-P-S basically solved. But yes, the idea is is HTDP was around to start with, and then people realized it wasn't secure enough.

So H-T-D-P-S was created. MCP at the minute in its current incarnation is not secure enough to be allowed out alone. It needs to have a chaperone,

riverside_peter_raw-video-cfr_the_agentic insider_0037: and, and and that's the second patent. The second patent is the, the chaperone, right? Because H-T-D-P-S ensures that you can have a, a, a, a confidential conversation, but it does not ensure that you have a right to talk about it.

riverside_alistair_lowe-norris_raw-video-cfr_the_agentic insider_0039: Exactly. So why don't we move on a little bit and talk about patent number three. Okay. Which is around the, the idea of the, the dock rocking [00:18:00] and the semantic compliance compiler and things like that. Why don't we talk a little bit about that.

riverside_spencer_bentley_raw-video-cfr_the_agentic insider_0038: Yeah, so I think it's, it's worthwhile, um not concentrating on, uh, a, a particular standard or even the standards. This is more general. How do you get a large corpus of information? And understand it in a reasonable size. Now, you could take the naive approach and take, take the thing you're interested in and include it in its entirety to the model, and that works. If the document is very large, the context may start to reduce the quality of the output from the model. um and if the, if the corpus so you've got a thousand documents, each one with a thousand pages, there's no model in the world that can take that. So what do you [00:19:00] do at that point? How do you understand the requirements of part?

Something like a a a ISO standard can have. a 70 page document with references to 50 other standards, each referring to other standards. And so you get this great bifurcation of, of documentation. How do you do that? So one of the, one of the realizations we came to was information in general consists of human beings, put them in in sentences, into paragraphs, into pages.

And we do that with almost an unconscious. Logic we group. Uh, topics into sentences. We group groups of topics into paragraphs. We group groups of ideas into pages, and we take advantage of the fact that that is embedded into the information we have in our world. We logically split the information up into its components, along with the rules [00:20:00] and the references, so.

Is there a reference to an external document? If there is, we make a note of that. Is there a diagram? Does it contain a formula? Has it got a table? We extract those details and store them formally. um and then the, once we've done that, we, we've then got a, a detailed list of all of the information and how it relates to other parts of the document, and then we can put all of that information into a graph database.

And we can build the relationships between the information. When you do that, you can arbitrarily add more documentation to that database and they will occupy the same links. so you can follow from an idea or a concept. Through the graph. And if we then do something called vectorization, which is where you take, uh, an embedding layer from a model.

It's a part of this retrieval augmented generation. Um, feature that [00:21:00] everybody uses, but that vectorization allows us then to very quickly find a location in the graph and then traverse the graph to understand the knowledge around that topic. And that lets us take very esoteric, very complex, um conceptual ideas and find the context for those and bring them into the AI in a compact, uh, form.

That is directly related to the question being asked rather than a general, here's the entire document you work out which bit's important. We are looking through the index, finding the thing that's important, looking at the index, then going, Ooh, that bit's mentioned in this document, this document, this document, pulling those pages together and then showing that to the AI.

riverside_alistair_lowe-norris_raw-video-cfr_the_agentic insider_0039: Mm-hmm. Rather than all of the documents, and that's ended up being extremely powerful and forms the basis for the dock rocking system.

Sure. I mean, I, I think here the big thing is [00:22:00] that there are hundreds of regulations out there around ai. I mean, there's hundreds of regulations and thousands of regulations, but other things. But for AI specifically, the EU AI Act is 500 pages. Okay. Colorado and Utah have their own, have their own per pieces of this as well around an AI act.

Um, Texas just put one in for the states. There's Cal, California has some, every single state is doing something. Singapore has, has, um, has, um, has, uh, regulations around this. So does Australia, so does the uk. So, and everybody is, everyone is looking at what other, um, other legislators have done. So governments are looking at what's been done in the past and in some cases they're referring to, you know, the same definitions, but in many cases.

Then not. And so you end up with when somebody say, when, when, when, uh, you know, one country A or region A says transparency. It's not exactly the same way that Country B or region B is talking about transparency. And if you are a company and you're an enterprise and you have to be compliant with all of these standards around the world and the [00:23:00] countries of which you do business, then it suddenly becomes more and more problematic.

And then, you know, in addition to that, just as you are saying, Spencer, there are, you know, hundreds of standards around ai, hundreds of best practices, more on top of this. And the complexity is almost impossible for people to be able to get. And I think if we are able to bring it all together and then harmonize it and understand where the edge is and where the, where the issues are, and then, and then, and then find a way to be able to help people navigate through that, suddenly we've got a, a corpus of knowledge, you know, to, to your point that that can be used.

To ensure or assure compliance in a lot more, uh, a better way because it's directly related to the, to the pieces that are.

É and, just adding as a comment here, right, that yes, the standards body is large. Your corporate and your corporation's knowledge base is also large, right? So you, you are going to want to, uh, restrict agents to proper behavior, not only to the external standards that you are mandated to do, but also to [00:24:00] your own ethics.

Absolutely. And you may have a code, may have a code of conduct or a way of working. For example, you want, you want the agent to be able to communicate with people, but the, you, you know, your organization uses Slack or your organization uses teams. And also the, the way that Disney has, you know, um, has, uh, has its employees talk to each other would be different than, than how Ford.

Has its people talk to each other and different than Coca-Cola because they're different companies. They have different codes of conduct, different international, you know, and enterprise standards. And it may well be that the, the, the marketing team in France okay, needs things done slightly differently than the marketing team in Germany.

And so now we have. Corporate standards, European standards, marketing standards, marketing in France, standards, and it has to be able to take all of those and overlay them, okay. And allow it to be able to have the right context for how to operate inside a company as well. Quite right, Peter.

riverside_phillip_swan_raw-video-cfr_the_agentic insider_0036: And flipping that on his head a little bit too. I it, I'm just putting it in human terms. Who would [00:25:00] hire a, a band of interns, 20, 40, a hundred, a million interns, and just let them loosen their company without any training. That's what you're doing here, right? So you have to train these agents. You have to give them a code of code, you have to give them the entitlements and permissions of what is, or

permissions, I

riverside_peter_raw-video-cfr_the_agentic insider_0037: you you don't just give them a rifle and turn 'em loose on the battlefield.

riverside_alistair_lowe-norris_raw-video-cfr_the_agentic insider_0039: No, this is the toddler with a chainsaw situation all over

riverside_phillip_swan_raw-video-cfr_the_agentic insider_0036: yeah. And yeah, and expect good things to happen. Exactly.

riverside_alistair_lowe-norris_raw-video-cfr_the_agentic insider_0039: So, I mean this, I think this, this is great because, you know, we talked about patent number one, which is around the, the mass fabric. Okay. We talked about patent number two, which is why entitlements are important. And then we've got the idea of, okay, now we can understand the, the compliance side of it and ensure that the things are compliant by design and compliant and execution.

But one of the huge. Huge, you know, breakthroughs that I think we had was, um, was why we decided to delve into patent number four. So, Spencer, can you talk a little in your, you know, because this, this really was your brainchild here, what was the problem? And then suddenly [00:26:00] what was the way that we saw through this that led us to patent four.

riverside_spencer_bentley_raw-video-cfr_the_agentic insider_0038: this, um. Has a lot to do with what is currently called vibe coding. Um, so there started to be, uh, models got good enough that you could, um describe what you wanted and the model on occasion w would actually produce the thing you wanted so that that happened. And as the models got better, that happened more and more of the time.

Um, but. Absolutely nowhere near good enough to put into an enterprise where you've got security regulations, you've got scaling rules, you've got, uh, code that needs to be built in a certain way. um all of the, the parts of an enterprise that make enterprise work are almost exactly opposite to the waying works.

So the question is how, [00:27:00] how could you get an AI to create code? That is structured and, and well-formed. um so my realization was that we have done this before. How do you take, uh, potentially unreliable intelligences, group them together and then make them produce something of high value? And the example I I give is, is a, a satin five rocket.

You've got, not sure of the number of people, but I imagine hundreds of thousands, if not millions of people worked on that rocket. And there are millions of bolts and threads and metal plates and all kinds of things that need to be done without mistake multiple times. How? How do you do that? And it turns out we have formalisms in industry that describe how you [00:28:00] tell, um.

Intelligences that may make mistakes. How do you monitor them? You have line managers, you have department heads of department. You have a structure that's built around managing intelligence and it seemed to be a logical leap to go, well, AI is a is an unreliable intelligence. Can we use the same ideas to push that towards enterprise quality output?

And the, the, the short answer is yes, you can.

riverside_phillip_swan_raw-video-cfr_the_agentic insider_0036: That's great. So now we're, now we're get, now we're getting into our complete, our complete solution. So in patent number four with the Iridius Solution factory, we, we've been talking about this from the very earliest days, Spencer, you and I, you know, at the product perspective, and it's, it's becoming real. What's got you, what's got you excited about this?

riverside_spencer_bentley_raw-video-cfr_the_agentic insider_0038: the more [00:29:00] general, um, view that if a person can do it, we can probably, um find the rules. The person uses and replicate, replicate those in ai. Um, which gives you a really effective way of saying, how can I do this? I don't know. oh, wait a minute, how did people do it before? And now it doesn't work for everything.

But there are lots of instances where you can say, oh, right, I see how it was before and how it should be. So one of the examples is. The way that, um large single model coding agents work is like A an artisanal, uh, craftsman. They do it very bespoke. They do it, uh, piece by piece. It takes a while. There's a lot of skill involved between the creator and the agent.

And you end up with a unique bespoke thing. [00:30:00] Fine. That's not what industry needs. Henry Ford, Spotted that and he said, ah, what we actually need is lots of small, dedicated agents that can do one thing reasonably well and put em all in a line, and that realization means that we can apply that to agents as well.

riverside_peter_raw-video-cfr_the_agentic insider_0037: I mean the, the, I, I like to think about it as quantization, and this is what dis uh, distinguishes what we are doing from the vibe coating, that we are creating a quantized, um, target environment of the multi-agent system that, uh, only allows you to do things that are good.

riverside_phillip_swan_raw-video-cfr_the_agentic insider_0036: So why does by coding not let you do that?

riverside_peter_raw-video-cfr_the_agentic insider_0037: Vibe coating is open-ended. Uh, so vibe coating lets you do everything in between.

riverside_phillip_swan_raw-video-cfr_the_agentic insider_0036: So by coding is great for rapid prototyping versus

riverside_peter_raw-video-cfr_the_agentic insider_0037: Yes. And, and, and the chance that you are going to step on a landmine I is, is much, much higher.

riverside_spencer_bentley_raw-video-cfr_the_agentic insider_0038: [00:31:00] Yeah. At the minute there is, there's two con, two conflicting forces. One says five Coding is awesome and it's the best thing ever. And then the other group goes. Vibe coding is gonna destroy enterprise and industry. It's, it's unsafe. You should never do, it. And as is typical, the, the truth lies somewhere in the middle. vibe Coding is incredibly powerful for, is this even viable? Can it be done?

riverside_phillip_swan_raw-video-cfr_the_agentic insider_0036: Yep.

riverside_spencer_bentley_raw-video-cfr_the_agentic insider_0038: not what you want to push into production, right? There are two. Those are two things. Stop conflating. Don't do that.

riverside_alistair_lowe-norris_raw-video-cfr_the_agentic insider_0039: Building a vibe. Coding is not aimed at building you a full enterprise grade safe compliance solution, okay? It's about quickly mocking up ideas, seeing if they would fly, and then saying, okay, now you know what we need to do. We need to build a real backlog with real tickets in it and real items, and we're gonna go off and build this thing properly.

So you've got a quick prototype be it didn't build a thing, okay? The, the, the intelligent [00:32:00] solution factory of patent number four says, Hey, why don't you take the principles. Of, of building something quickly. Okay? In a vibe in, you know, in that, that vibe's talking about. But this time, why don't you not just mock it up?

Why don't you build the entire enterprise safe compliant solution in one go. Take that. But that, that jump or that leap from, why don't we just do a mock up to why don't we have a fully compliant enterprise grade solution is such a huge leap that that's I think the big thing here.

riverside_phillip_swan_raw-video-cfr_the_agentic insider_0036: And this is really the last mile. So tying the four patents together, Peter and Spencer. from a customer standpoint, what does the future look like for our customers? And, you know, and, you know, in terms of the future of AI and, and what does, what do these patterns mean, you know, for our products as a whole and for our customers.

Mm-hmm.

riverside_peter_raw-video-cfr_the_agentic insider_0037: Life, the universe and everything. It's, it's a big question, right? Philipp? I, I think that one of the, uh, forces that the, uh, [00:33:00] collective, uh, patterns and, and our product will in, uh, impose on, on industry is a democratization of, of doing your job. And, and in, in the previous wave of this was sort of the co-pilots of the world where you said, uh, if you have something as a single contributor, uh, we, that you can, uh, automate a little bit, the co-pilot can do for you. That is, uh, limited by, uh, it, it's, it's, it's in front of a person. Uh, the solution is in front of, of the company. Right. The, the copilot is, uh, it requires you to say, okay, are you sure you wanna do this? Uh, the solution says, I have already gotten the permit permit to do certain things already and I'm going to proactively seek out other solutions.

So there's a level of, of scale in terms of organizational scale, there's a level of scale in terms of internet scale, in terms of the consumers. On the other end, there's a level of, [00:34:00] um. Authorization in, in, in, uh, imbuing, uh, the executing environment with preauthorized, uh, rights to do certain things very well confined, that allows you to, um, uh, just do things faster.

riverside_phillip_swan_raw-video-cfr_the_agentic insider_0036: That's awesome.

riverside_spencer_bentley_raw-video-cfr_the_agentic insider_0038: Yeah, I, one of the really important things for me was I, I've always found it really valuable to be able to try something out, and that's kind of what the Vibe code gives an individual coder, that is a really powerful tool. Could we give it to Enterprise, And I think the answer is yes. We can give them an almost. Zero cost it. at enterprise it, it's going to be effectively zero but in terms of speed resources um, and money you can try something and you can iterate on it very rapidly [00:35:00] and maybe produce something really valuable and that's something you cannot currently do in enterprise everything has to be done in a very slow methodical way and We think there's a way of doing that. Automatic,

riverside_peter_raw-video-cfr_the_agentic insider_0037: it challenges, the position of SaaS companies a little bit too, right? I mean, the existing SaaS companies are islands of stability. Uh, if we, if you say to the industry, actually, stability has been pushed into the wall socket.

riverside_phillip_swan_raw-video-cfr_the_agentic insider_0036: right. Alistair, any last thoughts?

riverside_alistair_lowe-norris_raw-video-cfr_the_agentic insider_0039: I think this is giving, you know, companies, enterprise companies, an ability to rapidly iterate and develop, um, solutions that allow them to gain a competitive advantage. I think this is something where they're going to be able to innovate. Faster and to some extent fail faster, but succeed faster.

They're gonna need to find those things that, that do not work faster and get rid of them in order [00:36:00] to be able to step towards the things that work. And I think that rapid iteration is huge, and I think they're going to be able to do it in a way that's, that's safe and compliant so they know they're meeting all of the right standards.

So they, they they have the confidence that even if they're in a highly regulated industry, okay, where lives are at stake and people are going to be affected, they're still going to be able to build things. Quickly and test things out with customers in the market. Okay? And, and, and fail faster. And therefore, without as much money and time and resources spent, which means that they're going to be able to innovate in such a freer way that I think as, as, as Peter says, because this is being pushed into the wool socket, you literally just plug a device in.

You don't have to worry about how the electricity works. The microphone connects the power. You know, you, you, you, you can plug a computer in, you take it out again. You don't have to worry whether the electricity is being fed to you in a certain way. If all of that compliance and all of that safety is being handled for you, then, then really the, it's up to you to develop and build, and you're only limited by your own creativity at that point.

So I [00:37:00] think it's a really huge opportunity for enterprises going forward.

riverside_phillip_swan_raw-video-cfr_the_agentic insider_0036: I agree, and the one thing it does not stop is a bad idea turning into a good outcome. So you still have to have good intent and good ideas to actually make it work. And um, I want to cautioned everybody on that one. Uh, my, my brother Spencer has, has brought that up and articulated that a, a couple of days ago, and I go like, he is absolutely spot on.

So. Thank you to the three of you and to the entire team at EDUs for the work that went into these, um, uh, these amazing provisional patents and the work that's going into our, our products to save our customers time, effort, and money, and also create the safe environment that they want for not only. Their customers and reputational risk, but also for employee safety. When we deal with, with firms like, you know, we are dealing with global firms in [00:38:00] pharma and, and energy and manufacturing and each, and in each of those scenarios, people die, um, as a result. So public safety is a very key, uh, key thing that, uh, we can help them survive. Thank you to our audience for listening. Join us again next week and thank you again, Spencer, Peter and Alistair. Thank you.

Speaker: That's a wrap on this week's episode of the Ag Agentic Insider. Thank you for listening. For show notes and more, please visit AgTech Insider Show. To learn more about arids visit arids.ai. We'll see you next time.