AGNT Podcast Ep. 13 with Gemma Allen & Raphaëlle d'Ornano

Welcome to AGNT, the podcast where enterprise tech meets the agentic era. I'm Gemma Allen, joined by my co-host Raphael Dornano, broadcasting from the New York Stock Exchange. And every episode, we unpack how intelligent systems are reshaping companies, markets, and the way real work gets done. From Fortune 500 boardrooms to breakout upstarts, we're digging into the strategies, technologies, and people defining the next chapter of AI. Let's get into it. Raphael, welcome back. It has been what feels like a century since you were here because so much has happened. Had >> a good summer. >> Yes, so much has. This summer has been crazy. >> Well, the markets have really been crazy and I know we're all waiting for Jensen to take the mic in a while um and see what happens with that video. I think a lot is kind of resting on what sort of earnings they have. But let's talk broadly for a second in terms of why this industry is so crazy and I want to get into a topic which I know is close to your heart and that is the world of intelligence. Intelligence happening in abundance and in that model who truly owns the receipts. >> So you have been writing a lot about this apocalypse. I know you have been somebody who's been very focused on this for quite a while and lately the viewpoint is that okay it's not necessarily which model is better. We know models are getting better. We know they're getting cheaper. They're becoming mass available. It's more the question of who owns the relationship, the opportunity after Claude gives you that result. Right. talk me through what you're seeing, what you're thinking, where you think this very nuanced industry is headed. >> Right. I think look we're almost we're a bit more than 6 months into the SAS apocalypse which happened in the beginning of the year and no one has forgotten right during this time uh intelligence has as you said has become abundant with and intelligence under all its forms open source closed source is now like uh an evidence and I think that the way that the software industry has evolved and what it has become is at a very interesting inflection point. What I mean by that is that I really saw that software was in a huge trouble at the beginning of the year pre-sass apocalypse and I had shared those views with you because we were seeing these AI labs and I when I say these AI labs entropic and openai of course really trying to build a moat by going upstream from intelligence into becoming this operating system for their clients and that was their only way of and is still their only way of escaping a form of commodatization that is now very clear. And so software at that time was in a position where those companies knew they were going to be disrupted quicker than they saw. And what we have seen is kind of like an interesting inflection by which the system of record which we thought was going to be a database that was going to be just queried by intelligence is mutating into the most important thing by which the system of record is actually what allows you to build from a very defensible position to use the intelligence that is now abundant by these models and now to own the front end. It's what we've seen with Slack code. M >> it's what we're seeing with workday and senna completely rewiring the interface and now we're in this new architecture by which some software companies have the proximity to the human intent are actually capturing intents are rewiring rewiring what needs to be done to whatever form of intelligence exists cloud open etc and are going and rewriting to the database that they own because they have the system of record so they're building this new execution loop which they have the entry point they have the system of record and I'm not saying all software I'm saying the companies that have the systems of records that are building this new front end which is exactly like claude and chpt being the front end the UI of uh of AI we're not there yet I'm not saying that this is solved I'm not saying that software is completely back I'm saying we're at a very interesting point by which these incumbents have moved very fast with a position that is very defensible. If they're able to evolve their business model and to capture the economics of this new architectural reality, we may be assisting to something that is very interesting over the next months. So what does capturing the economics of this architectural reality look like though, right? Let's take Slack code as an example. Okay, a lot of folks I mean it's obviously owned by Salesforce. Salesforce's had a shaky year across the board as have many of its you know peers >> right >> this idea that you know usage based licensing seat based licensing is kind of in somewhat of a paradox Salesforce the company has agent force you know it has Tableau great product too but again you know it's like one potential solution for one part of the business which is very much the communication side of that business right with Microsoft teams or whoever Microsoft Teams has co-pilot right like if we were to really, you know, kind of like boil this down. Like give me an example of who you think is defensible, like who you think is doing this well. like where you see an opportunity for a stock to bounce back perhaps because again you know there has been somewhat of a laggered response to the cannibalization that has been you know presented to us through the likes of anthropic and I wouldn't really say open to the same sent but it's certainly there right >> look I think you first I think you're raising a very important point on insisting on the economics part and I would start to say by by underlying that most of these software companies are still reasoning in terms of seats. At some point and when I say at some point I mean now we need to acknowledge that this is not the right unit anymore. I think this is point number one. We have acknowledged that the UI is not what it used to be and the UI has changed. So that's what's being done with Slack code slack sorry. This is what is being done with Microsoft now using uh like uh copilot as the interface which I think is is interesting. So this is okay. Now from a seed perspective if you believe in agents delivering tremendous value and capturing work hence the 30 trillion t that uh raised many questions from entropics IPO I don't know if you heard this yesterday like SpaceX 20 something entropics already. So if we believe in that, we believe that agents are going to be completing work that was previously completed by humans. So if you follow the logic, the more this gets successful, the less you have seats. So companies that are building per seats are naturally on a declining path. That's normal. So software companies today, the way that they're attaching consumption, they're still attaching consumption to seed revenue. So when you see oh this company has a lot of consumption this company has uh great growth in agents and you dig into this you can see that it's still linked to the seat. So I think the next act that we need to put forward is that we need to see a decoupling from effective consumption which will prove that aentic AI is working big time from seats and accept not to have seats go down but to rebase rebates I like this term the seed revenue naturally. So I think this is where everyone is kind of stuck right now. I'll give you an example. If you look at Senna and workday, Senna is really interesting and I think that maybe one of the reasons for a potential take private if that goes through with Silverlake is that workday would effectively be a very different company if Senna effectively became that new interface aetic AI compliant I would say. But when you look at the cross cell at the number of users that are using workday, the ones that are effectively using Senna, that rate is actually pretty low from my understanding. So we're not there yet. We're seeing a lot of burgeoning signs, but this is a race that is up. What I do not like is companies that are showing a ton of AI ARR because that's what's the most fashionable right now, but that have not changed their architecture, meaning they have not captured this user intent, the proximity to user intent. They're not uh they don't have a system of record that is a defensible entry point. So, they're still at the core of the SAS apocalypse. And this has only gotten worse. Just to be clear like the SAS apocalypse >> with the threat that existed at the beginning of the year is even bigger right now. So some companies are in the course of transformation. I mean workday, Microsoft, Salesforce which is doing amazing work with the announcements that we had from headless 360, Slack. So this is all going in the right direction. Okay. And at the same time, Entropic is moving big time. And Entropic is now explaining that well, their TAM is going to be the whole work, which I do find to be a big number, but which between you and I doesn't shock me. >> Well, Dario, I don't know if you saw or heard that he apparently said that Anthrop would be the only private company in the world in 15 years. Gavin Baker went to Unicorn podcast said that. thought there was a bit of a rebuttal online but I mean the ambition is obviously there right but let's stay on something for a second on this whole seatbased versus consumption based licensing let's just take again just because it's an easy one for me to remember the example of Slack okay and Salesforce >> we know that or I know I guess from just a number of conversations I've had that it has been a challenge to unbundle or to bundle Slack into the Salesforce broad integration at enterprise level because a lot lot of companies had a Microsoft house or you know a Google house whatever it might be and they had teams and you know and it's very difficult to kind of convince somebody to unplug one piece and plug in another right even when they're very happy you know an enthusiastic Salesforce users at least they were okay >> so um that whole opportunity to even like give it for free right like I remember having this conversation at Dreamforce last year does you not you know you get that early win that market early right but it never truly happened Okay. How easy do you think it is to un claim or or to reclaim your stack like and and to re-imagine like an enterprise agreement? Like who do you think that Tam is for this? Do you think that Salesforce should be going out to like their midcap SMB business trying to like basically position you know this opportunity as more consumption based or are you trying to completely work with like longtime enterprise customers and reimagine what the whole infrastructure should look like? Well, I think that this is effectively a profound architectural, you know, I like this word, architectural reorganization of the IT Slack. And last year at Dreamforce, Mark Binov had said this is going to take time. You don't rewire an enterprise stack like in two weeks. That that's wrong and that doesn't exist. So, let's put that aside. This is taking time and this is taking time by putting the pieces together. So, the piece that we had initially was the system of record. Okay, that's an interesting starting point. Now we have where does work not not where does work get done where does the intention of getting work done get expressed by both a human you and I or an agent. Now Slack is one example of that. Microsoft copilot is an other example of that. Senna is a third example and of course Claude and Claude code specifically Claude code is a perfect example that points is a structural mode for any company pretending to have pretending trying to have this position clearly and it doesn't happen overnight. I think that you raised an interesting point which is the pricing of Slack and of course I do not have information into that but it would make sense for Salesforce to have Slack being really the wedge being okay this is deep deep place where intent gets expressed human and and agentic one and then work gets funneled to whatever source of intelligence to the record that is written over amended and creates the loop etc. So that is a very structural position with everyone claiming for that. Then how do you monetize that? I think is a question that is not as important as do you have that strategic position or are you called upon by whatever company is going to have that. Remember when QuickBooks seated that position to Chad GPT? I think it was over one year ago. >> You you say, "Oh, this is great. I have now massive distribution. I have CH GPT. I have Claude who's going to distribute QuickBooks." But that's not so great because you're seeding your interface to intent to the LLM. So you want to be reversing that. You want to send work to the LLM, keep the interface. That's I think what is absolutely fundamental in winning this enterprise battle if it's a battle but I think it is a battle. >> It becomes an engine as far as to the operating system as opposed to actual throughput. >> Exactly. >> I want to talk about a couple of other interesting things that have happened since you were last in the show. So Oracle ask you about Oracle. Very interesting company. Okay. Has had a pretty rough few weeks on the market. >> Mhm. >> It's hard to underestimate Oracle because you know its founders one of the most successful men in the entire world, >> right? has all the right relationships if you want to define success in a particular category, right? Like with folks like Musk and others, you know, but still is like suffering because I think there is a challenge as to understanding how the Oracle of 10 years ago, which really did do very well in enterprise, right? Like as a ERP and a number of levels is going to compete with the future of this, you know, frontier model driven. who knows you know how we actually can describe it type of enterprise right they don't really have a message to what the next 10 15 outside of Oracle cloud which is also doing interesting you know how that looks where do you think a company like that is going wrong I'm like so interested to understand >> look I think that when people think of Oracle they first and foremost think about their infrastructure business which has driven so much growth OCI but which is where everyone freaks out because there's so much depth, so much dependency to open AI. The numbers are impressive, but right now it's much more of a burden that it is an opportunity. This is where we can see that the wind has changed. Not not over the past weeks, over the past year. Like in July, Oracle had very bad performance on the markets as the AI trade was questioned. So they actually have a very interesting software business. They do not have what I call this proximity to user intent. I'm not saying that to criticize Oracle. I think they do not they have not evidenced that in a way like Microsoft, Salesforce or work they has done for now. That's work in progress and of course the AI labs. So I would say they're kind of like they have a strong architectural position from the database part where they come from and they've done a ton of innovation on that part. They're not putting the pieces together where you you say, "Oh, I'm going to have Oracle be my number one interface for all of the work that gets done in the enterprise by humans or agents." It's more like in the process, but it's not the first person or first tool that you see. And they have the burden of OCI, which I think is what everyone thinks about when they think of Oracle stock. It might be unfair, but it's it's the reality. >> You know what's interesting when you say that Fox that a dependency becomes a burden. There is no unilateral metric or definition for the turnoff point for that though, right? Because in some cases, you know, and we see this a lot in the kind of newer frontier side of the house, you know, where you have a lot of like very fast followers who form relationships with folks and, you know, NeoCloud's perfect example. They have, you know, situations where they 70% of their entire revenue is with one hyperscaler. Yet the stock continues to to to go well, right? Like for in some cases it's very inconsistent like you know at what point it is that the markets view like tight relationships and as a dependency and as an economic burden as opposed to something that's seen as like opportunistic >> like I think we need to tie that to like and this might change but open AI to be fair with open AI because they're criticized like I mean the critics against open AI are like right now very abundant. I think they're doing a very good job in the enterprise which we have always discussed here is their only way only way to have a dupal mode and they've made great progress over the past weeks. Now that being said I think open AI is causing a lot of stress. When you think of open AAI you think of wow okay the growth may be impressive though we don't really know the numbers the losses are huge. How is all of this going to be financed? So I think any company that is strongly linked to open AAI which is not the case for entropic which is still seen very positive but that might change everything changes so fast right now when you're associated to I have a big dependency on open AI I would say that for the past months if not more the market sees that as okay that might be a risk factor a big time I want to discount that just that's my perspective I'm not representing the market I'm sharing my perspective and I think that this might change as entropy goes to IPO as the numbers are going to be disclosed in the S1. We're going to have like a first really in-depth view of what the economics of a frontier model company looks like. We had that with Miniaax. We had that with Zapu. But I don't know if US ambassadors were digging into the S1 of Chinese open source models, though we had interesting points. This is the first time that we're going to really open the S1. next week maybe we'll talk about this next time like this could be in the next days and see oh okay so this is what it takes to grow this is what we're talking about in terms of gross margins in terms of operating margin and that might be seen very positively or that might be seen as wow we need to take a step back those numbers are absolutely crazy and are we going to be able to finance this so openai is the poster child so it's it's easy to criticize open again I do think that they did not have a moat going in the direction that they were going before and my point of view is not a good point of view on OpenAI. My point of view is OpenAI right now is showing meaning meaningful traction in what matters so to be followed. I want I mean it'll be very interesting to see what's in that S1 for sure and hopefully we'll be able to actually break it down next week but the early estimations and the early speculation I mean it is pretty crazy numbers right like you know broadly speaking like what what are your thoughts do you think again we have another SpaceX you mean it's going to be even higher potentially right they're saying four trillion like there's all sorts of numbers being floated around like you know again as as someone who's you made your career on like investing money wisely. Do you think that this is a overinflated bet broadly? >> No. I I think that we need to be starting to be we need to be not starting we need to be asking the right questions. The right questions is not anymore. What is the growth of entropic? The question is what does entropic become? Like let's project ourselves five years from now, not even 10 years from now, just five years from now or maybe less. What does this company become? when we go back to what we were discussing at the beginning of the show which is does entropic need to become an enterprise software company or does entropic need to become what I have called the system of execution which is I would say what software is in the agentic era because that's the way that entropic actually grows into its two trillion maybe three trillion dollars maybe even more valuation why cap it at two trillion it could be more and this is where the whole notion of TAM gets into perspective what really uh I made me a bit mad with the Space X IPO was that we had this huge number, the 22.7 trillion on enterprise AI applications with literally zero justification. So we were asked to believe that this company was addressing this huge enterprise AI market and now I think the consensus is SpaceX is going to be the biggest NeoCloud which I agree with from a consensus perspective. But this is not what was said in the S1. Remember the the S1 had a part of Neocloud which was like small. So I'm not the one who wrote the S1. My point is the whole system right now needs a lot of coherence. You can believe in a crazy TAM which again might not be so crazy a couple of years from now. And I think that a lot of people are bashing AI in ways. I'm not like a cheerleader with my pompom saying AI is great. I'm saying you need to really understand what is the change that is going on. If work gets done by agents and you augment throughput massively, you need to think differently around temps than as we think currently. So again that may be credible even for space X. I don't know. I think no one knows. What we want to know as an investor is what is this company becoming? Is what is it what it is telling us coherent with the numbers that we're seeing and what is in the S1 and are we able to not just believe what someone says crazy numbers but are we actually able to chart a path in terms of growth and profitability that makes sense from a fundamental point of view. This is what is missing in the market today. I mean on the surprise we're actually a neocloud debate in some respects you know and I think we spoke about this at the time SpaceX went to markets with this idea that you know we're yeah we have infrastructure but really we're an orchestrator we are you know a leading AI like lab right and a lot of folks like well actually it's really an infrastructure play moonlighting is something else anthropic is very much we are like an orchestrator of enterprise like that is their leading message so if markets loved the SpaceX message being so focused on them, you know, as owning AI like owning, you know, the kind of endpoint for customer value, then potentially anthropic is in a very good place. But again, who knows? When you talk about fundamentals, like who truly knows? I want to go on to something though that I also think um is an interesting conversation and that is, you know, in NYC Wired, we interview a lot of founders, right? We talk to folks across the gamut of health tech, biotech, enterprise tech, fintech, all you know we might do 30 interviews a week sometimes. And one thing I one pattern I've really noticed is there is a lot of money being spent on that vertical application layer on startups that are spent investing heavy. They're not trying to be anthropic. They're trying to be a very niche player. Okay? They're not trying to replace everything. They're not saying you can, you know, remove your entire stack. They're saying, "No, we are going to be the best at underwriting. We're going to be the best at this particular, you know, very niche space." And it's interesting to me because in some respects, a lot of these SAS companies are kind of getting throttled for having like this vertical agenda, right? Where that vertical agenda lines up to payroll, let's say workday is a great example. Yes. HR benefits, whatever. But these incumb these new companies are like challenging the incumbents are getting a lot of enthusiasm, a lot of venture dollars. What are your thoughts there? Like what where do you think that the thinking breaks a little bit around well you can't do that because you've tried you've been doing it for too long and you haven't really created a defensible moat but you can even though you're you know it's your first ever company and you've got $40 million to do it. So that's that's a fascinating debate and I would say it it goes back to okay if you think about software and we had introduced that distinction when we talked about Figma a couple of months ago is you have those backbones so you have those platforms the the Salesforce the workday service now Microsoft let's let's qualify those as like the platforms or horizontal software to take a term that I do not like and then you have vertical which is I'm addressing a specific industry I'm appolio I'm viva systems I'm gwireire So companies that are addressing a specific like workflow in a certain domain and the AI native startups are going like full speed in that. They're really saying we're going to be building what they call these autopilots meaning agents that can perform work faster, better, quicker and that are working quite well. Harvey this week unleashed released it's not yet like it's not yet general general available a availability but it was a very interesting research where they showed that by really uh like fine-tuning the model they were able to have amazing results in the field of legal which is one of the fields that is attracting the biggest fundraising like Lego Harvey so many other companies like Thompson Reuters that announced their own large language model so that field that will finance is moving big time. So I do think that a lot of those companies will build very successful businesses. I do think that they have an advantage coming from we are born aentic AI native. We are born AI native. And this is where we go back to the SAS apocalypse. The SAS apocalypse is over only for those companies that are either a platform converting into this new system of execution which we mentioned or the domain specialist workflow specialist companies that are not this platform and that are moving very fast against the clock that is ticking ticking against those AI natives. there will be I think a lot of casualties in the specific parts because it's exactly what you're saying when you haven't been able to build strong retention when when your product is not the product that people want right now when your UI has not been reinvented maybe you have regulations that protect you those businesses will stick around for a while but if you're in a sector that is not like where you need to have the CRM where you need to have this thing in insurance. I would be a bit stressed. >> I have the same I suppose question about this that I do about cyber companies too, right? Which is no CTO, no buyer wants to manage another license, another even if it's consumption based, you know, it's just another line item on your P&L, something else to monitor, right? Like when we think about vertical application, as much as I love the enthusiasm, I think it's so important within tech that we have this energy and it's great to see because we need challengers all the time. I just think about that kind of, you know, simplification, mass integration thesis that we've come to really see as what people want as much as they can possibly have. And it makes me question, I don't know, the longevity of some of that thinking. But you know what? I actually hope they're successful because like I said, we need competitive market dynamics, right? >> And there's going to be a ton of M&A. We're seeing that already. I mean, I think yes, there's a lot of enterprise fatigue where it's true and it's interesting to see that platforms like Service Now, like Salesforce, like Workday have moved into domains that were not their initial domains like that have gone from ITSM to HR, from HR to CRM. So, we've already seen like a ton of that last year. I I think there's going to be so much M&A of these AI native agentic native companies that are going to be brought to offer like a more broader platform combined with companies that are going to be building their own agents. I mean, we're starting to see a lot of that like companies are going to not even have to go either to an AI native company or to platform, but they're going to be able to plug their own clouduilt agents, open AI built agents on top of this enterprise backbone that may be Microsoft, that may be Salesforce, that may be workday, I don't know. And this is moving fast and entropic goes to IPO in this context and has to prove even faster that they're going to be the winners. You know, I think what you say there, one thing that I really find interesting and I and I I have a real I guess be in my bonnet about this is this whole marketing message of AI native, right? What it means to be AI native like and I think in a lot of respect that also is causing some of this confusion. And I know we talk a lot about, you know, how I think that tech has a marketing problem, but I certainly think it to be true. Raphael, I know we're up on time. I think we're going to hear from Nvidia in about 10 minutes, so I'm sure you've got a little bit of trepidation. So much happening. Thanks so much for joining us. I look forward to catching up in two weeks. >> For sure. Thank you. >> Thanks so much for watching AT&T. We are talking about the agentic mode of this next year in tech. Stay tuned for next time.