Latest / Construction Technology Podcast / After Autodesk University: AI Assistants, Token Pricing, and Autodesk’s Durable Moat
Transcript
- Luigi La Corte: Hello everyone. Welcome to the construction technology podcast. I'm Luigi with Provision. I'm joined by my co hosts Andrew from Join and Mike from Billvision. What's up, boys? How are you? How's the week? Andrew Zukoski: Yeah. Mike Powers: What's up? I did not go to Autodesk University. Andrew Zukoski: My week was good. Luigi La Corte: gonna say yeah I'm surprised I'm the only one here who went to AU. It was really it was actually a really good it was a really good conference. I came away very bullish actually on Autodesk. maybe I thought I I don't know if you saw the the demo that was done by ⁓ by Andrew Agnost. But very impressive. Did you guys see it? Yeah, let's watch it now. Mike Powers: No. No. Luigi La Corte: Okay, I'm gonna share my screen. All right. Mike Powers: No sound? Andrew Zukoski: You gotta give us your best and your agnostic impression. Mike Powers: Yeah, give me yeah, give me the Andrew. Andrew's the goat. He's got his he's got his doctorate in like astronaut astronautic engineering. ⁓ smart guy. Andrew Zukoski: Damn. Luigi La Corte: Can you guys hear the audio, by the way? Andrew Zukoski: Trying figure out a joke about Mike Powers: No, and you and the camera and the video went away. ⁓ man, we're streaming. Luigi La Corte: ⁓ shit. What is this? This is ⁓ this is like amateur hour over here. Mike Powers: All right, as he pulls it up, here's how I'm feeling about the pre con market after AU. Luigi La Corte: That's yeah. Mike Powers: I would not say the pre-con market has a ten billion dollar TAM. Andrew Zukoski: Maybe globally if you include I don't even know. Mike Powers: But there's some good ones in here. Plan better, raise the series B. Build OS, raise the series A. Takeoff AI has a seed. Pre-type pre-contact changes everything. pre-con cloud here. ⁓ Luigi La Corte: Precon Cloud, come on. No, that's that's lazy. This is a good this is a good model. Mike Powers: So takeoffs, estimates, bid packages, plansets, value engineers, a big market, just not that big. I think is a good Andrew Zukoski: Has Sequoia done a construction deal post plan grid? Mike Powers: They invested in a company called my gosh, it is what was that company called? A an accounting software company. No. Adaptive just raised though, congrats to Adaptive. ⁓ Tidemark led. ⁓ Luigi La Corte: Adaptive. Yeah, they did. Thirty mil. I met with Tidemark at AU. very thoughtful investors. Okay, let's look at this. no, Grace ⁓ and believe Andrew. Mike Powers: So I David. Grace. Luigi La Corte: Okay, let's look at this let's let's look at this demo. Let's get some live reactions to this demo here. Okay, can you see this? All right. Mike Powers: Yeah. Luigi La Corte: So what are the thoughts? Andrew Zukoski: So I was at Join, I was at a computational design company called Flux and it like right when Dynamo was getting going and I don't even know. it means that like for for years, like has demos of technology where you're gonna press a button or you're gonna say a thing and all the is gonna get done and like the layout will be started from a place and you can tweak it if you need to. and so there's like starting of skepticism from me this time is radically different. But it is interesting that like You don't need like some crazy dynamo script that someone in the back office cooked up and will have to maintain. Like plausible, like how do you audit it? How do you audit the layouts? How do you get comfortable with them? How do you update them? ⁓ I think you still end up for that particular demo, ⁓ a bunch of stuff around manually and Revit. That's take. Luigi La Corte: I think okay, that's an interesting take. I'm actually I I I left the I left being quite bullish on Autodesk and ⁓ for two reasons. One, the move to formula means that everything's kind of packaged in this single price consumption model, which gives you access to all these tools. two, I think they're competing the the what's that? Tokens, ⁓ Second, they ⁓ it looks like they made the choice to complete with to compete with Claude and Mike Powers: Tokens baby, let them burn. I love it. Greatest bit of smell ever. Luigi La Corte: ⁓ codex chat GPT to own the the surface, So as opposed to let's say like Nemichek Bluebeam that decided to integrate with Cloud. I think it's really interesting because the fact that they can put all their apps now as artifacts allows you to simply prompt and get put into their tool without having to learn how to use Revit or learn how to use any of these tools, but then continue from where the the the model got you too. I think that's very interesting. ⁓ there's another con there's another yeah, go ahead. Mike Powers: So let's let's just dig it. Can we dig into that? I think that's like fa I think that's a fascinating perspective. Like, so you say that the data research team behind Autodesk is a better position than data research at Claude OpenAI? like where does where does that curve, where does that Pareto curve hit where like their industry expertise, their foundational model? Training. is that the ⁓ that your thought? Like that actually like Autodesk is going to be in a better position to train a better model than an anthropic and hire the right people to lead that team to ru train a better model. like ⁓ that it's a very, very big business bet. because like ⁓ my position like LLMs, I know I I know we we had ⁓ Augmenta on two weeks ago and his like, ⁓ well, LLMs can't fix this, and obviously like. Luigi La Corte: Mm. Mike Powers: anthropic and open AI are very L are LM heavy, but there are like the the world models that doing kind of the opposite. Like is the bet that Autodesk can be that better world model? curious to your thoughts on that, just like that layer deeper on that. Andrew Zukoski: Right? Luigi La Corte: I think Autodesk has built their own model, but I think they built a model for tool calling with Autodesk tools. I I don't think they're using an off the shelf like an Astra or like a fable 5.1. I'm they've probably taken a model like a Quen and they've fine-tuned it to be a good model picker or a good tool picker. what do I think this? So w let's say using okay, let's say ⁓ prompting in the Autodesk AI assistant, which is what they call it, right? And I wanted to do some some kind of tool for or some some kind of some action for me. Like, hey, I have an RFI here. I want to actually want to draft an RFI. Great. gotta use MCP. It's got to call individual tools. It's got to know the sequence to call this tool, then this tool, then this tool, right? they're probably they probably can just stand up an LLM themselves, right? Quinn is fairly easy to to to stand up. ⁓ But it now needs to call the right tools. I think what they've done, because they have the traces, I'm sure they have this data internally. They've probably just said, like, hey, 80% of our users go here and then here and then here, and they'll use these individual workflows. That makes sense from me, it makes sense that they would train a model to to recommend which which tools to call within MCP for an individual product. and then they obviously get the margin advantage. So ⁓ I bet you they just did it. But that's not like a very complicated model to build. So makes sense that they would do it. but I don't think that you need a very performant model to be effective in context. I just think you need to work within the places that general contractors and and subcontractors work. And like Autodesk has all those tools. So even if I had the best model in the world, let's say it was like, you know, GPT 10, if I was saying, Hey, I want you to send me this RFI. It's not gonna go build a project management system from scratch. It just needs to work within Planned Grid. And like Autodesk has that advantage. So they don't need the most performant model. They need the one that's like a very efficient, high margin tool picker. That's my perspective. What do you guys think? Andrew Zukoski: Yeah, I I know I'm expressing skepticism around like press a button, a building comes out in design, but also left bullish ⁓ on Autodesk's strategy because they have a very durable moat, especially on the design side. Like for the next ⁓ many years, potentially Revit is going to be where 3D models ⁓ of like buildings and anything that needs to interact with that is gonna have to like deal with Revit in some way. and that that gives them they get to control and choose how much of that they want to like own. Do w want to expose MCP tools? Do want to build their own harness? Do they want to like fine-tune their own model to do tool calling? ⁓ I think they've got the to do that. Like we can't we we don't have the the the scale To do that effectively. Like we we we cannot like fine-tune our own model and build a harness around it and do so in a way that will provide the level of customer experience that like our customers deserve. Like there's just too much work to do there and we don't have the scale to afford it. Autodesk does. so yeah, I don't think they need I don't think they need better, I don't think they need more data of annotated plans. I don't think they need like more data as in having more RFIs or more labeling. Like they don't have to compete with the foundational models that. ⁓ like unshakable wedge is the systems of record that people are going to have to interact with, especially on the design side. Luigi La Corte: Completely agree. Let's go into ⁓ let's go into a demo, more construction specific demo from the keynote. All right. So from a G C's perspective. Andrew Zukoski: God, the like off white backgrounds with rounded corner boxes, that's like so AI tool in twenty twenty six, right? Luigi La Corte: One thing to note here, like I I do like the way that they've managed space. Like they clearly have put a huge emphasis on artifacts and previews. Like it's almost it's more than 50% of the screen. ⁓ which again is I think why they wanted a kind of custom work surface and they didn't want to work directly within Claude Codex, right? They they know that they need to show off the documents. Andrew Zukoski: Yeah. Yeah. Mike Powers: Mm-hmm. Yeah. Yeah. Luigi La Corte: Okay, so they said, Hey, there's this clouded area, what's the impact on schedule? Andrew Zukoski: Hmm. Luigi La Corte: So, I mean what do you think, Mike? W w where does your mind go when you see this? Mike Powers: So I think structurally it makes a lot of sense. I think the execution of relying on multiple agents is the wrong interface. And what I mean by that is like they're defining ta my understanding is they're defining like specific tasks, like specific agents. And I don't think that's gonna work at scale because then you're relying on the user to decide which agent to engage. Again, there can be agents that you're like having a tool call, but like the what's I've seen successful in a world of like agentic layer is what Sierra has done. And there's one agent at Sierra that does every possible thing you can do. And it has the context to do those things. ⁓ I think that is the interface is the incorrect way to engage with this. I know there's other construction tech companies, we won't name any names that are doing the same thing of like launching specific agents for specific workflows. And I think that's a hurdle that's not going to be able to be overcome. And it's a structural thing that they're going to have to fix. Luigi La Corte: Sorry, what what's the hurdle specifically? Mike Powers: So I I you lose all the context across the different agents, and you're making the user select an agent to do a task. So you should never have to select an agent to do a task, it should just do the task. Because now you have assist, now you have agents, now you have whatever other tags they add on the other side. I think that's gonna hurt the customer experience. It should just work. And this is like the this is the drastic shift that they're gonna because right now Autodesk is a workflow tool. And if you want agents to do work, agents to communicate. And agents to do things across the different tools, that interface has to change. And it's a very, very hard thing to rip out. ⁓ I think that's the challenge. So like saying, hey, ⁓ that these there's a there's a there's a red box around this. The architect is saying to move the wall is reactive, it should be proactive. And I think that again, it's like very conceptual. I know that's very conceptual, but like that's like mentally the mental model around that is the best platform. Luigi La Corte: Mm. Mike Powers: Is going to be a tool that has an agent that does proactive things as opposed to reactive things. And it's hard for them. They're like, they're in this existential autodus is an existential crisis where they're like, hey, like if Claude can spin up a pro again, if Claude 10 can build a project for instantaneously, again, I'm not saying it can't, but if it could, that's an existential threat threat to Autodesk. So they need to figure out how to wedge into that world and they have to use their existing workflows and build agents on top of it. And I think companies that go the opposite direction that are more agent agent first are gonna be more could be more successful if that's the bet that wins. Andrew Zukoski: I don't know man. Luigi La Corte: Yeah, have two re two reactions to that. One, I think a lot fewer people at enterprises are using Cloud and Chat GPT than you'd like to think. I know you know this, but ⁓ like the the IT IT teams just get really uncomfortable with security with those tools. ⁓ two. Mike Powers: Every every Fortune five hundred company is getting called on by Claude and Open AI and the majority of them have subscriptions, I would say right now. If you pulled the top five hundred fortune five hundred companies, I bet you ninety percent have Claude A or open AI API AI access. Luigi La Corte: Really? You think that's in construction too? I'd say it's excuse to like mid market and S ⁓ B. Mike Powers: Correct. Yes. Andrew Zukoski: Yeah, I agree with that. But there's like a difference between a contract and like adoption and usage. Yeah. Mike Powers: But this but this pitch this pitch is enterprise. This is you're not a mid market general contractor is not doing this. Luigi La Corte: Maybe. Mike Powers: But ⁓ 'cause they're you they're using auto audit as tools, they don't revit. They're like like you're gett still getting the two D drawings. This is a direct pushback to like what Palantir is pushing out. Like this is it this is th this is straight enterprise. That's what the token model is, that's what the consumption model is. That's where Autodesk thrives is in enterprise accounts. Luigi La Corte: Well, like the benefit here though is like I mean you still have a team based project. People still are reviewing the same documents here. You need a workspace to save these documents. So like you have to agree with that. Two, you want to review the same documents. So if you can both read and write documents and you can make that more efficient here, it's just it seems like a win for Autodesk. If you're single player mode, I would agree. Like Claude, as long as I have the context, is great. But as soon as you're multiplayer mode, I feel like Autodesk has the advantage here. Mike Powers: So but then all right yeah, I I okay, so but then you're like, all right, I'm charging three dollar three tokens a day a seat for BIM three sixty docs form. Am I gonna be able to charge ten tokens a day for this? I don't think I think the answer is no. There is this gonna be the expect this gonna be table stakes across the platforms because Procore is gonna release something exactly the same. And then it's gonna have cost suppression at an enterprise level. And that that business mechanism is gonna be a struggle to scale through, which is gonna open up the opportunity for other companies to come in and do different things at a better capacity because of how the business model is structured. Andrew Zukoski: Wait, I don't get it. So Autodesk has all your data, all your docs, bunch of workflows around it right now. They're they charge three tokens a day for that. Hey, look, we want AI to be able to do things with that. Can we get 10 tokens a day? No, you can get three or three point one because it's an expectation. How does that open up budget for other people? To come in and like make money on top of the Autodesk data. Like Autodesk has this interesting frenemy relationship with the ecosystem that lives on top of their platforms. Like AutoCAD lived and died on extensibility and people working around it, Revit as well, Forma, you know, maybe one day. How does that open a lane? Autodesk being unable ⁓ Being told like, no, we just expect this functionality inside of the system of record for our documents. Like, how does that open a for someone else? Mike Powers: I I think the I think the three the three tokens a day is getting compressed down to two tokens a day. Andrew Zukoski: And then so so that gets that gets squished and now there's a token a day for this AI stuff and Autodesk will fight for it and will win sometimes and won't win sometimes. Mike Powers: So we'll have their opportunities, and then there'll be specific and the companies that offer outcomes on top of that are gonna be able to be more successful, would be my position. So hey, Autodesk and Pro Core continue to thrive and push towards continually capturing the project management space. The project management space ⁓ is not accelerating as quickly as AI spend. ⁓ because fees are getting compressed, because owners are getting smart. There's a lot of cost collaboration. An owner buys a seat of Autodesk that shares with the GC. So why does the GC need a protocore subscription? There's a lot of friction in that motion as we become more collaborative in the PM market space. So that fee compression goes down, that fee compression goes up, reducing cost or reducing what they can bill. So then it opens up space. And that open space is where companies that are outcome-based, companies that have a unique position of how to tackle things or sell the service, the like the service as opposed to the software, I think you're gonna be able to absorb that. ⁓ 'cause I don't think the TAM, I don't think the TAM for a general contractor is increasing. It's the price of project management they spend will decrease and then that open wedge is where the opportunity is. Andrew Zukoski: But if you're Autodesk, you're in in however long it's existed, Autodesk has not I guess they do payments now ⁓ with GC Pay, but I don't know if they've I think that might just be managed payments. I don't know if they actually did like what Procore did. Yeah. So like Procore took a look at insurance and then ran away. and they do do payments now, and that's a slow, slow burn to grow, but like is a real a real new business line. Autodesk Luigi La Corte: Mm. Mike Powers: Right. They're not. Andrew Zukoski: sells software. Like the business model shift has been going from like box software to a subscription. ⁓ and there's like some pricing changes there too. Twice they've tried to build a content library and they've run away from that. Like Autodesk is is it seems unlikely they will go in and say like, hey, we'll procure all your HVAC equipment for you. Like and we'll get you better than benchmark by two percent and then we want a quarter of the savings that we get or whatever. Like won't whoever these new service providers are Mike Powers: Mm-hmm. Yeah. Andrew Zukoski: The people who coming out and saying like, hey, you have an outcome you need, you need to get through permitting, we're gonna get you through permitting, we're gonna use technology to do it faster, better than you've ever seen it before, won't those companies be Autodesk customers as well? Mike Powers: Always company if you are less than us. I don't know. I don't I don't know. We churned off Autodesk. We were using ⁓ we were helping customers with Bidboard and their building connected account. ⁓ it's a good question. I think like the overall zoom is that there's a dollar, say for a contractor, there's a dollar spent on software, whatever the number is. Three dollars out of every ten is spent on software. What is it? One dollar of one of every hundred dollars spent on software. ⁓ that's the what Autodesk and Procore are fighting against. Luigi La Corte: Come on. Mike Powers: The opportunities are in the other ninety nine dollars. And that's a very hard that and and none of the none of this, this is not this is still in the dollar bucket. None of this is none of this is in the ninety nine dollar bucket. Andrew Zukoski: Yeah. Yeah. Mike Powers: I still had not I I until a large general contractor comes to me and says, Hey, ⁓ we had all this rework and the software helped me reduce this rework. And I, as a general contractor, made X dollars out of that in profit. And I shared that profit with the technology provider that did that for me. Until that happens, we're still in the dollar bucket world of the a hundred dollars. 'Cause you can say whatever you want about like, hey, we know we've reduced rework and we saved thirty dollars and that thirty dollars doesn't go to the general contractor. Luigi La Corte: Yeah it does. 'Cause it c Yes, it does. It comes it comes directly out of their it comes directly out of their contingent contingency. Andrew Zukoski: Yeah. But I I guess the question is like these other entities Mike Powers: Does not. Luigi La Corte: Andrew, can you hear me? Go ahead. Andrew Zukoski: sorry, I was just gonna push back on the like the other ninety-nine dollars if you're a general contractor ⁓ self-perform scopes, overhead, and things you've subbed out, right? And so like let's just focus on the like you subbed out some scope. Somebody's like ⁓ the design, fabricating, procuring, installing a scope of work. Like, aren't those people are those entities going to need to interact with like one ⁓ all the data of the project, all the documents, all the schedules, all the communications to all the other stakeholders on the project? Like aren't they going to need to plug into Autodesk, whatever's being used to manage, manage the project and find the data? Mike Powers: Yeah. And they're and they're doing that now. They're like it's bring your own seat, bring your own license. Andrew Zukoski: Like autodas might never get happening. Yeah. Mike Powers: It doesn't sh I just don't I don't think that dollar for the BIM seat that that user bought for the subcontractor to access the RFI tool in Autodesk becomes a two dollar seat. Luigi La Corte: Well, if they if well the bet Autodesk may be making, I don't know their business model that that well, but if they get people to prompt, move prompting from ChatGPT, everyone I know in construction kind of has like a side chat GPT account, right, where they're doing work. But if they can bring that token spend from Chat GPT into AI assistant, then yeah, they will be competing against the token spend at Chat GPT. They'll just do it where their work is being done. Mike Powers: But the token spend's not even but for the Claude in in open AI it's not even budgeted right now for a general contractor. A general contractor can't spend ten thousand dollars a a month for their t hundred employees to run Claude. There's like there's no budget there already. We're already seeing that friction. Luigi La Corte: I don't know. I I I see I see G C spending money on AI tools, right? So if they can compete against other startups, sure. Mike Powers: They have to. It's the cost of bit they have to. So like that that's just that's just gonna be fee compression. They're not making more money because they're using Chat GPT or Cloud. Luigi La Corte: I think GCs are becoming more efficient using these tools. I believe that. But also the other angle here is Autodesk is not the incumbent, but this is very advanced capabilities. So now if you're considering, okay, do I want to stay with ProCore? Autodesk is moving pretty fast. Plus, I get these really neat tools. It's an interesting draw. Like, look, part of Autodesk doing is like they just need to keep up. Part of it is we want to take market share from Procore. Part of it is we want to increase token spend, if I understand their business model. So okay, question for you guys. Do you think that this move helps them accomplish those three things? Mike Powers: Mm-hmm. Yeah. Andrew Zukoski: Yes. Luigi La Corte: Yeah, I think so too. Some of it's also really look at this. Mike Powers: The talking is outrageous. Like they just rolled that out like two weeks ago, I'm guessing, because there's a new model release. Luigi La Corte: Very fast. I know this is like a canned demo, but it's good. Very low latency. That's very cool. To take the the like the main call like the main tools from this app and put it into chat, I think is actually very cool. Okay, thoughts? Bullish or bearish? Mike Powers: So again, this so this is like the same thing happened when we went to when we went to BIM. now an architect and engineer is gonna be able to do a trillion iterations. It's not gonna make them any more money. It's awesome, it's beautiful, it's really cool. They're gonna engineers are gonna be able to give their customers and architects are gonna be able to give their owner clients really, really cool, ⁓ flexible outputs, but you're just increasing the complexity of the building and it just this it's a scope creep. Like the arc the Empire State Building built in nineteen thirty-four was very, very simple. World Trade Center One was very, very complex. We're gonna have this next leap to very, very complex buildings. Because they're gonna use all the time. They're not gonna reduce their billable hours because of this. They're gonna increase their billable hours because they're gonna be more complex projects. It's cool, it's great, but it is not increasing the profitability of an AE firm by doing any of this. Luigi La Corte: Okay, what if so you're talking about from the perspective of an owner architect engineer, right? Yeah, this that's right. This is a owner architect, right? I think they're talking so previously they're talking about the GC, where going through the the markup and the schedule impacts, right? Like even even that you have to admit that they'll be able to get more done on a project because of these tools. If look, if if if it if Mike Powers: That's what that's what that was showing. Yeah, that was showing hey I'm I'm working on some new designs. And what I grin, they're gonna make it more complex. Like, ⁓ hey, you wanna see the seven different angles that I can produce this facade at now? ⁓ let's go let's go spend seven hours sitting in a conference room doing these tweaks. ⁓ it's great. I'm gonna pull up this beautiful red. It's cool, it's awesome, it's great. But that doesn't mean that the person's designing faster. You're gonna fill up all the time. Luigi La Corte: But that's not the constraint the G C s that's not the constraint that the G C's solving for. Mike Powers: The constraint what's the in that so that the example we showed that was AE and the G C side, like you're not gonna have that superintendent or that project manager be on three projects now because they can use an AI agent to to do conflict resolutions on a project. Luigi La Corte: That's true. But can they catch an issue can they mitigate liquidated damages a little bit faster? Probably. Mike Powers: That's the expectation. That's the expectation of the contractor. ⁓ the my first project executive on a project, he goes, you know, the expectation is the project's perfect and the owner's rep's only gonna yell at you when you F up. So this how this reduces the F ups. But the expectation from the owner's rep is already it's a perfect project. Luigi La Corte: Yes. But they're not always perfect projects. Like they they're usually over schedule, over budget. That's what you're that's the baseline. Right. So if Autodesk can reduce reality maybe by five percent, that's money in GC's pockets, right? Like that's just cause the cost of that is profit erosion. Mike Powers: They're never perfect. Correct. Correct. I think the biggest value prop is the s ⁓ employees happy because now a PM is not as stressed because that job ⁓ brutal. Like going in and negotiating. Hey, I I I remember negotiating we were rerunning electrical conduit at the Philadelphia Eagles Stadium, and I sat with the project exec for the electrician, and he goes, It cost me a million dollars to rerun this. The owner's rep said, Hey, it's three hundred thousand dollars. Luigi La Corte: Yes. There's also that that's a great pitch too. Andrew Zukoski: That's big. Mike Powers: And we just argued back and forth for three weeks to figure out what the actual cost was. And I was looking at my timesheets, I was looking at my daily my DCRs, ⁓ like, ⁓ they had 13 electricians on this day. That s doesn't make any profit. That if that was just like it is actually five hundred thousand dollars, that just reduces my stress ⁓ the MEP super on that project. And that's what that does. That's great. It's be it's beautiful. Andrew Zukoski: And then you never would have gone to construction technology. Mike Powers: That was a great project. I loved it. I had to check every seat in the stadium. We installed what was it, ⁓ new seats, and I had to check to make sure they're all bolted down. Andrew Zukoski: Yeah, Luigi La Corte: Okay, so okay, after after the after this demo okay, so look, this is this is the demo, right? There's other things that happened at AU. I didn't catch the precon roadmap, Andrew, did you? Andrew Zukoski: ⁓ I did not catch the pre-con roadmap. I caught the AEC vision roadmap, which had a bunch of Forma stuff, a fair amount of AutoCAD stuff, which made me remember I don't think AutoCAD revenue has ever declined. Like AutoCAD is still very important in this industry. and then had the forma build overall roadmap, and I I thought it was super solid. It felt It felt like a collection of really pragmatic like innovations and investments in AI. It didn't didn't feel flashy. It felt like stuff that lined up with what we hear from our customers. you know, pragmatic integrations and improvements inside the platform. So like seeing building connected, this beloved product like make its way inside a forma, like that's the right direction. like I think I I I I I think Autodesk Construction customers should feel good. Luigi La Corte: Yeah, I think so too. Yeah, they're definitely gonna catch up, I think, to the general capabilities of models. Yeah, ⁓ I again I left I left bullish. Mike, what's your perspective? Are you would you buy or sell out a desk after this? Mike Powers: Buy the stock, 'cause the capital return in other areas is higher, is gonna be higher and accelerate faster. Luigi La Corte: Okay, so you're so you're short then. Mike Powers: Not short, I just would say that capital, the ROI the ROI is better placed in other areas than in construction tech software or AEC software, though the data center boom is booming. Luigi La Corte: Yeah, okay. I think it was right. Andrew Zukoski: ⁓ you're just short the whole sector. Mike Powers: I again if you're saying I get you I can I can put a dollar into anthropic at you know last year at ten billion dollar valuation and now it's worth two trillion dollars or I put a dollar in autodesk last year like that ⁓ that that calculation is it like having count that that calculation is very very hard to ⁓ rather just throw throw it into the mag seven or the like literally like are goo is Google gonna increase revenue faster than Autodesk right now? Yes. Luigi La Corte: Mm will they? Andrew Zukoski: Yes. Mike Powers: The amount of inference we're using b ⁓ my gosh. We're at one percent market saturation on L LLMs. The inference the inference it it's un it's cat it's free cash, baby. Luigi La Corte: I don't know about that. I wouldn't I wouldn't say we're at one percent market penetration for LLMs. Mike Powers: For inferen no no not not mark no for inference capacity for usage. That's that one percent market con p that that. Andrew Zukoski: Mm. Mike Powers: And if we get to five percent, ten percent at this time next year, every every every center builder that you That would be that so that's that would be my investment. I'm not I I think it's really cool what Autodesk does. I think it's very, very powerful. It's gonna make their customers very, very happy. It's gonna be able to have them reconcile their cost increases that they're gonna continue to push they're gonna say, hey, you need this tool now. You need to be AI forward as an architect or engineering firm. it now costs four dollars instead of three dollars a token. that's gonna force price increases across the board. It's gonna force AE firms to charge more to owners to build buildings now because of the Autodesk increase that is already a line item in their bill. stuff like that, it's great. That that will continue to grow. It's a great business. But ⁓ put a dollar there somewhere else, I'd probably do I'd put it put it in put it in Google. Luigi La Corte: Google's already at like a four point four trillion dollar ca market cap. Mike Powers: The thing is gonna hit ten it will i if if inference hits like it's supposed to and all these tools work in autodesk like it says and it makes engineers faster, that's a ten Google's a ten trillion dollars. Luigi La Corte: Mm. I think Waymo is also huge. Andrew, what do you think? Are you buying or selling? Andrew Zukoski: I would probably buy Autodesk. Mike Powers: Ha ha. Andrew Zukoski: Yeah, I would probably buy it. I don't know that you put all your your venture bet, maybe a retirement portfolio. Like it'll be interesting to I'm gonna need that money that we might put in Anthropic, you know, a few years from now. And I believe people will still be using AutoCAD and Revit ten and twenty years from now, but L Ms don't seem to be sticky. so we'll see. Luigi La Corte: Yeah, I agree. I I I agree with that take. I think if you're if you're hosting like if you have the compute, you're good. I think if you have the like the underlying tools, you're good. The LLM, it's a very it's a very sticky business. Like anthropic and open AI, I think the business model of training, like a frontier model, doesn't always hold if you're in the lead. but then again, look at what happened with Astra. Mike Powers: The inference though. Luigi La Corte: Like they've sucked a bunch of market share from Claude, people are using Astra a lot more now, too. So maybe that maybe that it Andrew Zukoski: How many times in the last two years has the market switched that way? It's like three times, right? Luigi La Corte: So many times. Mike Powers: But you but you but you know where it's still running. Gemini or sorry, GCP, AWS, and what's Microsoft? Luigi La Corte: No, exactly. Yeah, the compute, they're doing wrong. Okay, let's ⁓ let's cut it here. good AU conference. That was that was my first, by the way. actually very fun. And and it's great, you get just to see a bunch of people that you know and a lot people are chatting. It's definitely where the conversation was. Okay, let's cut it there. Thanks, gents. We'll talk next week. All right, see ya. Mike Powers: Doesn't matter. Andrew Zukoski: Cool. Thanks guys. Mike Powers: Yeah.