[ { "start": 0.0, "end": 4.76, "text": "This podcast is presented by NXAI, your partner for time series." }, { "start": 4.84, "end": 6.84, "text": "foundation models and physical AI." }, { "start": 18.92, "end": 22.68, "text": "Welcome to a new episode of the Industrial AI Podcast." }, { "start": 22.82, "end": 24.96, "text": "My name is Peter Sieberg And I'm your host." }, { "start": 25.8, "end": 28.08, "text": "Today i am going be talking to Bastjan Out." }, { "start": 28.12, "end": 30.56, "text": "He's CEO at Simcon." }, { "start": 31.6, "end": 40.16, "text": "Bastjan & I today are gonna talk about nothing less than The remaking of computer edit engineering." }, { "start": 41.0, "end": 41.6, "text": "Hi Bastian." }, { "start": 42.54, "end": 43.62, "text": "Great to be here, hi!" }, { "start": 44.06, "end": 46.2, "text": "Please introduce yourself to our listeners." }, { "start": 46.28, "end": 46.68, "text": "Bastian." }, { "start": 47.16, "end": 49.86, "text": "Yeah so I'm Bastionaud and the CEO of Simcon." }, { "start": 50.12, "end": 62.02, "text": "We make injection molding simulation software have been doing for decades really And we're in the midst of a pivot towards A completely new tech stack using AI methods." }, { "start": 62.5, "end": 65.04, "text": "Extremely exciting which is probably why i am today." }, { "start": 65.94, "end": 66.88, "text": "In terms my background." }, { "start": 67.26, "end": 75.52, "text": "before I joined Simcon was at McKinsey digital and analytics doing mainly analytics transformations for large corporates." }, { "start": 75.92, "end": 82.54, "text": "And I have a PhD at the intersection of psychology neuroscience in economics, yeah that's my background." }, { "start": 83.16, "end": 83.8, "text": "Sounds great!" }, { "start": 84.04, "end": 89.68, "text": "Tell us just a quick introduction we're going to go into details later about your company Simconn." }, { "start": 90.28, "end": 93.54, "text": "Yeah so we make injection molding simulation software." }, { "start": 93.82, "end": 101.1, "text": "it is used by engineers who design parts to make those parts." }, { "start": 101.54, "end": 113.4, "text": "And basically, it helps you understand what's going to go wrong during manufacturing because you would think you just inject liquid hot plastic into a cavity and then out pops." }, { "start": 113.44, "end": 122.16, "text": "the part has the same shape as the cavity but that's not the case Because as it pulls It turns out to shrink and I won't shrink by the same amount everywhere." }, { "start": 122.5, "end": 123.96, "text": "so You want a straight part?" }, { "start": 129.1, "end": 131.7, "text": "Let's stay with molding a bit." }, { "start": 132.08, "end": 147.78, "text": "I've been introduced into the technology couple of years back or have been following the technology for quite awhile, maybe not everybody should tell us gives it quick introduction to what is plastic mould and What Is The Central Role?" }, { "start": 149.8, "end": 149.84, "text": "So" }, { "start": 150.06, "end": 154.6, "text": "Plastic Injection Molding is A Way Of Manufacturing Plastic Parts." }, { "start": 155.16, "end": 159.98, "text": "It's Used When You Make Parts large numbers, right?" }, { "start": 160.04, "end": 166.7, "text": "So if you're in the hundreds of thousands or millions of parts per year then it's a very cost-effective way an energy efficient way." }, { "start": 167.48, "end": 180.42, "text": "Of manufacturing these parts almost all industrially manufactured goods have plastic injection molded parts like for example The lens and your cell phone camera the bumper on your car." }, { "start": 182.7, "end": 186.2, "text": "All of these objects contain Plastic Injection Molded Parts." }, { "start": 187.54, "end": 196.68, "text": "And the way it works is you take plastic granulate, heat up and inject into a cavity inside of mold." }, { "start": 197.12, "end": 203.74, "text": "The mold has a large chunk of steel that's shaped like the part in its center." }, { "start": 204.58, "end": 217.12, "text": "You inject the plastic there to cool down until solidifies then eject by opening And so you do this on repeat, right?" }, { "start": 217.3, "end": 220.56, "text": "So the things your optimizing for are number one part quality." }, { "start": 221.1, "end": 224.6, "text": "You want to make sure that the part actually has the geometry that you want." }, { "start": 224.96, "end": 225.86, "text": "is it in the right shape?" }, { "start": 227.16, "end": 229.86, "text": "and that's a non-trivial problem to solve." }, { "start": 230.5, "end": 233.0, "text": "but then also just basic thing like will it fill at all?" }, { "start": 233.36, "end": 236.2, "text": "these or highly viscous liquids need lot of pressure." }, { "start": 242.86, "end": 244.3, "text": "Let's say that you're LEGO, for example." }, { "start": 244.98, "end": 252.0, "text": "You are probably producing multiple bricks inside of the same mold in a single shot but will they all fill simultaneously?" }, { "start": 253.24, "end": 255.12, "text": "You want to be as synchronized as possible." }, { "start": 255.2, "end": 261.22, "text": "so because otherwise the slowest cavity kind determines overall cycle time and there is lots technical issues around it right?" }, { "start": 261.26, "end": 263.7, "text": "So its an entire field of study." }, { "start": 263.74, "end": 265.32, "text": "how do we understand this?" }, { "start": 265.4, "end": 278.9, "text": "well It really is art to build these molds while designing cooling systems And design the geometry and materials of these molds in such a way that the part quality works, so you minimize cost." }, { "start": 279.54, "end": 279.82, "text": "Right?" }, { "start": 280.9, "end": 284.78, "text": "The mold does not always work right from the beginning." }, { "start": 284.82, "end": 296.1, "text": "That's why your company simulates as good as possible what is going to be happening when building them all... When the mould has been built." }, { "start": 297.6, "end": 298.0, "text": "That's right." }, { "start": 298.08, "end": 306.32, "text": "So essentially if you build a mold first and then test it for the first time on your machine, And you didn't do simulation beforehand." }, { "start": 307.18, "end": 309.14, "text": "You're gonna discover that there are issues." }, { "start": 309.44, "end": 312.52, "text": "Maybe the part geometry isn't right maybe won't feel correctly?" }, { "start": 312.78, "end": 316.54, "text": "Maybe there are weld lines that form in places that are visible." }, { "start": 317.82, "end": 319.44, "text": "so all kinds of quality defects can happen." }, { "start": 319.58, "end": 325.74, "text": "and If you already have a physical mode if you want to fix those Issues than There two ways." }, { "start": 325.9, "end": 328.82, "text": "number one is you can try different injection molding machine settings." }, { "start": 329.26, "end": 330.32, "text": "If you're lucky, that might work." }, { "start": 330.6, "end": 331.26, "text": "usually it doesn't." }, { "start": 332.12, "end": 334.96, "text": "and then the other method is You correct them all right?" }, { "start": 335.0, "end": 336.76, "text": "So you ship it back to the mold maker." }, { "start": 336.98, "end": 341.42, "text": "they shave off bits of steel here in there And Then hopefully it gets better." }, { "start": 341.86, "end": 343.08, "text": "It's expensive if super slow." }, { "start": 343.18, "end": 356.9, "text": "it takes weeks To do these physical corrections once you've got that physical mode Right as so what our software does Is essentially it virtualizes The discovery of those errors and then searching for this solution Before you build your mold." }, { "start": 357.16, "end": 365.16, "text": "You run these simulations to understand what's gonna go wrong and you fix These issues virtually which is obviously way cheaper and faster than once." }, { "start": 365.2, "end": 366.04, "text": "we're in the physical world" }, { "start": 366.68, "end": 367.1, "text": "Very good." }, { "start": 367.24, "end": 370.5, "text": "Tell us a little bit more then about that simulation." }, { "start": 371.3, "end": 374.5, "text": "Does maybe in general, you know since when do we do it?" }, { "start": 374.58, "end": 375.72, "text": "What exactly is happening?" }, { "start": 375.8, "end": 377.58, "text": "And then may be want to move into." }, { "start": 378.04, "end": 383.04, "text": "you know Your specific Simulation platform on this sign called cat mold." }, { "start": 383.66, "end": 389.94, "text": "yeah so What this simulation software does is it simulates how the cavity fills." }, { "start": 390.42, "end": 398.56, "text": "So, its got a database of materials with measured properties on these materials that determine they behave physically." }, { "start": 398.92, "end": 402.38, "text": "and then you load in the geometry usually the CAD geometry at the part." }, { "start": 402.62, "end": 403.62, "text": "It's tessellated." }, { "start": 403.98, "end": 413.72, "text": "so essentially turn continuous geometry into something composed of finite elements because thats what math needs." }, { "start": 414.02, "end": 419.24, "text": "And then you simulate how it fills and all kinds of physical fields associated with that." }, { "start": 419.66, "end": 427.6, "text": "So temperature, pressure, shear rates are a type... Shear is the friction that happens in liquids." }, { "start": 427.78, "end": 430.74, "text": "essentially if not too much of that we can damage material." }, { "start": 431.36, "end": 432.78, "text": "things like that are simulated." }, { "start": 433.72, "end": 442.54, "text": "The way classical simulation works does this by solving large systems of partial differential equations, right?" }, { "start": 442.58, "end": 449.96, "text": "So like very advanced complicated compute intensive math which is comparatively slow." }, { "start": 450.44, "end": 452.94, "text": "Right so it takes a long time to run these simulations." }, { "start": 453.04, "end": 460.88, "text": "now if you look at classical simulation engines and ours as among the best in the world super fast and performant The numerical version." }, { "start": 461.28, "end": 466.0, "text": "but still for our large part It's on the order of hours To run a simulation." }, { "start": 466.26, "end": 468.14, "text": "If it's a small part, It could be I don't know." }, { "start": 468.42, "end": 472.6, "text": "Twenty thirty forty minutes if there is large party could be four or five six hours right?" }, { "start": 472.74, "end": 473.92, "text": "So that's the long time." }, { "start": 474.62, "end": 479.3, "text": "and so That means uh...if you want to after You've understood what your first." }, { "start": 479.66, "end": 479.8, "text": "Yeah!" }, { "start": 479.88, "end": 481.62, "text": "After you run Your First Simulation you're gonna Know." }, { "start": 481.84, "end": 488.24, "text": "okay i got a Warpage Issue for example But then you wanna fix it so you Want To Try Alternative Solutions." }, { "start": 489.22, "end": 490.88, "text": "That's Gonna Take Some Time Right." }, { "start": 490.96, "end": 502.0, "text": "so..you can Run A Bunch of Different Versions in Parallel shave off a little bit of time, but still every iteration is going to take on the order of hours." }, { "start": 502.6, "end": 508.42, "text": "And so that's just a consequence of the way that physics is simulated right?" }, { "start": 508.54, "end": 509.52, "text": "So what has changed there?" }, { "start": 510.3, "end": 515.9, "text": "for the first time we've now got architectures of neural networks." }, { "start": 516.539, "end": 530.18, "text": "you can train differential equations numerically, you just do feed forward through neural network which is way faster." }, { "start": 530.72, "end": 541.06, "text": "So essentially what we're building right now Is a way to predict the physics using a transformer model architecture rather than this numerical solving method." }, { "start": 541.4, "end": 542.54, "text": "Same inputs same outputs." }, { "start": 542.92, "end": 544.3, "text": "it predicts the same physical fields." }, { "start": 545.18, "end": 552.08, "text": "It Just achieves This in A different Way and The Advantage of Using Neural Networks for this Is Really Too Fault." }, { "start": 552.52, "end": 558.68, "text": "The first advantage is speed because you can run this on GPUs and in a massively parallel way, right?" }, { "start": 558.72, "end": 562.96, "text": "And then the lot of infrastructure has been built that really takes advantage at that." }, { "start": 563.32, "end": 567.32, "text": "You could get massive, massive acceleration three orders of magnitude easily." }, { "start": 567.78, "end": 568.92, "text": "so factor thousand." }, { "start": 569.76, "end": 573.92, "text": "actually internally we're up to a factor four thousand already On some cases." }, { "start": 575.38, "end": 582.12, "text": "So if he can run a simulation a thousand times faster that buys you not just time, it buys you better exploration." }, { "start": 582.2, "end": 582.32, "text": "Right?" }, { "start": 582.36, "end": 589.52, "text": "Because It means You can look at more alternative solutions in the same amount of Time and a thousand times as many is quite a lot." }, { "start": 589.6, "end": 603.44, "text": "I mean if you Just say The word A Thousand once it sounds like really fast but Like a thousand a thousand Times And then you get a feeling for just how much I mean At least looking at one block Of Manhattan versus all of Manhattan In the Same Amount." }, { "start": 605.58, "end": 613.12, "text": "So what you've shared with us is the introduction of AI into your simulation." }, { "start": 614.72, "end": 621.16, "text": "We're going to come and ask later if we can stick to injection molding or further, but in this specific field." }, { "start": 621.24, "end": 629.42, "text": "so I understand... ...and hear about maybe two concepts both agentic AI still also physics AI." }, { "start": 630.66, "end": 636.54, "text": "Can we look at what both of them do and maybe later on, are they contradicting or helping each other?" }, { "start": 637.28, "end": 640.08, "text": "Why not start first with the agentic AI." }, { "start": 640.66, "end": 651.66, "text": "How is that changing by how things have been doing until now... ...and how it's going to be done in future with Agentec" }, { "start": 651.72, "end": 651.78, "text": "AI?\"." }, { "start": 652.88, "end": 663.6, "text": "I think agentec-AI is this massive tsunami which will wash over pretty much all of human work and labor." }, { "start": 664.72, "end": 670.24, "text": "So one of the applications being in computer-aided engineering, it's going to change a lot right?" }, { "start": 670.94, "end": 672.16, "text": "It changed very fundamentally." }, { "start": 673.96, "end": 675.96, "text": "so first of all what do I mean by agentic AI?" }, { "start": 677.02, "end": 691.78, "text": "What i mean by this is using an agentic harness which Is a piece of software that uses an LLM or large language model for workflow management and decision-making." }, { "start": 692.42, "end": 695.24, "text": "And it unlocks tool use for that LLM, right?" }, { "start": 695.28, "end": 702.48, "text": "So this means your large language model can then use APIs to access other software and remote control it." }, { "start": 703.1, "end": 716.1, "text": "It has access to context from your enterprise so this could be accessed through the means of communication like email or Slack in Microsoft Teams chats." }, { "start": 716.92, "end": 724.66, "text": "It has access to your PLM where you store data from past projects that it can then reference and learn from." }, { "start": 725.1, "end": 726.02, "text": "And all of these things, right?" }, { "start": 726.06, "end": 736.82, "text": "They they all come together in this agentic harness... ...and this enables you to work with that large language model much in the same way that you would work with a human colleague." }, { "start": 737.38, "end": 743.36, "text": "So essentially instead of asking questions doing stuff yourself, right?" }, { "start": 743.62, "end": 745.3, "text": "Instead of it just being an advisor." }, { "start": 745.36, "end": 748.78, "text": "that sort of whispers in your ear and kind of gives you things to do." }, { "start": 749.84, "end": 758.06, "text": "You give it Things To Do And then It goes and does them by using Your computer In order to produce actual outputs." }, { "start": 758.72, "end": 761.02, "text": "So this has a bunch Of implications Right." }, { "start": 761.06, "end": 763.06, "text": "so number one uh...you would want to reflect." }, { "start": 763.12, "end": 770.8, "text": "This is the way That you run A software company obviously Because One of The things these AI Agents are excellent at Is writing code." }, { "start": 773.88, "end": 777.24, "text": "The pace at which that has improved is breathtaking." }, { "start": 777.78, "end": 803.44, "text": "Before, I would say October of last year we wrote about eighty percent or ninety percent of our code with human engineers and then twenty percent was auto-complete or copied over from a large language model where you sort of pasted questions into specific chunks of context to help people in particular function And you'd have to review it as a human." }, { "start": 803.66, "end": 806.66, "text": "and all of these things for quality control, because quality varied." }, { "start": 808.04, "end": 809.28, "text": "That has flipped completely." }, { "start": 809.66, "end": 819.2, "text": "I would say that we north of ninety-five percent of our code is now written by your agentic AI tools right?" }, { "start": 819.28, "end": 824.08, "text": "So like particularly Claude Code but uh... We also use codates from OpenAI." }, { "start": 826.1, "end": 828.36, "text": "They've gotten incredibly good, incredibly fast." }, { "start": 828.64, "end": 837.24, "text": "You know they have access to our repositories and they act much like a human colleague would in terms of the way you work with them right?" }, { "start": 837.28, "end": 846.6, "text": "So it's shifting the role of software developers away from writing code And into orchestrating agents." }, { "start": 847.0, "end": 851.86, "text": "It's much Like he would work With a bunch Of very talented team of junior colleagues." }, { "start": 852.6, "end": 860.4, "text": "Who need some who whom you Need To set goals and targets for, where you need some method of automating as much quality control." }, { "start": 862.38, "end": 869.82, "text": "It really increases the velocity of progress on new features by easily a factor of eight to ten." }, { "start": 870.88, "end": 871.02, "text": "Right?" }, { "start": 871.1, "end": 871.64, "text": "It's amazing!" }, { "start": 872.88, "end": 879.84, "text": "You look at roles that we have in our company right when we used to have product managers whose job it was." }, { "start": 881.34, "end": 886.78, "text": "They're sort of between the customer and product specs, right?" }, { "start": 886.84, "end": 888.1, "text": "So they talk to customers." }, { "start": 888.16, "end": 891.88, "text": "They figure out what the problems are that we should be solving." }, { "start": 892.64, "end": 900.28, "text": "They validate that were working on the right stuff And then translate it into ideas for features We could develop To address those issues." }, { "start": 902.02, "end": 907.38, "text": "In the past they would put a software developer to implement Those and show What was developed back to customers." }, { "start": 908.0, "end": 911.62, "text": "Do all That interfacing Between Product & Customer." }, { "start": 913.22, "end": 914.44, "text": "And they needed developers, right?" }, { "start": 915.5, "end": 925.92, "text": "They would work by essentially creating mock-up versions of features and use those to sort of communicate with tools like Figma." }, { "start": 927.4, "end": 927.56, "text": "Right!" }, { "start": 928.0, "end": 933.5, "text": "They'd build fake version that user interfaces then show this to customers saying is it what you want?" }, { "start": 933.7, "end": 935.64, "text": "Then say no I wanted different but better." }, { "start": 936.72, "end": 942.88, "text": "And iterate a bunch times before wasting pressure's developer capacity on features Right." }, { "start": 942.92, "end": 945.54, "text": "So, and that job has changed completely right?" }, { "start": 945.58, "end": 947.22, "text": "We no longer work with mockups." }, { "start": 947.54, "end": 959.64, "text": "we just I mean like a product manager plus quad design in cloud code who does not need the developer to implement sort of fake versions markup versions of new features." }, { "start": 959.7, "end": 962.8, "text": "would you make the actual feature In Like A Single Day?" }, { "start": 963.8, "end": 966.08, "text": "And iteration speed is gone up like massively." }, { "start": 967.92, "end": 968.52, "text": "and The roles?" }, { "start": 968.58, "end": 979.98, "text": "if like developers product managers and designers are fusing essentially to one role, which is a different you know it's-it's a different type of skill that is needed to do that." }, { "start": 980.06, "end": 980.26, "text": "well." }, { "start": 980.58, "end": 984.62, "text": "What I was going to ask is the relationship between product manager and developers?" }, { "start": 984.7, "end": 999.16, "text": "Which maybe in the past or whatever One to ten Is moving into Whatever one two but You say It's actually like everybody is becoming A. Maybe you call it something different, but is like a product manager or something who is responsible for?" }, { "start": 999.56, "end": 1001.7, "text": "A specific area of whatever." }, { "start": 1002.04, "end": 1004.48, "text": "You decide use the company want to work on" }, { "start": 1005.22, "end": 1005.54, "text": "exactly." }, { "start": 1005.72, "end": 1013.38, "text": "so I think that like in them in the medium term what this means Is The value that you contribute." }, { "start": 1013.88, "end": 1030.14, "text": "unless they're working long your deep overarching architecture type questions That the main value that we can contribute is through like super deep domain knowledge and an ability to communicate what is needed in such a way that an AI agent can implement it." }, { "start": 1032.0, "end": 1043.14, "text": "It shifts towards domain knowledge, communication skills toward customers with less technical prowess like mastery of programming languages architecture." }, { "start": 1044.0, "end": 1050.3, "text": "you can delegate those parts away You do still need the expertise but more senior level." }, { "start": 1053.24, "end": 1062.28, "text": "this amplifies the value of super senior developers who are good at quality control and orchestration, and sort of overarching architecture." }, { "start": 1062.72, "end": 1070.08, "text": "It amplifies a value with people have a good cross-section of empathy domain knowledge And it's sort of uh The attitude." }, { "start": 1070.12, "end": 1075.58, "text": "if someone likes to sort of fail forward and just run experiments You know try build it fail forward." }, { "start": 1075.96, "end": 1080.6, "text": "but that type of attitude is like extremely valuable in combination with these tools?" }, { "start": 1080.88, "end": 1087.0, "text": "We're going to be talking about organizational changes towards the end, but for a moment already." }, { "start": 1087.08, "end": 1090.3, "text": "I mean how do developers react?" }, { "start": 1090.96, "end": 1102.4, "text": "Senior developers have just started and they've learned Wherever and now they cannot, you know right if then statements or whatever." }, { "start": 1102.84, "end": 1108.84, "text": "They are supposed to be working at let's call it maybe a higher level how certain people Are very happy with that." }, { "start": 1108.94, "end": 1116.42, "text": "other people may be Very unhappy Maybe even those who have been doing this for forty years Or the ones just come from university?" }, { "start": 1116.54, "end": 1119.12, "text": "How do they react?" }, { "start": 1119.26, "end": 1121.64, "text": "Do we see is an opportunity as threat?" }, { "start": 1122.02, "end": 1122.5, "text": "Yeah, both." }, { "start": 1123.18, "end": 1125.16, "text": "I think... Both and rightfully both." }, { "start": 1125.9, "end": 1144.06, "text": "so i think it depends more on the person like on individual than it does on On The background that they have By which i mean you know some recent graduates from university are sort of not quite as polluted with old ways Of working Right?" }, { "start": 1144.14, "end": 1154.38, "text": "So They're More open Like there's like this naive experimentation That that they just bring, or they kind of go for it and then don't start with the belief this is impossible." }, { "start": 1154.44, "end": 1155.5, "text": "therefore I won't even try." }, { "start": 1155.98, "end": 1159.54, "text": "But rather let's see whether these works!" }, { "start": 1159.66, "end": 1160.54, "text": "Maybe we can do them?" }, { "start": 1161.94, "end": 1165.38, "text": "Every month passes makes me more likely to be able to actually do those things." }, { "start": 1166.04, "end": 1168.24, "text": "That has been one facet." }, { "start": 1168.7, "end": 1179.96, "text": "at the same time... The main risk you incur in this is when you wind up building stuff but isn't robust by other colleagues in the long run." }, { "start": 1180.36, "end": 1181.74, "text": "Maybe it's not production ready, right?" }, { "start": 1181.8, "end": 1191.58, "text": "It is a great demo but there are some architectural flaws that you as an expert were ill-equipped to advise Codd to avoid." }, { "start": 1192.5, "end": 1204.3, "text": "so I think there're advantages of having that level experience and depth Sort of the willingness to experiment with these tools, right?" }, { "start": 1204.34, "end": 1207.54, "text": "There is no like Gandalf out there worth ten years of experience." }, { "start": 1207.68, "end": 1221.64, "text": "With this tooling because The technology's too young and therefore I think a right way To discover sort of the jagged edge Of capability which is continuously shifting as to just work with it day-to-day And so it requires your willingness to Experiment." }, { "start": 1221.98, "end": 1222.26, "text": "chiefly" }, { "start": 1222.66, "end": 1223.08, "text": "very good." }, { "start": 1223.22, "end": 1230.72, "text": "another topic and Robert and i Just talked about the phenomenon in our last news update Is the fact that an you talk we talk ages when talking humans." }, { "start": 1230.82, "end": 1239.48, "text": "now The fact that today in DOS or any other there for user interface has been designed for surprise, surprise humans." }, { "start": 1240.14, "end": 1246.6, "text": "And now as agents take over the what I call nitty gritty mean it in a positive sense?" }, { "start": 1246.76, "end": 1249.5, "text": "I mean at details from coding." }, { "start": 1249.76, "end": 1258.02, "text": "if then statements and whatever details we've been doing Now agent does need to access you told us." }, { "start": 1258.86, "end": 1266.36, "text": "But the agent does not need to be information in graphics, in charts and whatever we humans have been looking at." }, { "start": 1266.68, "end": 1268.4, "text": "As an example on a shop floor or wherever." }, { "start": 1269.06, "end": 1270.7, "text": "what is that due too?" }, { "start": 1271.56, "end": 1280.14, "text": "The way of working means are working let's say the user interface basically off what I as human has been doing." }, { "start": 1280.32, "end": 1284.82, "text": "now the agents taking over from you with us it was do i needs differently?" }, { "start": 1284.94, "end": 1285.16, "text": "maybe" }, { "start": 1285.94, "end": 1288.0, "text": "yeah So exactly." }, { "start": 1288.3, "end": 1289.76, "text": "This is the second big chunk, right?" }, { "start": 1290.16, "end": 1294.64, "text": "One facet is like how does this change the way you work as a software company internally?" }, { "start": 1294.76, "end": 1298.8, "text": "but then their second question is How Does This Change Software per se Like for our customers?" }, { "start": 1299.04, "end": 1301.68, "text": "what will there interface be in future?" }, { "start": 1301.72, "end": 1306.2, "text": "What were they working for engineers In injection molding For example?" }, { "start": 1306.24, "end": 1307.74, "text": "it would look like in two or three years." }, { "start": 1308.68, "end": 1318.92, "text": "I think these changes are massive because As you rightly point out The ways that people have been working with computers so far has been optimized around." }, { "start": 1319.58, "end": 1324.96, "text": "basically graphical user interfaces that help people access the underlying functionality." }, { "start": 1326.16, "end": 1331.3, "text": "in a graphical way, the way that human would absorb information and input information." }, { "start": 1331.58, "end": 1333.92, "text": "Now these agents operate in different ways right?" }, { "start": 1333.96, "end": 1338.42, "text": "The best way for an agent to access software tools isn't through a graphical interface." }, { "start": 1338.48, "end": 1340.42, "text": "rather it is through programmatic access." }, { "start": 1340.64, "end": 1341.8, "text": "you need API access." }, { "start": 1341.88, "end": 1351.98, "text": "You need command line access I don't know MCP someway to essentially access you know, cursor and a click." }, { "start": 1352.34, "end": 1357.74, "text": "And so the place where human users interact with these tools is going to shift sort of one level up." }, { "start": 1358.08, "end": 1367.28, "text": "The main interface two computers for humans in the future Is Going To Be Less And Less About Graphical Tools Designed To Help You Tell Software What To Do?" }, { "start": 1367.92, "end": 1372.08, "text": "It's Gonna be More And More Conversational Interaction With Agents Right." }, { "start": 1372.16, "end": 1376.3, "text": "So Ultimately This May Be A World Where You Have One App." }, { "start": 1377.38, "end": 1378.46, "text": "That'S Whatever Your Agent Is." }, { "start": 1379.78, "end": 1387.04, "text": "graphical representation you need isn't generated by particular pieces of software, but by your agent on the fly." }, { "start": 1387.46, "end": 1391.92, "text": "You need some visualization—you just tell your agent to show you and it'll do it on a fly!" }, { "start": 1393.26, "end": 1404.3, "text": "There are some things that these agents can do this for natively... ...and then there are somethings that these agents buy themselves or unable-to-do without access to third party tools." }, { "start": 1404.86, "end": 1413.52, "text": "So in our particular area injection molding we believe that what these generalists, AI agents are great at is workflow." }, { "start": 1413.98, "end": 1415.96, "text": "And what they're not so great at his physics." }, { "start": 1416.26, "end": 1422.34, "text": "right because the physics that we are simulating our like super proprietary and niche." }, { "start": 1422.64, "end": 1431.1, "text": "there are no large publicly available training data sets that you could use to teach your agent The way that the data is structured of it." }, { "start": 1431.16, "end": 1432.32, "text": "You would need to train a model." }, { "start": 1432.96, "end": 1436.88, "text": "It'll suit it for a language as the main way to describe what's going on." }, { "start": 1437.06, "end": 1458.28, "text": "So if you're doing next token prediction in the language domain, but like your geometry is super large and described in very complicated non-languagey way some binary format And then we were predicting these physical fields In four dimensional space so three dimensions of space at one time." }, { "start": 1459.26, "end": 1462.56, "text": "These are also on order of gigabytes." }, { "start": 1463.22, "end": 1468.96, "text": "So like if you were to describe what happened at every location, at every point in time using language tokens." }, { "start": 1469.0, "end": 1473.46, "text": "that would be an extremely inefficient way of doing this and very hard to sort of." }, { "start": 1473.5, "end": 1474.92, "text": "teach the same large-language model." }, { "start": 1475.1, "end": 1480.8, "text": "That is writing haikus for you or writing C code on a particular domain right?" }, { "start": 1481.42, "end": 1483.08, "text": "Which brings me into conclusion." }, { "start": 1483.76, "end": 1495.2, "text": "I believe it's best for agents still being able answer your questions is to give it access to these more specialized tools that can predict the physics, right?" }, { "start": 1495.26, "end": 1503.78, "text": "So we think that the physics engine like... We want to provide the best physics tools for your agent through programmatic access." }, { "start": 1503.92, "end": 1509.22, "text": "Right so API access and what you're agent needs are tools." }, { "start": 1510.18, "end": 1511.56, "text": "obviously they need to be accurate." }, { "start": 1512.02, "end": 1516.52, "text": "We've got thirty-eight years of experience in calibration in how we do this." }, { "start": 1519.88, "end": 1527.32, "text": "then it needs to be fast because your agent, unlike you potentially can look at ten thousand different results and compare them." }, { "start": 1527.68, "end": 1529.64, "text": "And draw conclusions which no human could do." }, { "start": 1530.14, "end": 1535.18, "text": "but you can only exploit that if you are actually able to generate those ten thousands really rapidly." }, { "start": 1535.68, "end": 1537.3, "text": "otherwise You're not really saving time right?" }, { "start": 1537.94, "end": 1545.7, "text": "If It takes a year to generate about ten-thousand different alternative solutions you're Not gonna Actually have the speed in performance gain In The real world That you want." }, { "start": 1546.26, "end": 1561.9, "text": "so speed is important Because Speed means better exploration, then it needs to not break the bank on compute because ultimately if you're paying tens of thousands of euros for GPUs just to run the physics." }, { "start": 1562.2, "end": 1565.68, "text": "It may outstrip the savings that you get from the better engineering." }, { "start": 1566.28, "end": 1567.42, "text": "so there's a trade-off there." }, { "start": 1567.52, "end": 1571.2, "text": "and as computer efficient is possible uh is important." }, { "start": 1571.4, "end": 1582.32, "text": "and then of course it means to be feature complete which means like he wants address any type of question you have using your agent." }, { "start": 1582.42, "end": 1586.68, "text": "So therefore, an agent needs tools that can cover all the issues that you have." }, { "start": 1587.3, "end": 1588.66, "text": "so feature completeness is important." }, { "start": 1589.08, "end": 1599.02, "text": "and then of course they way interface has to be agent friendly which means it need to be a super performant API with infrastructure behind at scale as well right?" }, { "start": 1599.74, "end": 1602.98, "text": "This combination features ultimately what we are optimizing for now." }, { "start": 1603.62, "end": 1611.96, "text": "We've put both our numerical engine for a numerical engine, plus our new transformer model." }, { "start": 1612.24, "end": 1614.1, "text": "For the physics thousand times faster." }, { "start": 1614.58, "end": 1629.0, "text": "we've put that in single unified API which means you can have your agent do these two-stage workflows where it uses the AI model to do super rapid exploration of very large solution space." }, { "start": 1629.66, "end": 1634.8, "text": "try out a thousand different versions and then it can zoom on areas best performed." }, { "start": 1638.08, "end": 1645.4, "text": "do validation runs using the numerical solver without wasting compute on some optimal solutions, right?" }, { "start": 1645.46, "end": 1651.62, "text": "So basically you first explore in a fast way and then validate and confirm." }, { "start": 1652.34, "end": 1657.6, "text": "Using this thirty-eight year validated method to get the confidence that you need." }, { "start": 1657.9, "end": 1663.3, "text": "And you would all not by clicking around in a graphical interface but talking for example to Claude." }, { "start": 1664.1, "end": 1665.34, "text": "You say Claude here's apart." }, { "start": 1666.06, "end": 1669.3, "text": "Can you please help me figure out what size machine I need, right?" }, { "start": 1669.36, "end": 1672.24, "text": "And then it goes and that sets up looks at the geometry." }, { "start": 1672.74, "end": 1679.52, "text": "It runs a bunch of simulations to understand what clamping force you might be too To build this thing on there for what size Machine You Might Need Right?" }, { "start": 1679.6, "end": 1682.1, "text": "or you can say okay called will i have a warpage issue with This Thing?" }, { "start": 1682.34, "end": 1682.46, "text": "right?" }, { "start": 1682.52, "end": 1682.56, "text": "so?" }, { "start": 1682.74, "end": 1684.08, "text": "and Then it Goes in." }, { "start": 1684.32, "end": 1692.84, "text": "it know everything he knows it loads In it Runs A Bunch Simulations if Understands What Issues We Have With Default Settings That It Varies Them Etc etc." }, { "start": 1693.38, "end": 1696.28, "text": "much like Delegating to a Human Colleague." }, { "start": 1697.28, "end": 1699.5, "text": "It turns out, Klaus is a pretty good plastics engineer." }, { "start": 1700.42, "end": 1705.76, "text": "We were surprised about this actually because we didn't anticipate it being that good and it's really quite remarkably..." }, { "start": 1706.38, "end": 1711.74, "text": "By the way you mentioned Transformer couple of times as the neural network need to be a transformer type." }, { "start": 1711.84, "end": 1714.14, "text": "I happen to know some guys with different approach." }, { "start": 1714.36, "end": 1719.22, "text": "they seem very successful specifically in predicting time series as well." }, { "start": 1720.16, "end": 1723.92, "text": "So i think its less functioned architecture." }, { "start": 1726.64, "end": 1730.72, "text": "what are you teaching this architecture, right?" }, { "start": 1730.76, "end": 1736.88, "text": "So we found the main challenge actually not to be around like how do you exactly construct a neural network." }, { "start": 1737.6, "end": 1747.8, "text": "Whatever architecture you pick needs to be performant enough to capture the output side variability that comes from the input side variability it needs to able to do that." }, { "start": 1747.86, "end": 1748.78, "text": "mapping all right." }, { "start": 1748.82, "end": 1756.98, "text": "so need sufficient level of complexity the types of physics that you're trying to capture." }, { "start": 1757.78, "end": 1760.78, "text": "Once, you are confident your architecture can in principle do this?" }, { "start": 1761.02, "end": 1767.1, "text": "The main challenge becomes about data set like training dataset engineering right." }, { "start": 1767.14, "end": 1777.22, "text": "so because these neural nets don't learn physics through memorizing equations they learned it by example Right." }, { "start": 1777.28, "end": 1786.04, "text": "So You Have To Show Your Neural Network Examples Of Good Designs but you also have to show it examples of bad designs." }, { "start": 1786.8, "end": 1792.72, "text": "because like one of the main things that are trying to predict is what's going to go wrong, right?" }, { "start": 1792.78, "end": 1796.0, "text": "So if you only train on parts that actually went into production." }, { "start": 1797.0, "end": 1800.1, "text": "It will think everything goes well most time." }, { "start": 1800.8, "end": 1810.32, "text": "so you need to construct error cases and inject them in a way which takes real domain expertise." }, { "start": 1811.4, "end": 1827.42, "text": "So I think that the main challenge in these physical domains, particularly in engineering where you're trying to sort of predict and virtualize this error mode discovery and fixing is having the domain knowledge like the right intersection of pre-existing training data." }, { "start": 1827.92, "end": 1834.1, "text": "And then the right way to modify that training data to build in mistakes that in most data repositories are no longer baked in." }, { "start": 1834.84, "end": 1841.56, "text": "You have to construct it yourself with engineers who know what might happen And we have sort of the right didactic sense." }, { "start": 1841.68, "end": 1845.36, "text": "I've like, okay if what would i need to see in order To get it?" }, { "start": 1845.64, "end": 1848.38, "text": "so there's a real art form that's really baked into That?" }, { "start": 1849.68, "end": 1852.82, "text": "So you introduced two major shifts." }, { "start": 1853.04, "end": 1856.52, "text": "one We call agentic yeah The other physical AI." }, { "start": 1856.64, "end": 1864.14, "text": "now we're gonna go with little bit more details Of What You as A Company Have Been Doing In The Last Couple Years In Exactly Those Two Areas." }, { "start": 1864.22, "end": 1871.38, "text": "Now Assuming That Maybe The others see a e-vendors you know and says semen's that so all to death." }, { "start": 1871.42, "end": 1872.28, "text": "there is a couple of others." }, { "start": 1872.88, "end": 1875.04, "text": "Maybe they're doing similar things." }, { "start": 1875.5, "end": 1878.12, "text": "what does it mean toward these?" }, { "start": 1878.34, "end": 1885.98, "text": "vendors are developing how day will behave in the market, but indian is going to be their key different shader." }, { "start": 1886.08, "end": 1891.68, "text": "if i'm assuming that many of them are working in some directions" }, { "start": 1892.34, "end": 1894.22, "text": "I mean, i would be amazed if they were not." }, { "start": 1894.7, "end": 1894.96, "text": "Me too!" }, { "start": 1896.18, "end": 1900.24, "text": "We are the first in our domain to have a model of this kind." }, { "start": 1900.72, "end": 1909.3, "text": "it took us about eighteen months uh...to get where we are right now and of course we're not gonna stand still at this point." }, { "start": 1909.52, "end": 1921.84, "text": "so like First Model will train in public for the filling phase of injection molding but of course that goes through later stages of physics after you fill the cavity you know, packing and cooling." }, { "start": 1921.9, "end": 1923.62, "text": "And then shrinkage in warpage right?" }, { "start": 1923.66, "end": 1929.74, "text": "So we have those models were training them Right now scaling the training data set but We know that architecture is capable of it." }, { "start": 1930.48, "end": 1940.32, "text": "so I think That The main mode that You can Have as a company In this field isn't Isn't the idea that like Your are going to be the only one forever who has This?" }, { "start": 1941.2, "end": 1941.88, "text": "i Think that would Be naive." }, { "start": 1942.44, "end": 1950.74, "text": "um i think that the Main Thing You Can Work on Is Velocity Of Progress a one-trick pony in terms of domains and physics." }, { "start": 1950.8, "end": 1952.56, "text": "We are laser focused on this, right?" }, { "start": 1952.6, "end": 1958.9, "text": "So like there is nothing more important for us as a company than to get this right, so we're advancing the model." }, { "start": 1959.04, "end": 1969.72, "text": "We were advancing our agentic harness where putting it... It's deployed with real customers in the real world who are using this productively today that can create a flywheel effect around community." }, { "start": 1970.74, "end": 1970.86, "text": "Right!" }, { "start": 1971.5, "end": 1980.56, "text": "All of them have to learn much Like The Rest Of The Planet how to work with these agentic tools, and we can be a vector for them to experiment with this." }, { "start": 1980.92, "end": 1988.56, "text": "And understand what new processes and types of organizations laborer could look like in a broader sense right?" }, { "start": 1988.6, "end": 1997.94, "text": "So the questions our customers have at some of them are about you know the specifics of the simulation engine that we have but a lot of them also just more general about." }, { "start": 1998.04, "end": 2001.8, "text": "How do I do context management In a compliant way in my enterprise?" }, { "start": 2002.34, "end": 2010.36, "text": "How do I link this in such a way that my, you know, Claude knows how to work with information from past projects and the right way?" }, { "start": 2010.7, "end": 2016.7, "text": "How do i share information between colleagues who are using different instances of these genetic harnesses." }, { "start": 2016.98, "end": 2018.5, "text": "In any way it works etc." }, { "start": 2018.72, "end": 2018.8, "text": "etc." }, { "start": 2019.0, "end": 2025.2, "text": "Right so some of it is around The particular project but A lot Of It Is Just About This Migration Into This New Wave Working." }, { "start": 2025.6, "end": 2037.84, "text": "If You Look At These Broad Generalist Multi Product Vendors I think the main thing that differentiates us there isn't around talent or capability to ultimately do this." }, { "start": 2037.92, "end": 2039.66, "text": "It's going to be around focus, right?" }, { "start": 2039.74, "end": 2046.12, "text": "So if you were Autodesk and probably have bigger fish to fry than in particular domain." }, { "start": 2046.4, "end": 2054.199, "text": "You've got like kind of five hundred different products And one team is able build a model." }, { "start": 2054.9, "end": 2056.219, "text": "where would we deploy them first?" }, { "start": 2057.04, "end": 2062.659, "text": "so order it by business impact, ultimately for your customers probably and for yourself." }, { "start": 2063.239, "end": 2066.52, "text": "And so I doubt that this would be the first domain they will pick right?" }, { "start": 2066.56, "end": 2080.199, "text": "So i think that gives us some sort of our ability to focus like entire company on this and be nimble in user or comparatively smaller scale For a hardcore very determined pivot in its direction." }, { "start": 2080.699, "end": 2089.219, "text": "it give us speed advantage which can compound into just velocity advantage by time we start quite a bit further along." }, { "start": 2089.659, "end": 2090.739, "text": "Very good, let's take it." }, { "start": 2090.92, "end": 2094.4, "text": "step back and look at the history philosophy of your AI." }, { "start": 2094.56, "end": 2096.06, "text": "solve your API toolkit." }, { "start": 2096.139, "end": 2096.38, "text": "maybe." }, { "start": 2096.56, "end": 2098.88, "text": "since when have you been developing?" }, { "start": 2099.14, "end": 2102.66, "text": "another question where did you get training data from that?" }, { "start": 2102.74, "end": 2104.6, "text": "You used to train your model." }, { "start": 2105.42, "end": 2106.18, "text": "Yeah I mean so." }, { "start": 2106.36, "end": 2110.78, "text": "we've had this ambition to train an neural network on the physics for quite some time right?" }, { "start": 2110.82, "end": 2126.32, "text": "So i think our first initial discussions around is were in like twenty-twenty three And And we then thought hard about what's the most resource efficient way to get this deepest gradient of progress." }, { "start": 2126.64, "end": 2130.98, "text": "So how can we make progress as quickly as possible without breaking back?" }, { "start": 2131.88, "end": 2141.18, "text": "One option would have been hire a bunch of data scientists and go explore, but that seemed inefficient given at the time." }, { "start": 2141.24, "end": 2143.78, "text": "also it was unclear whether this actually feasible." }, { "start": 2144.1, "end": 2145.46, "text": "all right nobody had done." }, { "start": 2146.28, "end": 2148.58, "text": "So how would you justify the resources?" }, { "start": 2148.96, "end": 2155.1, "text": "But we knew that if successful, this will be a massive game changer in terms of capability plus business models." }, { "start": 2155.2, "end": 2155.8, "text": "it unlocks." }, { "start": 2156.34, "end": 2163.12, "text": "A customer benefit is super clear and way to use as an unique selling proposition also super clear." }, { "start": 2163.5, "end": 2182.06, "text": "And so they approach through partnership at time with ME AI who were a freshly founded start-up And so we were one of the first partners that they worked with, which gave us access to their AB-UPT architecture." }, { "start": 2182.28, "end": 2208.64, "text": "So this is a particular type of transformer model where the encoding and decoding stages are well adapted to ingesting the types of geometry that we work with together with them to our domain and co-built the first versions of this model." }, { "start": 2209.56, "end": 2224.94, "text": "This was designed as a sort of build, operate transfer type collaboration right where we built the first version's together And then took it in house afterwards so that most recent versions are fully built in house with their own resources." }, { "start": 2225.62, "end": 2236.02, "text": "Emmy has been recently acquired by Mistral as a one hundred percent acquisition, forming the nucleus of their own industrial AI team." }, { "start": 2236.38, "end": 2239.34, "text": "And so this model is now fully in-house with us." }, { "start": 2239.6, "end": 2244.32, "text": "we are fully dedicated to the development of it and deeper stages." }, { "start": 2244.44, "end": 2246.52, "text": "We've already built versions like." }, { "start": 2246.94, "end": 2252.92, "text": "there were a couple issues in the original versions that our collaboration was unable to solve." }, { "start": 2255.3, "end": 2260.36, "text": "You know, this was like an accelerator for us to sort of get going quickly." }, { "start": 2261.04, "end": 2266.22, "text": "I got the fundamental architecture bootstrapped in a super efficient and rapid way." }, { "start": 2266.4, "end": 2280.92, "text": "And now the main challenge is really around not so much architecture but data set engineering because The way that model makes progress by showing examples Of things it hasn't yet seen." }, { "start": 2281.36, "end": 2289.02, "text": "There are many different sort of special processes in injection molding that you kind of have to show it chunk by chunk, right?" }, { "start": 2289.08, "end": 2292.28, "text": "So for example there are materials that foam." }, { "start": 2292.48, "end": 2292.6, "text": "Right." }, { "start": 2292.64, "end": 2304.6, "text": "so sometimes we want foaming material because they wanna make a part lighter while still being rigid and your model won't know how deal with the materials until you've shown them some examples." }, { "start": 2306.0, "end": 2309.0, "text": "And so this means you, but it's also not the first thing." }, { "start": 2309.04, "end": 2312.6, "text": "You would train right because if you've got a model trained on is like a two percent use case Right?" }, { "start": 2312.66, "end": 2326.52, "text": "So Like you train that for us while you've gotta model addresses to present in use cases you start with the stuff That Is frequently used and then With time you kind of graduate To rarer more niche types Of applications." }, { "start": 2327.2, "end": 2336.98, "text": "um...and so uh..that sort of is The next frontier that we're working On as sort of chunk by Chunk Expanding multiple capability by expanding the training data set in those directions." }, { "start": 2337.22, "end": 2341.96, "text": "Yeah, which again is less an architectural challenge and more a domain type challenge" }, { "start": 2342.26, "end": 2342.5, "text": "right." }, { "start": 2342.62, "end": 2345.28, "text": "so where Where is then Simcon today?" }, { "start": 2345.5, "end": 2350.94, "text": "Where do you see your company moving and strongly moving and changing market?" }, { "start": 2351.04, "end": 2353.52, "text": "And maybe let's combine it with what I announced." }, { "start": 2353.56, "end": 2359.02, "text": "kind of share with us a little bit how it is and how you do rebuild." }, { "start": 2359.36, "end": 2362.84, "text": "I understand the thirty year old German middle stunt company." }, { "start": 2364.0, "end": 2367.66, "text": "two words, working around agents." }, { "start": 2368.96, "end": 2372.74, "text": "what does this job as CEO but for complete company?" }, { "start": 2372.86, "end": 2374.02, "text": "What does that include?" }, { "start": 2374.1, "end": 2374.52, "text": "implying" }, { "start": 2375.08, "end": 2385.28, "text": "once you realize your as a CEO just how profound impact will be Overall, I mean it's just crystal clear." }, { "start": 2385.44, "end": 2389.94, "text": "You have to reorient the company around this change right." }, { "start": 2390.0, "end": 2395.26, "text": "if you miss This change like they're gonna miss The most fundamental thing that's happened in software." }, { "start": 2395.62, "end": 2403.16, "text": "In as long as i can think Right because of me the implications are Are not Just about ways Of working and how?" }, { "start": 2403.26, "end": 2407.4, "text": "You build your product And It's About what the Product is right." }, { "start": 2407.48, "end": 2417.94, "text": "so the Product used To be desktop Software for humans to use through a graphical interface, monetized through seat-based pricing." }, { "start": 2418.2, "end": 2428.08, "text": "So essentially the thing you're selling is human access to a graphical Interface and then number of humans who have access determines Number Of Units You're Selling." }, { "start": 2429.42, "end": 2434.28, "text": "And that's not A sustainable business model in The Long Run." }, { "start": 2434.98, "end": 2450.3, "text": "if you believe That Most Use In The Future Will Be Intermediated By Agents most usage in the, not even that far future is not going to come from humans through a graphical interface whom you've sold a seat too." }, { "start": 2450.82, "end": 2464.14, "text": "It's gonna come through agents that your human is talking to and then numbers relation is going to be dramatically in favor of these agents because like one human can orchestrate a thousand agents right?" }, { "start": 2464.2, "end": 2467.16, "text": "So they're agent will spawn subagents who do subtasks et cetera." }, { "start": 2468.26, "end": 2481.92, "text": "it's a one-to-end type So, which means that theoretically as a company you could have like a single seat but a thousand agents sharing that seat and working through a bunch of problems." }, { "start": 2482.96, "end": 2488.86, "text": "This creates a fundamental challenge to any software business model based around monetizing seats." }, { "start": 2491.02, "end": 2496.56, "text": "Once we realize this it's important for us to think hard about what is the thing I will be selling in future?" }, { "start": 2497.12, "end": 2498.94, "text": "And so it's probably not seats." }, { "start": 2499.8, "end": 2503.32, "text": "It's more likely to be things like usage, right?" }, { "start": 2503.38, "end": 2505.42, "text": "So much like Anthropic and OpenAI." }, { "start": 2505.56, "end": 2509.72, "text": "ultimately sell tokens... Like what is the equivalent of that in our domain?" }, { "start": 2510.68, "end": 2518.1, "text": "Simulation runs.... Things like that compute hours or you need to reprise around outcomes Right!" }, { "start": 2518.28, "end": 2523.28, "text": "You could sell instead of access to a system To solve your problem yourself ...you can just sell solutions." }, { "start": 2523.76, "end": 2527.92, "text": "Give me part I'll give you a fixed I'll give you a mold, right?" }, { "start": 2527.98, "end": 2529.5, "text": "So something like that." }, { "start": 2530.86, "end": 2533.94, "text": "But every software company needs to answer the question." }, { "start": 2534.28, "end": 2537.52, "text": "so now immediately what are we aiming for?" }, { "start": 2538.32, "end": 2546.62, "text": "We made a decision to aim for being the tool of choice For your agent To be able to address this domain of injection molding." }, { "start": 2548.32, "end": 2556.18, "text": "Your agent wants to help You do design for manufacturing for plastic injection molded parts, it's going to need a tool to predict the physics." }, { "start": 2556.38, "end": 2563.84, "text": "We want provide that tool and we wanna make as simple fast convenient accurate compute efficient is possible?" }, { "start": 2564.52, "end": 2579.16, "text": "And we wanna be first really develop in super rigorous way which why built entirely new tech stack cloud based API dual engine with numerical end other engine one different pricing model around that." }, { "start": 2579.42, "end": 2586.6, "text": "We're providing it to other software vendors as an SDK so they can use it as a plug-in into their own solutions, right?" }, { "start": 2586.66, "end": 2591.72, "text": "So I don't know for example if you are in marketplace for injection molds where people buy and sell injection molds." }, { "start": 2592.34, "end": 2600.48, "text": "You may want some quick assistant To help you assess how expensive should this mold be based on the issues you'd anticipate?" }, { "start": 2600.86, "end": 2605.1, "text": "Well we could provide us super fast quick way to generate that information." }, { "start": 2605.5, "end": 2612.28, "text": "Like, or let's say that you are a CAD software company and your part designer designing the next I don't know." }, { "start": 2612.34, "end": 2619.26, "text": "BMW three series interior wants to know before he sends it to The rheology of plastics experts." }, { "start": 2619.3, "end": 2621.62, "text": "You know is this basically an okay park?" }, { "start": 2621.92, "end": 2623.2, "text": "like will they be able To work with it?" }, { "start": 2623.84, "end": 2629.28, "text": "And then in the past like you couldn't have A non-expert do This because They just Don't Know Enough About Plastics." }, { "start": 2629.86, "end": 2636.44, "text": "But if They've Got CAD plus Claude Then Potentially they can because caught is a pretty good license engineer right." }, { "start": 2636.48, "end": 2637.72, "text": "so not experts." }, { "start": 2637.78, "end": 2638.96, "text": "Can you use this functionality?" }, { "start": 2640.16, "end": 2642.36, "text": "You as plug in into other tooling that are using." }, { "start": 2642.48, "end": 2642.54, "text": "Are" }, { "start": 2642.58, "end": 2645.58, "text": "you providing this API to humans through companies?" }, { "start": 2645.8, "end": 2648.4, "text": "or I mean, it's going go do your market for agents." }, { "start": 2648.54, "end": 2652.88, "text": "how're you going to provide the solution like these two agents?" }, { "start": 2652.92, "end": 2655.56, "text": "through whatever kind of API mcp etc." }, { "start": 2655.82, "end": 2655.92, "text": "etc." }, { "start": 2657.0, "end": 2665.82, "text": "So we are providing it right now in early access to companies, um In a way that makes their agents capable of working with it." }, { "start": 2665.9, "end": 2666.0, "text": "Right?" }, { "start": 2666.04, "end": 2670.1, "text": "so ultimately We're selling Access To Their Agents But It's Company-Specific, right?" }, { "start": 2670.14, "end": 2677.02, "text": "So you need Ultimately You Still Need Credentials To Use Our API And Then These Credential Are Tied To Particular Companies." }, { "start": 2677.94, "end": 2679.16, "text": "Maybe I Was Thinking Too Far." }, { "start": 2680.46, "end": 2685.72, "text": "which I like doing is that the next step was going to be, you know just a marketplace for agents right?" }, { "start": 2686.1, "end": 2687.68, "text": "The agents they will find out themselves." }, { "start": 2687.74, "end": 2689.38, "text": "But let's stay here for them." }, { "start": 2689.44, "end": 2690.26, "text": "on very good." }, { "start": 2690.8, "end": 2694.9, "text": "and then we talked about only one very obvious thing We talked about." }, { "start": 2696.1, "end": 2698.18, "text": "maybe it's not so much That's another thing." }, { "start": 2698.6, "end": 2701.46, "text": "Robert and I discussed jobs change." }, { "start": 2701.56, "end": 2709.72, "text": "Maybe sometimes tasks change but having said What I hear you talked about is really about jobs changing very radically." }, { "start": 2709.78, "end": 2710.78, "text": "That's only one thing." }, { "start": 2711.38, "end": 2716.34, "text": "where certain people are very, very happy taken as an opportunity the other ones see there's a threat." }, { "start": 2716.42, "end": 2719.4, "text": "maybe look for another place or once come to." }, { "start": 2719.46, "end": 2725.3, "text": "you will come that at the end of this while your looking and hiring uh... Is it a central one?" }, { "start": 2725.4, "end": 2726.26, "text": "what about yourself?" }, { "start": 2726.46, "end": 2729.12, "text": "Are you eating your own dog food?" }, { "start": 2729.2, "end": 2735.12, "text": "so-to say using agents do your work." }, { "start": 2735.22, "end": 2735.66, "text": "Absolutely," }, { "start": 2736.46, "end": 2736.9, "text": "absolutely." }, { "start": 2737.04, "end": 2742.04, "text": "I mean my most important assistant right now is Claude Cowork." }, { "start": 2742.36, "end": 2744.82, "text": "i use it with basically Right Now." }, { "start": 2744.86, "end": 2750.38, "text": "I'm using Andre Karpathy's LLM Wiki method of managing its memory." }, { "start": 2750.74, "end": 2758.92, "text": "so essentially I tell Claude Cowork to maintain knowledge in a wiki like structure, right?" }, { "start": 2758.96, "end": 2764.64, "text": "So it's an index plus a bunch of markdown files that it maintains to sort of keep track of important concepts." }, { "start": 2764.68, "end": 2768.92, "text": "That I have and then I tell it to ingest new information into that wiki." }, { "start": 2769.12, "end": 2769.24, "text": "Right." }, { "start": 2769.28, "end": 2772.72, "text": "so maybe i've read an interesting article or listened Toan." }, { "start": 2772.84, "end": 2790.62, "text": "Interesting industrially I podcast with Peter And Then I Tell Claude you know can You please take the transcript and then ingest The Information onto the write sub pages inside by wiki so that and then anytime I prompt it to do any task, It sort of looks at the wiki first." }, { "start": 2790.76, "end": 2791.82, "text": "Is there any context in here?" }, { "start": 2791.86, "end": 2794.52, "text": "That is relevant to this task And only than performs a test right." }, { "start": 2795.0, "end": 2797.18, "text": "So This is pretty cool for a number reasons." }, { "start": 2797.26, "end": 2801.42, "text": "Number one It means that you're not polluting your context on an individual prompt." }, { "start": 2801.54, "end": 2818.88, "text": "because let's say that You put a PDF into your contacts For specific tasks that pdf files like three to four megabytes just bombing Your Context window or no good reason because most of that is bloat, the actual content and concepts in that PDF are like much more compressed." }, { "start": 2819.36, "end": 2822.76, "text": "In the form of text uh in a markdown file right." }, { "start": 2822.8, "end": 2830.36, "text": "so like by sort of separating the ingestion from the use you incur that token cost only once." }, { "start": 2830.64, "end": 2830.82, "text": "all right?" }, { "start": 2830.86, "end": 2839.48, "text": "The first time you show it that pdf It's going to just ingest it and put it into the right places of your wiki but then the next time You use that you retrieve That information." }, { "start": 2839.52, "end": 2851.56, "text": "its not retrieving this source PDF Its retrieving the content of that PDF summarized into the concepts you actually need for your work, which is much more token efficient." }, { "start": 2852.16, "end": 2859.62, "text": "And this also means instead writing prompts two pages long where I'm like oh yeah i'm Bastion and I am the CEO." }, { "start": 2860.24, "end": 2861.44, "text": "here's what our investors want." }, { "start": 2861.56, "end": 2871.06, "text": "Here are some things we're working on Now, please can you do x for me instead of writing it like that which I used to?" }, { "start": 2871.1, "end": 2871.66, "text": "Do now." }, { "start": 2871.74, "end": 2876.54, "text": "I could just write You know Can you please do X for me and it knows all the context because he can just look at up dynamically." }, { "start": 2877.44, "end": 2882.96, "text": "It also makes a portable right Because it means that like if i decide To work with." }, { "start": 2883.3, "end": 2891.6, "text": "you know codex because I don't know Open AI released astro or whatever And then it's just more capable For The next couple of weeks." }, { "start": 2892.52, "end": 2897.82, "text": "My memory isn't stored inside clawed my memories stored in a bunch of markdown files on my hard drive." }, { "start": 2897.9, "end": 2907.44, "text": "So I can just point any agent at that same memory, but so it makes it portable which is sort of liberates you from dependency on an individual vendor Which You've seen what?" }, { "start": 2907.86, "end": 2914.02, "text": "You're seeing the Trump administration turn off Claude Fable for awhile and given the geopolitics i think he want that flexibility." }, { "start": 2914.4, "end": 2915.42, "text": "Last time we need to jump in." }, { "start": 2915.5, "end": 2919.62, "text": "We need to finish within five minutes if not Robert's going to slap me on my fingers." }, { "start": 2920.04, "end": 2921.96, "text": "We need To stay with the hour." }, { "start": 2922.12, "end": 2924.96, "text": "two more topics And you can share It up as you wish." }, { "start": 2925.34, "end": 2931.58, "text": "We've heard a lot from you of the changes, the amazing change that I have been talking about." }, { "start": 2931.66, "end": 2937.96, "text": "And it seems to be still happening." }, { "start": 2938.66, "end": 2942.62, "text": "nevertheless Tell us what is your view on the next couple years?" }, { "start": 2942.7, "end": 2952.56, "text": "What are we going to do with engineers and how will this person be educated in terms of education for society?" }, { "start": 2954.02, "end": 2958.86, "text": "And secondly, please tell us a little bit more about your company where you based." }, { "start": 2958.94, "end": 2960.58, "text": "You know who are your colleagues?" }, { "start": 2960.74, "end": 2961.92, "text": "I see that you're hiring." }, { "start": 2962.04, "end": 2964.72, "text": "so those two topics for the next five minutes please." }, { "start": 2965.32, "end": 2969.16, "text": "all right i'll do my very best to fit it into five minutes." }, { "start": 2969.66, "end": 2974.22, "text": "yeah i think that as a species we are at an inflection point." }, { "start": 2974.62, "end": 2977.04, "text": "and what are we going to do with this?" }, { "start": 2977.08, "end": 2985.64, "text": "technology is like akin the invention of, you know or the discovery that are a fire or electricity." }, { "start": 2986.1, "end": 2993.2, "text": "This is a general purpose technology and it's in its larval stages right?" }, { "start": 2993.24, "end": 3008.48, "text": "So what we are... The pace of progress is exponential And we've got like multiple compounding exponentials multiplied That are going to hit our species In these upcoming years." }, { "start": 3008.66, "end": 3018.76, "text": "I don't think anybody on the planet has the cognitive capacity to understand what the interaction of all these things is going to mean in a long run." }, { "start": 3019.46, "end": 3027.14, "text": "We just don't know, we are gonna be unable even imagine this because too many things changing at same time right?" }, { "start": 3027.18, "end": 3040.96, "text": "So if you hold everything constant and change one thing I think your brain kinda thinks that i can predict whats gonna happen there but were changing too much for sort of valid long-term predictions, right?" }, { "start": 3041.02, "end": 3049.14, "text": "It ranges from sort of you know Star Trek to extinction which is a pretty broad spectrum of possibilities." }, { "start": 3049.48, "end": 3053.78, "text": "So I wouldn't venture to... Frankly i just don't know what's going to happen in the long run." }, { "start": 3054.54, "end": 3055.06, "text": "In The Short Run!" }, { "start": 3055.68, "end": 3057.36, "text": "...I've got a bit more visibility Right." }, { "start": 3057.4, "end": 3073.5, "text": "so I think that in the short run by Which mean couple months time horizon What we're gonna see as senior roles particularly with deep domain knowledge And judgment and the ability to instruct an orchestrate teams are going to become more valuable, right?" }, { "start": 3073.54, "end": 3086.94, "text": "So if you know how to instruct a junior colleague on average You're gonna be better at Instructing these agents because you'll have thought about what is the minimum set of information I need to give To someone in order to execute a task well." }, { "start": 3087.64, "end": 3096.24, "text": "How do i judge The quality Of What pops out those like super critical sort of properties for working with These agents?" }, { "start": 3097.26, "end": 3102.4, "text": "It's also going to be rewarding for people who have sort of a startup-ish mentality, right?" }, { "start": 3102.46, "end": 3106.92, "text": "So if you like the build stuff experiment fail forward." }, { "start": 3107.14, "end": 3116.44, "text": "You're okay with running multiple experiments at the same time four which failed and then one of which succeeds And your happy about that one not unhappy about other four." }, { "start": 3117.16, "end": 3117.4, "text": "Great!" }, { "start": 3117.54, "end": 3126.32, "text": "That is really useful quality for the upcoming couple months, you need to be experimenting with this because it's the only way to learn what these systems are capable of." }, { "start": 3126.66, "end": 3129.5, "text": "I think we're gonna see reorganization in companies." }, { "start": 3130.32, "end": 3134.24, "text": "It's going to be more slow than that technology can do Because of inertia." }, { "start": 3135.0, "end": 3137.08, "text": "Organizations are notoriously hard-to change." }, { "start": 3137.26, "end": 3147.76, "text": "People don't like to change And if we just froze technology and implemented Only What We Have Today Even then The Impact Would Be Monumental." }, { "start": 3148.2, "end": 3150.72, "text": "But it would take some time because, yeah." }, { "start": 3150.76, "end": 3152.46, "text": "It's just how humans work right?" }, { "start": 3152.5, "end": 3153.38, "text": "The diffusion... Okay" }, { "start": 3154.08, "end": 3157.02, "text": "one minute on the final point then Because we're coming there exactly." }, { "start": 3157.18, "end": 3159.6, "text": "I see that you are hiring an ML engineer." }, { "start": 3159.64, "end": 3159.88, "text": "I think" }, { "start": 3160.08, "end": 3160.22, "text": "Yes!" }, { "start": 3160.28, "end": 3160.6, "text": "That is right." }, { "start": 3160.66, "end": 3171.04, "text": "So when were expanding our AI team The ML Engineer Is mainly going to be working On sort of platform and scaling Right." }, { "start": 3171.08, "end": 3184.72, "text": "so like How do As we scale deployment of our model and tooling to more customers, like how do we ensure that it sort of scales in a cost-efficient and performant way while keeping it secure?" }, { "start": 3184.9, "end": 3186.42, "text": "Plus... That's going be one part." }, { "start": 3186.64, "end": 3187.58, "text": "so just deployment." }, { "start": 3188.5, "end": 3189.3, "text": "And production." }, { "start": 3189.68, "end": 3192.32, "text": "Another part is around the training infrastructure." }, { "start": 3192.72, "end": 3207.54, "text": "So How Do We In An Efficient Way Provide GPUs For Multiple Training Runs To Simultaneously Execute With free model choice during deployment so that you could pick which model do want to run in a flexible way." }, { "start": 3207.94, "end": 3221.4, "text": "Tasks like that are going be front and center, we want people with sort of start-up type mindset technical excellence who ideally already using these new tooling in such constructive ways." }, { "start": 3222.12, "end": 3223.36, "text": "Thank You very much!" }, { "start": 3223.42, "end": 3226.66, "text": "We almost made it only couple minutes beyond the hour." }, { "start": 3226.78, "end": 3236.26, "text": "It was real pleasure talking that we could have continued talking for not a couple of hours, which maybe you and I will be doing." }, { "start": 3236.32, "end": 3241.94, "text": "Not necessarily ours in the piece but we would see each other at what we call The AI At The Volcano." }, { "start": 3241.98, "end": 3242.5, "text": "That's" }, { "start": 3242.54, "end": 3242.7, "text": "right" }, { "start": 3243.32, "end": 3246.38, "text": "You'll be the theme captain on exactly your topic." }, { "start": 3246.56, "end": 3247.04, "text": "A.I.," }, { "start": 3247.2, "end": 3248.48, "text": "an industrial simulation." }, { "start": 3248.6, "end": 3257.56, "text": "so... Your chance dear listener if want to meet Bastian face-to-face But i'm sure If not then Bastian is happy to talk with you otherwise as well." }, { "start": 3258.22, "end": 3259.62, "text": "So Basjan, see you then." }, { "start": 3259.8, "end": 3260.22, "text": "Good luck!" }, { "start": 3260.56, "end": 3264.94, "text": "Fun with remaking of computer." }, { "start": 3265.08, "end": 3266.56, "text": "sorry added engineering." }, { "start": 3267.06, "end": 3267.52, "text": "Thank You Peter." }, { "start": 3267.88, "end": 3268.42, "text": "pleasure to be here." }, { "start": 3268.56, "end": 3268.92, "text": "See you" }, { "start": 3269.84, "end": 3271.08, "text": "soon bye-bye." } ]