Chris Hay:  I don’t think this is a finite game i don’t think there is a winner i think this is the classic simon sync infinite game i think um the players are going to play until they run out of resources and they can no longer play the game and i think we’re going to win right i think it opens up a creator ecosystem what i hope that it breaks up is all these kind of web 2 massive companies controlling everything and we can get a a more uh surrounded marketplace.

Tim Hwang: All that and more on today’s Mixture of Experts. I’m Tim Hwang and welcome to Mixture of Experts. Each week, MOE brings together a panel of the smartest and most charming thinkers in technology to distill down what’s important in artificial intelligence. Joining us today are three incredible panelists: we’ve got Chris Hay, Distinguished Engineer; Lauren McHugh Olende, OpenEDGE Program Director, AI Open Innovation; and Volkmar Uhlig, VP Core AI and Watson X AI. So, this is our Thanksgiving episode, and we’re going to change up the format a little bit. Rather than ticking through the news of the moment, we’re going to take a step back and have a focused discussion about the bigger picture. But as always, we’ve got the headlines with...

Aili McConnon: Hi everyone, I’m Aili McConnon, a tech news writer for IBM Think. As always, I’m here to cover your top AI news of the week, but instead of running through a bunch of headlines today, we’re actually going to focus on one: the big news of the week, Anthropic’s new Claude 4.5 Opus model, which just dropped. And to do this, I’m joined by our expert, Mihai Creţu, Distinguished Engineer for Agentic AI.

Mihai Criveti: When I heard it was about the latest model from Claude, I couldn’t resist. So happy to be here.

Aili McConnon: What would you say is the most important thing that users need to know about Anthropic’s new Claude 4.5 Opus?

Mihai Criveti: I think it’s just how efficient it is with its tokens. It’s 50% more efficient than Claude Opus 4.1. So even though it’s a reasoning model, even though it can be a fairly expensive model, it’s cheaper, but it also consumes 50% fewer tokens when reasoning. So it’s one of the most efficient models out there. So don’t be afraid to use it, give it a try. I think the best way to use it is through Claude Code and leveraging those capabilities, and it really performs very well for how token-efficient it is.

Aili McConnon: Mihai, what are some of your initial reactions to the model after playing with it?

Mihai Criveti: Yeah, it’s all kind of fresh because I think it all happened like 21 hours ago, so this is all fairly new. But I’ve already been using it. I’ve been using it with both the desktop application and with Claude Code, and I can say this is by far the best model for coding, and the bar has already been set quite high. As you know, Google has released Gemini 3 Pro quite recently, and I think two, three days ago, OpenAI released GPT-5.1 Pro, which has great reasoning capabilities, but also they’ve released GPT-5.1 Codex Max, which makes their Codex agentic platform really, really good. I suspect, at least from my initial testing, that it was on par or even better than Claude Code with the previous Opus 4.1 or 4.5 Sonnet models. But now with Opus 4.5, I believe Anthropic has regained the lead in terms of the best model for coding. Still doing some initial testing on it, but it’s performing really well.

Aili McConnon: That’s super interesting. And you mentioned this comes shortly after the release of Gemini 3, and we’ve had some other big models. What do you make of the timing of Anthropic’s release? Obviously, it’s been a busy fall for big releases of coding agents. This can’t be a coincidence.

Mihai Criveti: I’m wondering if these vendors just have these models ready to go. They just might not have the best price-performance, or they’re waiting for another announcement from their competitors before they put them out to market, because the timing was really, really good. I mean, within the span of three days, we got three world-leading models for code, all outperforming each other in various benchmarks, which is quite interesting.

Aili McConnon: Are there... you mentioned three strong-performing new agents being released in quick succession. Is there any aspect of the Claude 4.5 Opus that is different? Obviously, it has slightly superior performance, but what makes this release different, from your perspective?

Mihai Criveti: I think it’s the pricing as well. They’re able to reach much better price per performance or number of tokens used than the previous Opus 4.1. I was at some point where I was just using Opus 4.1 for the planning or the more complex tasks and using Sonnet 4.5 or the previous models for the actual work, because they have a very large context window, they were cheaper, they were faster. I think they’ve been making some optimizations in terms of things like pricing and performance for Opus 4.5, and it feels you get a bit more bang for your buck than the previous versions.

Aili McConnon: And can you talk a little more about how they’ve achieved that lower pricing? Are there innovations in how they’re approaching things that have helped them to do that?

Mihai Criveti: I’m not sure it’s necessarily to do with innovation. Maybe it just has to do with more availability of the hardware. You know, they’ve recently announced some very strong partnerships with both Microsoft Azure and with Google to use more GPUs, more TPUs. I think part of it just has to do with having more availability of the infrastructure and being able to reach a bit further and say, “Hey, we’re putting some of our best models out there as the default.” I was actually quite surprised, and maybe even shocked, that when I opened up Claude Code, it recommended Opus 4.5 as the default, which is somewhat unusual. It might say something like, “Hey, we’re going to use Sonnet for thinking, and then we’re going to transition into Opus,” but outright they’ve launched a new version of Claude Code, and it used Opus 4.5 as a default.

Aili McConnon: And I guess it’s hard to predict, but how long does Anthropic have the lead? We’ve had... is it 48 hours? Will we have something new by next week? Or does this set itself apart such that you’ll be using it yourself as your top choice for at least the coming weeks?

Mihai Criveti: I think the way I see it, and I’m using all three models. I’m using Gemini for a lot of my deep research, my advisory. I’m using Claude Code with Opus for writing code or for writing test cases, and I’m using Codex with GPT for reviews or for anything else. So I’m actually using all three. But I would say, by number of tokens, it’s still the Anthropic models I’m using the most tokens with. They’re writing the vast majority of the code for me because they perform the best for this particular use case. I don’t see it necessarily as a situation where, if I had a different use case, I would still use the models from Anthropic. If it was summarization or content generation or creative writing, maybe I would lean more towards GPT-5.1 Pro. But at least in the area of code, it could also be a combination with what they put in the Claude Code tooling. It still seems to outperform Codex, at least for my use cases from my personal experience.

Aili McConnon: Do you think for that specific area of coding, does this have enterprise application or usefulness? Do you think that’s part of the play in trying to bring the cost down to make it more enterprise-friendly?

Mihai Criveti: Yeah, I think the strategy of making this available through the various hyperscalers at a reasonable cost is going to help with enterprise deployments because many enterprises are never going to consume the models directly from the provider. You’re not going to go off and consume this on OpenAI; you’re going to consume them likely on Azure from Microsoft, not necessarily directly from ChatGPT. Same thing goes with Claude; you’re going to consume it through maybe AWS Bedrock. But there has been somewhat of a limiting choice for enterprises with its availability through Microsoft Azure, for example. This has really opened it up to enterprise customers and consumers.

Aili McConnon: Is there anything else about the model that’s worth talking about that we haven’t covered yet that strikes you as interesting?

Mihai Criveti: Yeah, I think clearly they’ve optimized it for agents, they’ve optimized it for computer use, they’ve optimized it for coding. But I like the fact they’ve also optimized it for things like building PowerPoint slides, which was maybe something that even Microsoft was looking at these models for use in Office 365 or PowerPoint generation, slide generation. So they’re not looking just at software development use cases; they’re now starting to tackle a lot of other enterprise use cases: being the best model for generating a PowerPoint slide or generating a Word document or working with XML or working with the schema required to build those documents. So I’m pleased and happy to see that models are being optimized for these enterprise use cases. I’d be happy to have a model to do my PowerPoint slides as well.

Aili McConnon: Yeah, 100%. Thank you, Mihai, so much for joining our conversation.

Tim Hwang: And now we’re going to return to our special Thanksgiving episode. Happy holidays, everybody! Thanksgiving, I think, is a really good time to be talking about agents because, of course, agents have been very much hyped in 2025, but agentic commerce has been one of the things in agents that people have been excited about. And you know, this week is going to feature Thanksgiving but also importantly Black Friday, which is one of the biggest shopping moments of the year. So, I guess maybe Chris, I’ll kick it to you first. Do you think this week is going to be a breakout moment for agents in agentic commerce? Why or why not?

Chris Hay: No, I don’t think it will be. I think we’re probably another year away from that. Why not? I think all of the ingredients are getting in place. So if you think about what OpenAI’s done, they’re now bringing on board the ability to shop in their channel for commerce products, and they partnered with Shopify, etc. But there’s so many commerce retailers that they’ve not onboarded yet, so I think that’s really early, and it is US-only at the moment. And then Google has released their agent commerce protocol, and again, that’s really early at the moment. So I think agentic browsers haven’t quite taken off yet. I just think we’re about a year away from that. Now, where I do think it’s going to become relevant is utilizing web search and deep researchers from within ChatGPT to find the products that you want. That is going to be big, and that is disrupting retailers. But I don’t see a massive effect on Black Friday this year.

Tim Hwang: And what’s sort of interesting, and I’d love to parse that out a little more, is I guess, Chris, you’ve listed a couple key components. It’s almost like the agentic browser is not quite there; the partnerships are not right there from a business standpoint. I guess, Lauren, do you kind of agree with this assessment? Do you think this week is going to be big for agentic commerce? I guess Chris is almost saying there is some, in so far as people are using it to find products, but it’s not obviously what we were promised in the exciting early days of 2025.

Lauren McHugh Olende: Yeah, my feeling is it’s also not going to be so different from last year. So the protocols that Chris talked about will help in automating the actual checkout once you’re using ChatGPT and you find the thing you want; you can automate the checkout. But I’m not sure that was really ever the biggest problem. I didn’t have a big problem putting in my credit card manually once I get to the link. They spent a lot of time making it easy to spend money on the internet. I mean, we’ve had automated checkout for a while; it was very hacky. That’s why it’s hard to get concert tickets, because it is possible to build browser automation to buy things automatically. So I don’t see a big revolution coming from the simple act of being able to check out once you’re in the AI application.
And I do think, too, that even the product research capabilities are a bit underwhelming. When you’re looking for something specific, like size dimensions or a style or something, it’s not always easy to find that just through the kind of interface that we have now. I think a lot more work could be done on both training AI models on e-commerce relevant information—so, what are the input-output pairs of “this was the customer intention, output this is what they ultimately selected”—and allow the AI model to build those pattern recognitions. And then I think there’s also, when you fit those models into an agentic pattern, how do you prompt it? How do you build in the steps of that process? So, start by identifying the retailers you want to look in, then get the information you need on their products, then compare them. That’s a whole flow that using a general-purpose chatbot is not made specifically for. I think if we had more workflows that were built specifically for that, the performance could go through the roof, so when you have an intention, you can get a specific link to that product right away. So I think that’s really where I’d like to see more improvements.

Tim Hwang: Yeah, definitely. Volkmar, I’d love to bring you in because I think your angle on this is really interesting. When we talk about agents, obviously we tend to talk about higher up in the stack. It’s like the application’s not quite there; even the business partnerships are not quite there to get this to work. Is there a hardware limiter to the world of agents really taking off, particularly in commerce, or is this not almost even in the picture?

Volkmar Uhlig: So I would take a completely different stand. I think it is actually the year of agents, and the reason is very simple. If you look at Black Friday, 15 to 20% of the stuff gets returned. So I think the agents are not the consumer-facing agents, but the agents are actually the back end, and I think this is where the true adoption happens where people are returning stuff. It’s the same after Christmas. Statistics are unclear, somewhere between 15 and 25%; it depends on the product category.
So if you look at Amazon today already, if you want to return something, it used to be you click five buttons and then they’re like, “Okay, good, ship it back,” or “No, we reject it,” and you had to make a phone call. Now that’s all done through agentic workloads. So I think the big retailers are already in that motion of actually optimizing that backend flow. I do not know what they are doing when the product hits their shipping center; if they have agentic workloads there, I’m sure they do. But that first customer touchpoint, effectively doing the return, I think that’s where the majority of the labor is on their side, because the front end is very optimized. So we just don’t see it as a consumer, but we see it indirectly because actually, the return is easier.

Tim Hwang: Yeah, definitely. And I think this is a pattern I did want to talk about. Zooming out from e-commerce, I think it’s almost easy as a consumer to think agents are the dog that didn’t bark in 2025, because most of the people I know are not using agents every single day. But it does seem like on the back end, in the enterprise, there is a lot of agent activity. So it’s interesting that the public face of agents is underwhelming, whereas, Volkmar, there’s stuff running under the hood for returns that is very much identified. So we have this split screen happening in agents that might actually fool us about just how far this thing is going.

Volkmar Uhlig: And I think that if you look at programming agents, it’s pretty complicated what you need to do, and you need to coerce it into doing the right thing. So I think the consumer will always consume agents indirectly through products. We have apps on the phone where I can automate stuff. I have one automation on my iPhone which is, when I’m driving close to our community, open the gate. That’s the only automation I have. Humans typically want to have a packaged product which just solves the problem. I think the beauty of what ChatGPT did is giving you that one line everybody knows how to use from Google over 20 years, and it’s really easy to consume. Now they’re building similar to Google all these capabilities in. So I think that’s how we, as humans, will consume agents. But then in the enterprise, you take your business process, and every place you have a human, you can try to put an agent. So I think we will see adoption indirectly, but not directly.
Now, is it the year of the agents? I think we are still in the PC phase. In that sense, I just want to have a contrarian opinion. To a certain extent, it is probably the year of the agent PCs, let’s call it that way.

Tim Hwang: Yeah, pilot agents. Chris, go ahead.

Chris Hay: Yeah, no, it’s the year of the agents. I don’t know, I disagree. It’s the year of the agents; you don’t need to caveat that. It just is. I mean, we need to think about this for a second. If we look at what has happened with ChatGPT, and we’ll start from there and move outwards, integrated both in it, Claude, Gemini, you’ve got web search capabilities, which is tool calling. You’ve now got the model catalog so you can hook up things like your Jira, and pretty much anybody who’s got a service on the internet you can hook up as a connector now, and that is basically tool calling. Everybody’s offering deep researcher, which is an agent behavior. And then probably the biggest star of them all is going to be the coding agents. That just went crazy, especially things like Claude Code. If you think about things like Cursor, etc., everybody’s using coding agents to get their work done. And the biggest thing that’s made the difference there is giving access to tools.
So I think agents are here. As I said at the beginning of the year, “Year of the Super Agent,” the fact is that with planning and reasoning, these agents have become really, really capable. So it can still feel PC-like because everything’s maybe not agent-agent-agent in the way that you think, but we are all pretty much using agents every day; we’re just not thinking about them in that way.

Tim Hwang: Yeah, and I guess they’re not woven together in the cohesive experience that we’ve been promised. Chris, you began by saying, well, it’s not like Black Friday agent commerce is really going to be happening everywhere because all these pieces are still missing. They’re there, but they just haven’t been orchestrated in a certain sense.
Lauren, I always joke the agentic consumer demo is always, “You need to book a trip,” and then you push the button and the trip is booked. I guess what Volkmar is saying is that isn’t happening and might take a really long time. Do you think it eventually will? Will we get to the much more consumer-y agentic experience that is the source of all these splashy videos and startups? Or is that... Volkmar, what I heard you saying is almost like the future may not actually look a whole lot like that just because of all the things you need to package, so all the agents might always be a little bit in the background. I’m curious how far to the consumer you think the agent experience will look like.

Lauren McHugh Olende: Yeah, I think the trajectory of just LLMs standalone is a really interesting one to compare this to. LLMs: we had 2017, the transformers paper; 2018 was the year we got BERT and GPT-1; and then 2022 was when we got those things available to the end consumer in a very easy way, in the form of a web app or a mobile app. So I feel like where we are with agents is maybe that 2018—not purely research paper level, but still not 2022, not in the hands of every single person. We have our GPT-1, BERT kind of demos and things to look at. The big question is, will it take four years to get into the hands of everyone like it did with LLMs? There’s definitely reason to think that across the board these timelines are accelerating, so maybe it could be less than four years. We have way more attention and investment in this technology than we did with LLMs back in 2018. So could it be faster because of that? Or could it be longer because maybe it’s going to turn out to be a lot more complicated than getting LLMs into production and into everyone’s hands? So I do like the idea that background hype on AI makes all downstream AI things happen faster because everybody’s paying attention to it now.

Tim Hwang: And I guess, Lauren, I think this is one thing I did want to ask you about: a big part of this acceleration is whether or not it’s easy for people to develop agentic platforms, tools, applications. Do you want to give a flavor of the state of the developer ecosystem right now? I feel like that’s a critical thing. Volkmar said a moment ago that getting these to work still takes a lot of work, which limits our progress because the number of organizations and people who can actually do this is small. One way you increase progress is you make it easier for people to develop for it. Interested in how you see the developer ecosystem evolving.

Lauren McHugh Olende: I think it’s a really fun time to be a developer if you want to try and experiment, and you can. I mean, you can do that no-code. There’s things like Langflow that you can use visually to build an agent and drag and drop—that’s super cool, that helps you not waste a lot of time coding something. All the way to the pro-code: there’s LangChain, LangGraph, Crew AI, AutoGen, Semantic Kernel—I mean, your choice of things. Some are a bit more easier and abstracted to use; some give you full control if you want it. So if you want to try, you have all the tools to do that; that should never be the problem.
I think if you want to actually deploy that and take it out of a very tightly controlled environment with a very precisely specified use case—which is probably “book a trip”—if you want to expand beyond that, actually have it hosted somewhere where you could invite your friends to try it, that’s where it immediately gets very complicated, and there’s far fewer obvious options of what you’re going to use. You might have to figure out where to host the agent logic itself, which is not really LLM-type workloads, and then a separate environment to host the actual inference, and then patch those two things together. So it’s really not ideal.

Volkmar Uhlig: So I think it’s also one of the inhibitors. Right now, we don’t have this packaged all-happy solution, and it’s the entry barrier. We are not at the Shopify level, where a mom-and-pop shop can say, “Hey, I want to have an agent, and it should deal with something.” There are some projects we have in IBM where we take the flows and the business description and we’re converting that straight up from English into a Langflow. So that transformation, when we are at the point that you can actually use English to describe what problem you want to automate and not know anything about programming, I think then you get it to the masses, and then you can do it on a cell phone. It’s like, “Hey, when I come home, I want the lights to be on,” and not build automation. The interface is a baby programmer interface for people who can program, and that’s why nobody uses it. But I can describe that logic in English. Now you need to be very explicit, but I think the models can already fill in the gaps; they are smart enough for that. If you can get to a point similar to right now we are doing English-to-code, if we can get English-to-agent, then we are at a point that it’s mass-consumable. Right now, the interfaces are still built for programmers; it’s not built for consumers.

Tim Hwang: Yeah, that’s right. And I feel like that vision almost short-circuits it, which is, well, do you need a developer ecosystem for a whole set of applications? Which I think is pretty interesting.
Chris, maybe to bring you into the conversation, going to what Lauren is saying about how right now there’s lots of ways of prototyping an agent, but the minute you want to do anything more complicated or scale it, there’s just this gap. Do you have a sense of what’s necessary to mature that right now? Are we still waiting on the companies and platforms that are going to make that happen?

Chris Hay: Yeah, I think so. I mean, I think to take things from an MVP to scale is a hard problem because consumers do crazy things. So you start to have to say, “Am I putting the LLM right in front of the consumer?” And if you are at that point, then you need to guardrail it, and that could be things like guard models; it could be running deterministic flows in conjunction with the AI to keep it on track.
To Volkmar’s point about text-to-plans: if you look at something like Claude Code, if you look at something like Cursor, Windsurf, etc., almost all of these things have a built-in planner. And so when you ask a question, the first thing that happens is it goes to the planning module for anything complex, and then the model is kept to the plan. You see that; we talked about Manus early in the year—same sort of thing. You ask a task, it goes to the planning module, the planning agent kicks in, creates the plan, and then the agents execute to the plan. And there’s a good reason that exists, which is if you give an LLM agent a big list of tools, who knows what tool it’s going to pick.
My favorite one at the moment for this is the Kimi K2 model. I love the Kimi K2 model. It can call 200-300 tools; it has a long sequential range of tool calls and can do a massive amount. But you give it a tool, it’s gonna call it. It’s like every tool that it’s got, it’s like, “I will do it this way, I will do it that way.” It goes off the rails. It’s a phenomenal model, but it goes off the rails because it can’t keep itself on track. And then even when you’re executing to the plan, quite often the models will either use its own memory, or it will not even bother updating the progress. It will be like, “Oh no, I know the answer to this,” and then just answer it, as opposed to, “No, I need you to use the tool; your information isn’t enough.” And it’s like, “No, I know this, I know this,” and then gives it. And even if it does that, once it’s done the task, when you’re following a plan, you want to go: executed step, executed step, executed. And again, the model, if you’re not deterministic and you leave the model on its own, it will skip steps in the plan or not even update it. So actually, to that point about frameworks, when you want to start to get to production, that’s where those frameworks become really important. But the reality is that there’s not a lot of... you then go back to a developer mindset to be able to put those frameworks in place to deploy; they’re not out of the box. I think when we see that being a mass thing, those frameworks are either just going to be part of the platform and ecosystem you deploy your code onto, or that’s going to be solved at the model level.

Tim Hwang: So, in the last few minutes, I want to talk a little about what we’ve been talking about technically, what needs to happen for 2026 to be the real year of the agent. I’m interested a little in winners and losers and platforms here. The question is: are the winners in agent land from a platform standpoint going to be the winners in AI in general? Like, is it going to be OpenAI and Anthropic that end up dominating the agentic ecosystem? Will it be some of the cloud players that really end up doing this? Does anyone have strong priors on who’s well-positioned to be the major platform for this space?

Volkmar Uhlig: I think there are two questions to answer. One is, what model or model zoo do I need to use to actually get good results? And what Chris just said was that these things go off the rails, so you need to babysit them into giving you the right answer. I had a case where I tried to program something, and it had an API call and the API call didn’t work, and in the end the model just decided, “Oh, I’ll just stub it, I’ll just call my own function,” and it’s like, “Look, I’m done, it works,” and didn’t do anything anymore. The solution is, “Don’t do anything, and then I’m good.” It’s like, “Success, congratulations!”
So I think the whole “how do we manage the model” is a hard problem in itself. Building a model—right now, the state of the world is you need these frontier models because otherwise the reasoning capabilities are not prevalent enough. I think that probably next year we will see people just building planning models—just focus on one thing, get the planning right, and then of course the models underneath to execute the plan and not go off track. Right now, I don’t think we have done that. So the frontier models are really the only place where it can go right now, but of course with humongous cost associated with it. So we will see smaller models being specialized for planning.
I think the second question is, how do you execute this and where do you execute this? My belief is—and this is where I’m also taking our product—AI is everywhere; there is no place where AI is not. The idea that we are just putting a bunch of H100s or H200s in the data center and that’s where all the AI will happen, that’s just not true. We will see pervasive application; it will happen on your cell phone, it will happen in the data center. So the real trick here is who can make those agents cost-effective, because in the end, right now, the work is done in business scenarios by labor; there is a person who is currently doing it by hand. And what we are hoping for is that agents replace the people who do it by hand so that they can do better things. The other one is we want to have agents in the hands where right now the work doesn’t get done at all or poorly, so we want to get more choices. So that one is really now a cost optimization problem. I think there’s an industry at the bottom of, we need to efficiently run that capacity, that infrastructure, so that we bring down the cost of these agents by 10 or 100x, and if we hit that, then it will be pervasive. Right now we are using it primarily for high-value tasks which are incredibly labor-intensive, or which are very controlled so I can actually say I have thousands of people doing this, but I can put an agent behind it because it’s a confined enough problem space that I can supervise and watch it. The moment these things get more powerful and we bring the cost down, I think then it will be a more pervasive application.

Tim Hwang: Yeah, that’s right. I like seeing that market sort of dividing between the existing frontier AI model companies going more agentic, and then there’s a whole cost-efficiency universe that emerges. The frontier model companies might also get into that, but it also looks like a very different kind of market and ecosystem. Lauren, I’m curious how you divide up the future agentic market. Is it one model to rule them all? I’m sort of interested in how you forecast here.

Lauren McHugh Olende: Yeah, I think whoever can make something repeatable will win, because it really feels like this moment of agents right now is like traditional AI 10 years ago, where it was really cool that you could build an AI model to do anything, but you had to do that from scratch. You wanted to predict education outcomes, you had to find the data, train the model just for that, refine it, and then package it up and use it. And then if you wanted to do something else, you had to start over; it was a whole end-to-end process every time. And that’s kind of what agent building is right now, and it’s even more painful because it’s not just code-based, it’s language-based, so much of that rebuilding is prompting it and figuring out how to nudge it. So I feel like if there was a breakthrough with traditional AI, it was foundation models. We then trained bigger, better models because we could; we had more data, more compute, and then that one model could do different things because it knew a little bit of everything. I think if we similarly had some concept of foundation agents, it could work similarly and then reduce that friction of having to build from scratch every single time.

Tim Hwang: For sure. And you think the winners that are best positioned would be the existing leaders? They take their model, polish it off, and then it’s the foundation agentic model—orchestration of multiple models plus other constraints. I really don’t know if it would be the existing leaders, or is it going to be some dark horse that builds one agent initially to do one specific thing, but then takes the pieces of that and uses that to do a second thing and a third thing? That was kind of the AWS story: they were building for themselves to do something specific initially, but then a lot of that cloud infrastructure could be used for other things beyond that. So I think there is a scenario where that happens, where someone just commits to a use case and initially they’re kind of looked down upon because they’re using AI to just do one thing but do it well, but then they realize what the pattern is to expand to other things and eventually build something that’s more repeatable and more of a platform.

Tim Hwang: Yeah, that’s very rich. I never really thought about that as basically, what’s the specific agentic problem that if you solve, unlocks the largest number of subsequent agentic problems? It’s interesting to think about: is that the travel planning one? What is that use case? Chris, do you want to give us a final thought here before we close up?

Chris Hay: I don’t think this is a finite game. I don’t think there is a winner. I think this is the classic Simon Sinek infinite game. The players are going to play until they run out of resources and they can no longer play the game, and I think that’s what’s going to happen. So I think a lot of the technology and techniques is known; I think it’s well known across the world, and the limiting factor is resource. But in an agentic world, the models need to get smaller and smarter, and in the future, be able to fit on a chip. Therefore, I just don’t think there is a winner in this scenario.
So who do I think is going to win? I think we are going to win. I think it opens up a creator ecosystem. What I hope that it breaks up is all these kind of Web 2 massive companies controlling everything, and we can get a more surrounded marketplace. That’s what I believe in. And the biggest thing, if I think about what I think is going to happen for ‘26 and ‘27, is you remember the Rick Rubin episode where I was frantically Googling who Rick Rubin was so I could give an intelligent answer to your question, Tim. I’m obsessed with Rick Rubin at the moment because actually, I think composition is where we’re going. I think ‘26 and ‘27 is going to be about marketplace, but I think it’s going to be about being producers. You’re going to say, “Okay, I’ve got this model over here, and I’ve got my piece of data and my brand and my style and these five tools, and then I’m going to combine them together into my ecosystem, and I’m going to create something new and beautiful, and that’s going to be my product.” And so I hope that’s what happens, is that it’s not this one model or whatever is the winner—that’s a depressing future. What I’m hoping is this vibrant, amazing ecosystem and marketplace where everybody’s got a chance to use AI to improve their lives, personalize it to them, and create their own company’s products and data without the limitations that we have today. So we’re going to be the winners.
But I think these model providers, they’re going to come and go. We saw that this year. Who was Moonshot and Kimi? Ask that question six months ago. And if we go back last year, who was DeepSeek? Same sort of thing. New model providers are going to come in. You remember when we were super excited about Manus—again, I expect them to come back at some point. People are going to come in and out, and it’s fine, and it’s okay. But yeah, who knows for the future, but it’s going to be fun.

Tim Hwang: Nice. Well, on that hopeful note, I’m going to let you all get to your impending holidays. Volkmar, Lauren, Chris, thanks for joining us. And thanks to you listeners for joining us. If you enjoyed what you heard, you can get us on Apple Podcasts, Spotify, and podcast platforms everywhere. We’ll see you next week on Mixture of Experts.