When AI Meets Sales, Support & Supply Chain: Omnae & Bardin AI

Episode Summary

AI is spreading beyond CAD and simulation into the industrial back office — pre-sales, application engineering, quoting, procurement, supplier collaboration, invoicing, and customer support. In this episode of AI Across the Product Lifecycle, Michael Finocchiaro sits down with Fay Goldstein, Co-Founder and CEO of Bardin AI, and Scott Lionello, Co-Founder and CPO of Omnae Technologies, to talk about where industrial AI actually creates value and where it can quietly destroy trust.

The discussion focuses on determinism, auditability, and human-in-the-loop workflows as non-negotiables when AI touches contracts, general-ledger data, or supply-chain commitments. The guests examine why token-burn pricing is unsettling procurement teams, how mid-market manufacturers may leapfrog large enterprises on adoption, and why the "OpenAI moment" for supply chain will arrive in segmented waves rather than a single breakthrough — with real value showing up first in the unglamorous parts of the business.

Bardin AI is building an application engineer for industrial automation sales and support teams, helping non-experts answer complex engineering and application questions without escalating everything to senior engineers. Omnae builds supply chain collaboration software designed to work across many companies, including smaller suppliers, while giving AI agents a safe and deterministic environment to operate in.

We discuss why industrial AI cannot just be “ChatGPT bolted onto a workflow,” why auditability and trust matter when AI touches contracts, engineering claims, supply chain commitments, or financial systems, and why the best AI use cases may start in the unglamorous parts of the business.

Topics include:

• Why Fay was “nauseatingly bullish” on AI from the beginning
• Why Scott remains more skeptical when AI touches the general ledger
• Why industrial AI needs determinism, traceability, and audit paths
• How AI is changing software development inside startups
• Why founders can now prototype product ideas much faster
• How AI changes agile, product management, and engineering team structure
• Where AI sits inside Bardin and Omnae
• Why not all AI in industrial software is generative AI
• Why human-in-the-loop workflows still matter
• The economics of LLM usage, API keys, and token burn
• Whether supply chain will ever have its own “OpenAI moment”
• Why entry-level talent needs AI fluency, not AI fear
• Why mid-market manufacturers may adopt AI faster than large enterprises
• Why startups may outperform incumbents in delivering practical industrial AI

This conversation is less about AI hype and more about execution: trust, workflows, liability, integration, adoption, and the hard operational details that decide whether AI actually works in industrial companies.

Timestamps

00:00 — Intro: switching from CAD/BIM to sales, supply chain, and pre-sales 00:32 — Fay Goldstein introduces Bardin AI 01:16 — Scott Lionello introduces Omnae Technologies 02:26 — First reactions to the OpenAI moment 03:36 — Why AI touching the general ledger requires skepticism 04:01 — Trust, determinism, and industrial AI 05:00 — Auditability, knowledge graphs, and regulated AI decisions 06:25 — The missing “black box” for AI and robotics 07:41 — How AI coding tools changed software development 08:37 — Scott on smaller, faster engineering teams with AI tools 09:34 — Fay on AI enabling non-engineer founders and product prototyping 11:50 — Bardin Flow: internal AI systems for startup operations 13:17 — Is AI changing agile and software craftsmanship? 14:37 — Freedom versus discipline in AI-assisted development 15:37 — Why AI agents need skepticism 16:26 — Where AI sits inside Bardin and Omnae 16:43 — Bardin as an AI-first industrial application 18:12 — Omnae as a “jungle gym” for supply chain AI agents 19:36 — Human-in-the-loop workflows and trust 20:50 — Why accounting personas resist uncontrolled automation 21:22 — From AI distrust to “prove to me it works” 22:18 — The economics of LLM-powered industrial software 23:11 — Usage-based pricing, user-based pricing, and uncertainty 24:17 — Omnae’s approach to API keys, rate limits, and token control 25:17 — Bardin’s long-term infrastructure vision 26:09 — Token-burn culture and AI leaderboards 26:45 — Will supply chain have its own OpenAI moment? 28:21 — Fay on segmented AI moments across industrial workflows 30:31 — Dark-stack AI in procurement and supply chain 32:01 — Lessons from legal, medical, and regulated AI adoption 33:06 — Advice for young professionals entering the AI era 33:40 — Fay: show your AI stack, not just your résumé 34:47 — Scott: people skills, product sense, and initiative matter more 35:35 — Digital maturity in industrial companies 36:19 — Enterprise stacks versus how work actually gets done 37:03 — Why manufacturers may use agents more deeply than expected 39:10 — Can startups move companies faster than incumbents? 39:57 — Fay: the goal is relief, not demo theater 41:04 — Scott: startups can prove value in smaller deployments 41:38 — Why top-down enterprise AI rollouts often fail 43:41 — Where to meet Fay and Scott 44:57 — Sponsor mention and wrap-up

Please don't forget to click on this link from our sponsor AWS for access to an exclusive webinar! https://pages.awscloud.com/awsmp-gim-yngd-webinar-aim-enterprise-ai-and-data-leader-panel-lt-panel-1.html?trk=730a334a-28e3-4e91-960a-fc94de422926&sc_channel=el

---

Full Transcript

Speaker

And we're live. Hello everybody. This is Michael Finneran on the AI Across the Product Lifecycle podcast. The second one of today. We had earlier Conic and Raven talking about BIM and and CAD, which is good. Now we're kind of switching to the dark side of marketing, pre-sales, supply chain. It'll be a lot of fun. Hey, why don't you introduce yourself and Bardeen?

Speaker

All right. My name is Faye. I'm the co-founder and CEO of Bardeen AI and essentially we are building an application engineer that sits in the pocket of every industrial automation sales and support team member. So essentially all those really complicated engineering and application minded questions that folks that are buyers of industrial automation systems have no longer need to only be escalated to your top engineers, but your sales folks and your support team could actually take and answer and support those questions and those queries that they've got help them along the sales process and the post-sales process as well.

Speaker

Oops. Sorry. The tab I hope it didn't have wasn't muted so I need to fix that. Sorry, go ahead Scott.

Speaker

Yeah, my name's Scott Linello. I'm a co-founder and CPO at Omni Technologies. We provide many-to-many supply chain collaboration solutions that notably actually provide a service that works for the small businesses you're trying to collaborate with and allows you to get agents playing in your supply chain in a way that's safe and deterministic enough to be workable today.

Speaker

Awesome. I'm really glad to have both of you guys here today and I'm an international group too. I've got Vancouver and Tel Aviv so pretty far apart.

Speaker

That is must be the biggest time zone spread you've you've hosted in a while.

Speaker

Most likely, most likely. Yeah, I don't think Yeah, I've never done with I've done I've never done with an Indian company yet. Probably an Indian and California would be pretty close.

Speaker

Oh, yeah.

Speaker

Yeah, I can hop pan in Australia and you got the there.

Speaker

Well, there are some startups there, but anyway, we're about what 4 years into the whole Open AI revolution, right? The this before and after. Um, I'm wondering as founders, were you guys skeptical or you super bullish as soon as it came out? What what was your reaction to the Open AI moment?

Speaker

I remember it clearly November 2022. In fact, I left my job in venture capital with the belief that I would be able to finally take the dream of wanting to be an engineer and build things without having to have gone and and and actually studied it. Um, I I I remember my first engagement at that point with ChatGPT. It's interesting because it's like I think a lot of us as consumers start the moment from November 2022, but anyone that has been in the machine learning side knows that this goes so far back that, you know, AI is just a euphemism in a sense for really just, you know, machine learning. But I think the moment really that everyone's talking about is that November moment and I was incredibly bullish. Incredibly bullish on how to think about it, how to approach it, how to learn as much as possible about it, how to use it on a daily basis, and still am nauseatingly bullish about it.

Speaker

Thank you. Scott, were you as in nauseatingly bullish?

Speaker

In in our very specific world of things that touch the general ledger, there was a lot of skepticism and there kind of still is in the world where if it has to be penny perfect every time and there is zero fault tolerance, You have to do a lot work before you can let it loose. So, it's it's it's coming together, but it's it's been a slow burn for us, I would say.

Speaker

Yeah, I I think that's interesting because there are those two perspectives. How do you look at it from a consumer perspective when you're just an average person using it versus when you're building with it and the importance you had mentioned this Scott something that you had said was building out the deterministic capabilities. And I think especially in industrial, it's like you can't just have that chat GPT moment go across the companies that we're building for. It just doesn't work. We even see, you know, Copilot's great in certain ways, but it's not what you're going to be building your systems on. It's not something you're going to be relying on in order to make uh whether it's, you know, financial supply chain decision or whether it's engineering decisions. Um it's that the moment needs to go beyond a moment and it needs to go very deeply into confidence and trust and um like you said determinism and not just prediction.

Speaker

Yeah, and I I imagine we're we're we're both working on workloads that are bearing on contracts and non-trivial liability, right? So, if you're making representations about the capability of an industrial system or in our case like accepting orders or asking to make an order, etc. like you can't make a mistake.

Speaker

Yeah, you can't make a mistake with scoping projects, selling projects, supporting it, saying what goes wrong. And even what's interesting to see is like this future of, you know, where AI is going to go in sort of in terms of even regulatory issues related to it towards AI like to AI when it comes to industrial space and auditability of the answers that AI is giving, which is like something we're deeply thinking about now is when you're building out your the AI in order to give an answer, how do you trace back that response in order to audit it eventually in order to say this is why it gave this suggestion of this configuration in our case or this is why it gave this suggestion to the customer to fix it in this specific way. And when you come into this level of regular regulations in AI and industry, you need to make sure you can find an audit path and you can go back and say and in our case we're building a knowledge graph for that. One of the one of the many reasons is that you can go back and say AI the LLM or AI went to this decision not just because it decided that it was the right interesting and it matched all the specs, but this actually is worth and this is why it made that decision to actually kind of think about it and traverse back across those decisions. So I think that's where a lot of people need to think about AI in a different way when they're building any regulated industries like ours.

Speaker

Mhm. That's really great observation and it reminds me like last year um I was at the Capgemini's Engineering Horizons conference and there was this great talk by professor from Oxford that was working on robotics and she her thing was like we have black boxes, right? So an airplane, no matter what country it's made in, who's flying it, if it crashes, we can figure it out because we find the black box and we can no matter what we can figure it out. Where's the black box for robots? And where's the black box for AI influencing engineering decisions, right? We don't have any standard for that. It's all just up So anyway, I thought it was a super interesting thing and she was arguing for like an e-black box standard for robots because someday a hospital robot is going to kill someone and no we're not going to figure out if it's parameter number 11,752 or was it some idiot human that just clicked yes and didn't read what the actual was happening? I mean we're going to need that, right? So yeah, it's a very valid point.

Speaker

I've seen a new wave of startups actually building insurance for AI, which was an interesting Yeah, I don't know much about it, but it's just something I saw recently, which is like

Speaker

Yeah, I I I hope they're very confident in their actuarial math as they they they head down that path.

Speaker

Yeah.

Speaker

But talk a little bit about software development because uh that's one of the In fact, it was interesting earlier today. I I always ask this question later about the open AI moment for engineering or for supply chain. And it's true someone mentioned we had the open AI moment for programming, right? That was this January. Cloud co- co- co- Well, it was anti-gravity superseded like immediately by cloud code, which is just So, how how has that changed you you guys are both founders and you have a programmers working for you? How has that changed fundamentally how you look at managing a company, managing programmers, managing meetings of programmers? I mean, has the whole waterfall versus agile gone out the window because now it's all agents do setting up their own scrum meetings? I mean, just tell talk to me about that a little bit.

Speaker

Scott, you want to go first?

Speaker

you. Sure, yeah. I mean, for For us, it is it's been wonderfully helpful in allowing allowing us to, you know, change our team structure from kind of one big scrum team of like eight people, um because we're relatively cash constrained, into breaking it so that it's, you know, one PM type of me, uh somebody else, and two two engineers that are relatively full stack working with cursor and cloud, I think Opus 4.7. Um and then kind of one floating designer that's managing the design system and can come in and just do front-end cleanup across the whole system without needing to bug anybody so that the engineers can focus on the hard stuff, which has made it way faster to deliver. And it's also changed from a product management perspective the amount of saying no I have to do, which has been nice cuz that was most of the job for the last couple of years. But now I can be more flexible. I can let the system handle edge cases for more people, which is, you know, bringing my TAM into a wider space, which is very nice.

Speaker

Very cool.

Speaker

think I think for us it's interesting cuz I'm I'm wearing hats. One is, you know, the the CEO and I'm not I'm not the CTO. I've got an incredible co-founder that is managing our our dev team. So, I'll speak both on behalf of me, Faye, and what I'm now able to do as a non-engineer CEO, but also product-centric, and what my co-founder has been able to do. So, first of all, like you had said, the like, you know, we are able to do a lot more with our young team, but we're also able to now trust juniors in a way that was interesting, at least for front end. That moment that you were saying with flood code and and really the ability to actually have the engineering moment with code, we definitely have felt it. Um, for the first time, our very, very classical back-end engineer actually was willing to try um, AI tools. Initially, he was like, "Hell no, am I letting this near my code base?" And, you know, about a year and a half ago. And now, there's actually a little bit more of a flexibility of seeing that the capabilities, at least for back end, as well we always were very bullish on on front end um, using these tools to be able to build out. Um, um, really the the product interface. Um, and we're also able to do a lot more with a lot less, like you had said, Scott. It's like we are actually able to do a lot more um, coming out of uh, of of even prototypes very quickly. And that kind of leads me to where I'm loving it as a non-engineer, not building out, you know, our our products directly, but my ability to actually take these dreams that I have after customer conversations or after seeing new things on the market, being able to very rapidly create a prototype that I can then speak to my dev team, that it's not just a written spec that I'm doing, but actually show them my vision. Um, which I think has changed a lot of the ways that um, um, there's collaboration between product teams, dev teams, um, um, customer-facing teams, and seeing a lot of opportunities. Um, and and I think that's really where I'm finding a lot of love. Sometimes, my my my co-founder is just like overwhelmed with like, Faye has another another prototype she wants to to show us another idea she wants to show another thing she wants to do so it's always trying to also figure out how do I can how I can hone in um and and kind of not not let that overtake everything. Um but overall um it's given us the ability to do so much more. Um and then even on my CEO side um you know there were so many things that we were doing and trying to figure out which CRM we wanted to use we were going to continue using notion we were having all our Fireflies transcripts and then we were going in and trying to figure out how we were going to follow up the leads and my co-founder he just built us we call it Barton Flow which is our internal system that is able to both take our scrum board and all of the product road map be able to map it against the the bug reports or the feature requests of our customer conversations be able to loop that back in into my reengagement with them afterwards. Um and I we now have an internal system that runs all of those things for us with us and collaborating you know collaborating on finally merging those worlds together which has in the past been so isolated. I think that merging of worlds is really where the value comes.

Speaker

Wow.

Speaker

Yeah it's cool admirable working on the internal tools. We haven't done much of that ourselves kind of just you know used it to wire what exists together a little better and to run some automations but we haven't tried building any like scratch CRM for ourselves just yet.

Speaker

Oh yeah.

Speaker

I've heard of that. I think that's part of the whole SaaS apocalypse thing right at the fact that so easy to do. It makes me wonder too I remember I mean I haven't done a lot of programming but I've been a lot around a lot of departments doing programming and I remember there was quite a lot of criticism of Agile because it became just a tool for managers to beat the out of the heads of the the programmers. And I'm thinking this AI stuff is probably because then there was this craftsmanship movement right which was trying to get back to the original ideas of enablement as Agile as enablement for programmers. I wonder if AI has really brought that back. It is it become more of a fun thing and it's less of manage a tool, you know, is it easier to is it easier life for the programmer because it doesn't necessarily have a billion scrum meetings and is it whole day is it just cadence by all this crap he has to give back to management to prove he's actually doing his his or her job, right? Have you Have you seen anything changing there?

Speaker

I don't know if I can speak to change in the industry. I can speak to change in my world, which is we were pretty strict agile practitioners for a little long time. Now it's well, partially because of the the change of pace and what's possible. The like story points were made up nonsense before this change and they've just become even more unknowable, whatever the heck vibe estimation of a person day you want it to be. So, they're kind of pointless. So, for us it's we we take chunks of the road map, we decide how we're going to handle them incrementally and we meet twice a week for half hour to see if there's any like walls we need to run through and we develop and that's been working very nice. Now, our our teams are very small, so it might only work on a small scale, I don't know, but it's been refreshing and a lot less bureaucratic.

Speaker

Thanks.

Speaker

And you know, from our side I think we've kind of discovered or are still trying sometimes to discover that sweet spot between being able to just go out and build versus also knowing that we need to somehow sometimes still spec things out, especially ones that have a lot of complex back end work or a lot of APIs that need to be built or a lot of things that need to be done that demand more than one player. Um and so there's always finding that that that line of like let them go build, but also let's make sure that they're building something that is on both sides compatible with each other. And so I think that's where we go back to kind of the basics of making sure I don't believe that now a bunch of agents are going to be able to only build companies. I still believe like, you know, Scott you had said in the beginning is like you kind of are a manager, you know, the the the the lead still needs to manage a lot of these decisions and still go ahead and be involved in those daily um decisions. So, I would say there's a lot of freedom now, but we can't let that freedom overtake smart sensibility in building the right thing for the right customer in the right process.

Speaker

Yeah, and it's especially important given AI's or especially agents kind of lack of skepticism, the the tendency to tell you your your ideas are really good. Like you you you need to challenge yourself and have someone challenging you or you're just going to run off down a tangent, burn a ton of tokens, and not get any get anywhere.

Speaker

Yeah.

Speaker

But that that that reminds me of uh I've been saying for a while I think we need some Maimonides agents. And I was just in uh Granada, so I actually saw the statue of Maimonides uh that's there. Um like an agent that basically just expresses doubt. Like, I'm not really sure. Are you really sure because we need agents like that, not just the one that your idea's the best idea I've ever heard in my entire life kind of thing. My life has only been the last 2 seconds since you started a prompt, I think. Um So, in terms of the the the stack that you guys are building respectively, um where where does AI sit? Is it part of the user experience? Is it a fundamental model? Is it the entire DNA of the stack? What Where does it actually sit inside of Bard and inside of Omni? As a technology?

Speaker

Within Bard it sits in almost the entire product. We are an AI-first product. Um the way that we're building our database is a knowledge graph based on, you know, very, very complex embeddings. Um we're running AI from the start to the finish. That being said, not all of it is generative AI. So, there's a lot of things that are going into it that are classic machine learning elements of AI, breaking things down a little bit more on that side, and not necessarily always is it, again, an LLM coming in and giving an answer. Um and I think that's something that's important to think about when you're thinking about AI is that like people think of it just as a chatbot and giving answers and making, you you an LLM going in, but there's a lot of things related to AI that are involved in products um that are not necessarily generative. Um so, that's number one. We definitely are an AI first product. We are building with that in mind and with that focus. Um but then again, we're building to still solve a very non-AI use case, right? We're talking about classic engineering um support tickets and queries and pre-sales scoping of large industrial automation projects. We're using AI as an enabler um to help solve human problems that are currently now bottle you know, bogging down the processes that humans need to take. So, our goal is not to necessarily replace those folks with AI agents, but to be able to enable them to do their human job a lot better. Um but AI is definitely core to how we're building, how we're thinking about it, how we're approaching it, and how we're seeing um opportunity.

Speaker

Thanks Fay. How about it? I know Scott, you said you're not going to have that

Speaker

same approach. And yet, there's still AI in there.

Speaker

I I look at our core system as the jungle gym that companies AIs can play on in their supply chain. And then we we have some native deployments of like ML algorithms in our the way once you've got a knowledge graph of how your supply chain is configured behind a bomb, like the actual line items that gets you this thing, which is never a clean map to the bill of materials on a product. Um like that the thing that goes out and tries to solve those dependencies for you when you cut an order or you have demand is not an LLM based system, but it is algorithmic. Um we've got some ML workloads especially in data onboarding and normalization cuz everybody's data is a different kind of mess and we can't have them going in there and like dragging around and coding in how all these columns match up. We just try and figure that out for them. And then the the current native LLM use cases are kind of monitor the the email for supply chain signal gateway which can go in your email and find all the supply chain signal and get it in the platform and draft for change orders, orders, sales orders, invoices, your vendor saying they're going to be late, your vendor asking if they're allowed to be early, that kind of thing. Um cuz not everybody's going to play ball right away, right? Like I would love everybody to just hook in their systems and let everything run lights out, but it's going to take time. So that's a a stopgap measure for us.

Speaker

Yeah.

Speaker

Okay.

Speaker

The human in the loop in this world is something that is really important to keep in mind. And building things that are task-based workflows that are classic that humans are doing, that they know that they can, again, not to come up with this audit conversation, but that they can actually measure and say, "This is how I would do it. These are the steps. This is where it can go wrong." and be able to come in in different places within that. Um so, you know, again, our with Warden, the first step for us is ingesting that data, like you said, normalizing it, structuring it into the knowledge graph, so it really truly understands how are these products working, what are their really the the the specs about them, what are the manuals speaking about, how do we engage with those things, but then so much far beyond that is then understanding what are the humans doing and what have they done in the past, how have they solved for my how have the humans solved tickets in the past, how have the humans built industrial projects and spec'd those out and built those bombs in the past in order to then have those be really the confidence level, you know, barometer of saying, "This is what the AI should be thinking about as well." Um so, I think that's a really important thing is to not just um go ahead and say AI knows everything, um but to say, "Humans have done it before. Let's really make sure that we're learning from that."

Speaker

Yeah.

Speaker

And there's there's a layer on top of technical doability and safety, which we've kind of let run into the hard way of will the humans accept it, especially the accounting persona.

Speaker

Yeah.

Speaker

Built for accountants before, but they're very control oriented. And if If in and say, "I'm just going to automate every interaction between your your GL in the outside world and turn it on." They freak out and hit the stop button. So, we had to go back and turn all the automation off and let them turn them on one at a time as they start to trust it, which was a lot more work than just automating everything in the first place. But, these are the things we learn.

Speaker

I think there's a lot of things that a lot of companies that have like the first thing they did with AI was build an AI chatbot onto their website that expected it to somehow magically answer all of their customer questions, and they realized it was just going ahead and hallucinating all those answers, and they're like, "Oh, wait, maybe this is not the way to go. Maybe this is not the smartest thing that we need to do." Um and I think that's kind of right now we're probably going to see uh and and we're we're seeing it already is there was like a distrust of AI because no one knew anything about it. Then there was a real big excitement of let's get it into everything, and then there was also now then slowly a uh uh a hesitancy because they know the mistakes it could make, and now I feel like we're in this space again of, "Okay, we know we need this. We know we want to engage. We know what makes our lives better, but prove to me it works." Um and I think we're seeing that cycle, which I think is an interesting thing. Um don't like it, love it, interesting, now let's actually build it properly.

Speaker

Yes, and make sure my dealership chatbot doesn't agree to sell a guy a truck for a dollar. That's happened a time or two.

Speaker

And then you see also what when I ask you guys about the changing economic models though, right? Because this stuff isn't cheap. I mean, it whether you're using cloud cover development or you're using an LLM that your customer's using. How how have you guys changed your the economics or the I mean, have you thought of you you know, creating your own LLM that understands supply chain or AI or you know, are you allowing customers to bring their own uh Anthropic API key? I mean, how how are you dealing with that? Because otherwise, the operating costs are going to be just astronomical, right? If you're paying the bill for Amazon and AWS and Anthropic and ChatGPT, I mean, that what's what's left, right? To pay salaries. I just I was just wondering how do you guys deal with that cuz that's sort of a new economics that's coming out in the last year and it I'm not sure any we have a consistent answer yet.

Speaker

I I I don't think I I I know I don't have a consistent answer yet. I'm still discovering it, still working at it, still trying to understand, you know, everyone's talking about outcome-based or usage-based pricing for customers. In the end of the day, industrials still understand user-based pricing. They still want something that they can actually plan on their, you know, understand what their spend looks like. So, it's an interesting thing I think as an early stage startup, we're still navigating and we're still exploring and we're still seeing things. Um we're still seeing what other folks are doing in the space. So, to be honest, I don't have um don't I don't know yet um and we're still learning.

Speaker

But for now, the people are using your API keys or they can bring their own uh API key?

Speaker

For now, we are taking care of it. We're still in that stage of doing it on our own in the beginning. Uh we are connecting in their we're connecting it their data is connected into our system. We're actually um able to maintain hosting in their cloud, but sometimes we're moving it into a managed cloud. Um for now, we are we are swallowing those costs. Um

Speaker

Okay.

Speaker

We'll see what the future holds.

Speaker

For us, it's a bit of both. Uh so, for our for the data onboarding ML stuff, the dependency solving, we eat that. For the email the uh email gateway, that's pay as you go on our LLM. And then if they want to scope in their own API key and let it do things, we're working on that to basically give them the ability to generate the API on top of our API. That means that it's scoped very pedantically down to what they want it to do using plain language and that it is rate limited, so it's not going to light a gazillion tokens on fire if they let it play with only. Um and then on the development side of the fence in terms of managing costs, we haven't done too bad space in on way we're working, which is very incremental, trying not to get Claude to go run off for a day and burn a million tokens, and then throw that code in the trash. So, like, want trying to know we're going to actually want to use something before we go get it built, beyond the looks like it works pretty prototype stage, um, has kept our internal burn pretty reasonable.

Speaker

Yeah, I think it's interesting I'm going to add something to that. In in in general, like, when we're building out Barton, and the way we're building it is we understand that there is a future that folks are going to want to be building out their own agents and using infrastructure. We're building it out eventually to be the decision infrastructure for industrial automation decisions when it comes to selling and scoping those products. Is really how we're thinking about it in the long-term opportunity. And we know in order for that to happen, we need to be building the infrastructure now for them to build their own agents on top of that, connect their own models, connect their own things. And so, we definitely are building with that in mind. Currently right now, we're not there. Um, but that is something that we're planning for and thinking about. When it comes to, you know, our our internal team using it, it's so funny to like see, and you can tell that we're an early-stage startup, um, because you hear all these huge massive companies that brag about the millions and millions of tokens, and they're like, "You're not a real developer unless you're burning through X budget of tokens." And I'm just like, "Wow, you know, like

Speaker

are hilarious. Like

Speaker

Yeah. Yeah. Leaderboards, exactly. And it's like, you know, I I I want a leaderboard of building out something right.

Speaker

What's also interesting that we're we're we're coming into a space where a rate limit on an API is a feature and not an annoyance. Mhm. Where like you say like giving your client the ability to let your scope of their API not light it 10 grand on fire in an afternoon. Like as a safety guardrail. Is a you're welcome, not a you're annoyed, cuz you're trying to get something to work with the integration.

Speaker

That's crazy. I never thought of that. Um So, we've been like 4 years into this AI thing, right? So, 2022 to now. Um and we've seen just an absolutely mind-boggling amount of change, right? Um and yet I'm not sure I mean we and we've seen an open AI moment in um in in manufac- in I saw many open AI Yeah, I saw an open AI moment in manufacturing when I saw at It's Prove It that they could deploy an entire factory of of sensors in 3 minutes. I was just like a totally open AI kind of moment, right?

Speaker

That's that's Jeff's system, right?

Speaker

Jeff and and Litmus and and Inductive also. But and then I we definitely saw an open AI moment programming light in January of this year. Um we haven't really seen it for engineering, have we? Or even for supply chain. I mean, do you think we'll actually be a moment or it'll be a series of moments or that these things just move too slow and the data's so complicated that we may never see I mean, what what do you think about that? Do you think we're going to have an open AI moment for what we do?

Speaker

I think it's going to take time. Like in the in the space of big enterprise supply chain software, these are deployments that traditionally take years. On like in a good case, like trying to deploy SAP supply chain is a 24-month at minimum frame off rebuild of your business. Um and in that world, like you can deploy quick wins at the edges, but you really can't do it at the core. Um but I think once once some of those deployments actually get done, you'll see some companies just lapping other people in ways that you've never seen before.

Speaker

Hm.

Speaker

I think when you're looking at it, I would segment it. Saying an AI moment in supply chain is touching a lot of players in it. You're talking about procurement, you're talking about the sellers, you're talking about the folks in the middle, you're talking about distributors, you're talking about the suppliers. Like we it's it's hard to say that one would be a cookies of moment because there's always an easier entrant into saying we are going to now make a a large shift into the way these people work. Um and I think we're going to see it segmented. So, we're going to see AI moments in in our case, let's say, the sales and support teams that are really focusing on the engineering questions and being able to optimize those folks. And we're going to see a lot more openness. And then we're going to also see things separately on the procurement side. And we're going to see separately things when it comes to invoicing or etc. I also think that those moments are going to be segmented by mindsets inside like first we're going to have the moment of I recognize AI will change my business fundamentally. That's a moment. Then I recognize that I can implement this into my business. That's a moment. Then we'll say, I now recognize that this is actually implemented and is actually making a change. And then we'll have the moment of I cannot do my business without this. And I think each of those are going to be moments. And I think if we're looking at it waiting for a collective moment, then we're not looking at it um uh correctly because I think we are all looking at this ChatGPT moment in November 2022 as like a moment, but you speak to anyone building in the space, they're like that has been worked on in small moments for for years. And if you listen to the podcasts about, you know, the folks that are Google DeepMind and how they were thinking about it and how they were approaching it, we were talking about a lot of very very big moments separately um that you know don't even think about as the moment. Um so, I think we need that way.

Speaker

But my point is though is that we didn't think about AI the same way after November, right? I mean, that was There was a way we worked before and then there was the open eye moment and it's the way the year the way afterwards. I mean, nothing is the same, right? I mean, the way you think about your user interface is not the same. You expect to be able to talk to your app. You don't want to click anymore, right?

Speaker

Yeah, that's

Speaker

My my my supposition, though I don't have any proof, is that especially in the procurement world, there's a lot of dark stack AI going on where line purchasers, buyers, reps are just using it and there's a fair amount of risk there post-mortems on why things got agreed to that were impossible and that's why lawsuits are happening. We'll start to point people towards trying to deploy it properly or in more risk-averse companies try to step on it for a little while. Um, but like this the the way the whole supply chain is oriented around the buyer persona within businesses is basically they are the interact like the the real integration point between businesses buyers and reps where they kick out emails saying, "Hey, where's my stuff?" They get a response. They pack that into a spreadsheet. That spreadsheet gets fed into SAP once a day. And that's how the thing works. Um, and that that call and response and the the knowledge graph of who makes what and how it feeds into the dependencies of actually getting something built needs to come out of those people's heads somehow and like shameless plug that's that's what we're trying to do. And you know, get into some kind of many-to-many system that can actually make things work in between companies that aren't just an entire profession which we don't want to fire those people. We want to make them a lot more effective, but they're also going to try and do it on their own dark stack and do a little bit more chilling before it's going to end up in some kind of proper system. It's going to take some time.

Speaker

So So, speaking of

Speaker

Go ahead.

Speaker

No, I was just going to say I think there's a lot that we can learn building in industrial in industrials and building in manufacturing and supply chain from the way that these moments have happened in um legal spaces that these moments have happened when it comes to you know, hospitals and medical spaces and the way that the vertical I AI has actually worked in other spaces and the the the pathway to make it work and have those moments. I think I spent a lot of time looking at the way um some companies have been doing it, you know, in the in in the medical space AI doc or a doc um in the legal tech space, Legora or Harvey, or, you know, when you're looking at the way that they've been building things, there's a lot of lessons to be learned. Um, and I think you're seeing regulated industries, um, adopt AI. And whether it's, you know, governmental, and I'm going to, you know, another Israeli startup, Darwin, right? They're building into government, etc. There's a lot of ways to think about building well unsuccessfully in our areas, and it all boils down to trust. Um, and I think that's the most important.

Speaker

Um, the and the demographics demographics of the people that watch this podcast, there's actually quite a lot of younger entry-level people. And I'm wondering, um, well, I'm I'm imagining that they probably have a lot of anxiety because, you know, the all the dire predictions of AI destroying all the entry-level jobs. Um, as founders and as people hiring younger people, what kinds of things are you looking for in those profiles that you're like, "No, definitely this person is better than an AI." Like, what should they be focusing on so they are employable in this AI era?

Speaker

I don't want someone that's better than an AI. I want someone that knows how to use AI very confidently. I want someone that will come to me in the interview and show me what they've coded. I want someone that has shown me if they're looking for a marketing or sales role, the way that they've built or managed agents in order to do the dirty work that they don't want to do and that they could be a manager of those agents. Um, and I think that's the opportunity for younger folks is don't be afraid of it because it will only take your job if you don't know how to manage it. But if you become the master of the AI solutions, of the AI tools, and you come to your interview saying, "Here's my stack." Instead of saying, "I know Microsoft PowerPoint and I know, you know, Word and I know Excel." But you say, "I actually know GitHub, I you know, Git, I know how to build on Vercel, I know how to use Cloud Code even to build something." Um, I even as basic as I've played around with N8N to automate certain things, and here's my workflow, then I'm like, okay, I can trust this person to to to be part of our fast-paced, forward-thinking, AI-first company. And I think that's what I would suggest. YouTube, watch YouTube videos and practice and come with that.

Speaker

Yeah, I think building on top of that, the things that are important to me moving forward is less technical prowess, more people skills, product sense and initiative, if that makes sense. Where you Not everybody that's on my technical team needs to be like a deep-dyed in the wool, close to the metal nerd. But they do need to have people skills, cuz ultimately the economy, like it or not, is about people, all right? And all of the business you're going to do, at least in the near term, is going to be people convincing people that your thing is good. The personality becomes more important than the ability to take initiative and not be told what to do. Is that a good answer?

Speaker

Those are both really original answers, so that's great. Cuz I usually get, you know, the fundamentals of physics and and things like nerdy answers. Um So, um as we move towards the the end of the interview, I wanted to ask about um Well, I want to discuss about digital transformation. Um I I look at digital maturity of companies on a scale of like one to five simplistically. One being like, they're still using Excel for almost everything and communicating by email as opposed to we have these fully agentic, adaptive, autonomous digital twins. And I don't think anybody's at five No, I on that scale. I think most people are between one and you know, one and two. But I'm wondering in your experience talking to your customers, there Do you find that they're between one and three, between one and two, or are there some of it four? I mean, how would you how would you look at that?

Speaker

The So, in enterprise, the the built, paid for stack will be a two and a half to a three. The how the sausage actually gets made will be a one. In the mid-market, you're kind of closer to the the two world where people are very curious and I've I've found this this change in consumer trend at least in the mid-market where people have just suddenly woken up to automating stuff they could have automated 15 years ago because they perceive the friction to be lower and they perceive anything that's in any way automating anything to be AI and they're just they're very curious and they're very ready to actually make a change whereas I could get to a yes five years ago but getting them to actually do some change management was the hard part. And that's gotten better.

Speaker

Yeah. I I would say quite similar and I would I would just add one of this an interesting thing I saw recently was a Microsoft report about um and and Industry Analyst shared it and kind of re-re-republished it about looking at the percentage of um agents being used individually in companies and surprisingly manufacturing had the highest share of agents being used in individual companies. Now, only 10% of manufacturers were using agents but those that were were using it at a a way higher percentage than even regular software companies were using it. Meaning once you get in to an industrial, once you get into a manufacturer and they're open to AI, their usage of it and their implementation of it is quite heavy. And what was even more interesting is when you were looking at the table and you were looking at the numbers, they basically were showing showing not only the amount of agents being engaged but also how many prompts were being plugged in and the prompts that were being plugged in and engaged with in this industrial or in manufacturing space was actually quite low. What we're seeing then is you're having a lot of agents that are not necessarily chat interfaces but these are actual practical daily workflows that are being used or tools that are agentic tools that are being used and thought about in the industrial space. And so when I'm thinking about that in answer to your question is I'm think I'm I'm really seeing and it's always this interesting thing cuz people are talking about, you know, industrials or manufacturing or supply chain being so old school. And then you look at these companies and they are so freaking incredibly brilliant. They're building robots. They're building the most complex industrial products often times and yet still their back office or their support system or their sales teams are still being bogged down by this not necessarily as innovative as the actual manufacturing floor is. And I think that's really where the opportunity is and I see like Scott, that's where you're building and that's where we're build we're building in and I think that's our goal is to get them from the reliance on Excel to the understanding that Excel is great, but let's automate some of those processes so you don't always need to just go into that Excel. Um and I think there's huge opportunity there.

Speaker

So So my premise also one of the premises on this webcast is that um if companies do wish to move that needle to the right uh towards full you know, fully automated digital twins, then using an Omnia or using a pardon is a much better bet than waiting for SAP, Oracle, DS, Siemens, PTC to help them with it. So I'm wondering like have when you guys have have shown people this way, has it been an epiphany? Have people said, "Holy cow, they mean if I broke the barriers between engineering manufacturing, I could actually go faster and get more profitability and more you know, higher quality and etc." Have you seen that a sort of epiphany or a sort of a a realization because they're using the your software?

Speaker

I would say that it is not an epiphany. It is a really pleasant, oh my god, kind of moment of like I need this, but it's not like a oh my, wow, my world is changed because I think that oh my, wow, my world is changed is kind of a little BS sometimes and it's like really nice for a demo, but when you come in and you actually start doing these processes, our goal is more like, "Wow, now Todd or Joe or Jeff or Rick could retire without being super stressed about it." Because they know that the juniors that they're bringing on to their team could actually scope that very complex project well, and that their customers will actually be supported without it having to escalate too far and wait for, you know, Japan to wake up because they're the ones that know all the details of that that's of that specific AMR. Um and so, I think it's more about like a sigh of relief that I'm looking for rather than a wow epiphany. I just want people to sigh with relief that they've got extra hands and extra capabilities that they never had before.

Speaker

Similar for us also, um I I'd say it's Bigger companies are liking us in that they can test us in small scale much easier than they can test a solution from a Siemens or a PTC or an SAP, or they can deploy us into either like a small business unit or into their prototype shop or something like that and solve a lot of problems there, especially in worlds where the big systems just don't like dealing with like pre-part number, pre-release production things like that, and then they can roll us out more broadly once they've proven it in that faster-paced environment.

Speaker

Yeah, and I I completely agree with that. I think one of the things that a lot of enterprises um do but I believe shouldn't do is they're attempting to roll out AI from the innovation level top-down kind of approach and say, "We want the same thing and it'll run across everything, the same system to run across everything." And then when you're looking at the way the market works, and you're looking even in the industrial market, the folks that are building the sensors in theory have the smart engineers that they also could be building the pneumatic valves, right? They also could be building the pumps, they also could be building. But, their focus has been and will be the sensors or the vision sensors. And I think that's something to keep in mind is that the folks that are trying to, you know, now take one AI solution and assume it's going to work just as well for accounting as it will to go ahead and build your bomb as it will to go ahead and answer very complex engineering questions, um, is is naive, I think. Um, and I think let the ability to go in and say we're going to solve this use case, we're going to start with a small, very focused use case, we're going to roll it out incrementally across your team, and we're going to be the experts. We are the experts in this specific space, in this very, um, core part of your business, but a element of your business, I think, is the right way for folks to do it. Um, and I think that's why I have the feeling we're going to see mid-market companies actually excel in the, um, adoption of AI solutions and because they have that ability to kind of take incremental steps and don't need to wait for a top-down mandate to roll out AI across all divisions.

Speaker

And you can also kind of see it it from the big tech side, too, of the the tendency for the big providers to try and just cram an LLM into everything and just force it on people. And then that tends to just create a bunch of pushback and little usage and a lot of cash being burned instead of trying to find like real wedges and wins and roll them out one at a time. Uh, which would be better for everybody.

Speaker

Yeah, I think so, too. Um, well, that's been great. I really appreciate, uh, the, uh, candid answers and and the the discussion. This has been fantastic. Um, I'm wondering, uh, where can we see you guys this year? Um, Scott, I got to meet you down in Threaded in Miami, which was awesome. Uh, and then Fay, you said you were about to head to New York. So, where where could people meet, um, Fay and Scott, uh, in the rest what's left of the this year?

Speaker

Yeah, I'll be next week in New York. I will be following that in Minnesota, following that in Detroit, and the best place, you know, right after that would be to meet me at Automate. I'm going to be at Automate in Chicago for uh the full week, and I would love for folks to connect with me. I'd love to meet you in person. Um and I'm an an an avid engager on LinkedIn. So, you can find me Faye Goldstein um uh LinkedIn and uh at there.

Speaker

And maybe in Tel Aviv, right?

Speaker

Yeah.

Speaker

And Tel Aviv, yeah. Sometimes but when I'm here.

Speaker

Yeah.

Speaker

Sounds good.

Speaker

Oh, no problem. Uh we're still kind of working out our conference schedule for the the back half of the year, but um like as as Faye is, I'm very active on LinkedIn. You can get me at scott@automate.com if you want to just send me an email. I'm happy to talk to just about anybody.

Speaker

Awesome. Well, I just I think I forgot to call out our sponsor AWS. So, there'll be a link in the comments to download the white paper or I think there's actually a webcast that they're promoting now about Agentic AI and uh leveraging AWS Marketplace for that. Um I want to again thank uh Faye and Scott for their time today. It's been really interesting. Uh a lot of fun. A lot of very unique answers to these questions today. I really appreciate that. And um I hope to hope to talk to you guys soon. It's been fun.

Speaker

Likewise, good catching up. Nice to meet you, Faye.

Speaker

Nice to meet you, too.

Speaker

Thanks, everybody.

Speaker

See you.