Latest / Unharnessed / Unharnessed Ep. 5: Agents Are Easy. Context Is Hard. w/ PMG’s Mike Treon
Transcript
- joseph hirsch: Okay, by the way, I'm sure y'all did this this weekend, but I just was so deep in Muse and I cannot believe what Muse is doing. I have all Mike Treon: Yeah. joseph hirsch: of these new Muse feeds where Muse is like searching the web. Like Muse is the first technology I've ever used that produces a crawler scraper on demand for a consumer. Mike Treon: Yeah. I I think the b the best response I had was I've hit a captcha that asks if I'm a human, would you like me to proceed? It was like, I just need to check. And then it was like flying through like a captcha to search for a new car for me. But I I found the on Yeah. Yep. joseph hirsch: Okay, so I I have I have a new car feed. I have I have two. I have one specifically for 1970s Broncos, right? Mike Treon: Mm-hmm. joseph hirsch: And then obviously a similar one for 80s Defenders. And you can like tell it, you know, sc scan, scan daily. And what's crazy is it will crawl all of Craigslist. I mean, it's it's so similar to my thesis. It advances my thesis on ad tech a thousand X but for the open web where the primary user of the web becomes an agent instead of a human like you don't have to use the web and then here here's the craziest part right like if you're an agent and you're using the open web and you go to a website there is a non-zero chance that the website counts the view as a human shows the human an ad and then records an impression right and Mike Treon: Yeah. yeah, I'm sure. Yeah. joseph hirsch: like Because that that's what Muse is trained to emulate a human. Right? So then you have to think to yourself: well, what IBT technology vendor will be the first vendor that figures out how to classify Muse volume as its own column? Because that's equally as valuable as the human impressions. Mike Treon: Right. joseph hirsch: I mean, not equally as valuable, but like if you if you think about Amazon. Like you I'm sure y'all saw Amazon blocked Muse, right? If you think about Mike Treon: Mm-hmm. joseph hirsch: Amazon and why Amazon blocked Muse, it is detrimental to the $70 billion a year advertising business they have. Because the $70 billion relies on you going to Amazon and seeing a sponsored listing before you shop. If you're an agent and you go and buy and then you don't see the ad, the $70 billion advertising business dies. Mike Treon: Yeah. Yeah, no, there's no decisioning. Yeah. joseph hirsch: Like the agent is an existential threat to every web advertising dollar. Moffie: We have another episode of Unharnessed. We have Mike Treon from PMG, and we're gonna be talking all about the next evolution of agents. Welcome, Mike. Great to see you. Mike Treon: Thanks guys. So excited to be here. love you too. Glad you paired up and I'm s ecstatic to be joining you this morning. Moffie: All right, so obviously we had a great combo just about Muse and how agents are getting much more proactive doing things. If you give them email access, I just had the agent. I looked, I have a flight later today, and my TSA pre wasn't added to my boarding pass. Small thing. Sometimes really annoying to get it added. And I was like, Muse, get it added. And Muse's like, on it. Got it added in like three minutes. Hit me back. joseph hirsch: Okay, let's just talk about it for a second though. Did you have to credential Muse to your loyalty program in order for it to do that? Moffie: So that's what's weird. No, I didn't give it access to this was Air France. I didn't give it access. I just said, here's my flight, here's my booking number. I need TSA pre-added to my boarding pass. And it just did it. joseph hirsch: Which is wild, right? Because Mike Treon: Mm-hmm. joseph hirsch: like the probability that an agent commits a crime at this point is now a 100% likelihood, right? Mike Treon: Right. joseph hirsch: Which then begs the question like, is John I don't know who John is, but Jonathan Moffey, are you responsible for your agent's crime? Or can you like blame the agent? Moffie: I'm gonna blame Meta. I'm gonna blame Meta Mike Treon: Plausible deniability, yeah. Yeah. Moffie: if it gets to that. I should have views review the T's and C's and get back to me on my culpability versus theirs. but yeah, it is a wild world. And I think what's happening right now, and this is why we love having Mike on. Mike is the super AI pilled leader at agencies, and we had a great convo in Cannes and kind of had a snapshot of where. Agents and agentic advertising was then. And now at that time, people that saw these agent capabilities were like engineers. They saw OpenClaw and you had Grockbot more recently come out, Codex. But now I would say this is like agents for normies. Muse and Instinct are just really too really easy to set up and just get rolling and doing like all this proactive stuff for you. So, like, you know, what is that doing to the PMG roadmap, Mike? Mike Treon: yeah, I mean, I think it's hopefully reinforcing what we're doing. You know, I think we're trying to build we're very focused on context and the data that's going into whatever agents develop and trying to build the big best foundation to to build the foot on. And if you think about Muse, how quickly it is to get up and running when the first thing you're authenticating is your entire meta history and your email. Right. Like the context around me is pretty complete with an email or a social circle that I've touched. And I think we're very much trying to think about that for you know marketing and our advertisers. And long, you know, long has PMG been focused on like data cleanliness and data ownership and even semantic classification of this of media channel data. I think, you know, that. we look at as a better a better mousetrap for any agent or any workflow we're going to apply is being able to contextualize all that disparate data, especially in programmatic, where, you know, you you literally have to solve, you know, the difference that two platforms call things, you know, campaigns versus ad groups versus I.O. versus order. We we got a lot, you know, a lot of variance in the, you know, semantic layer, particularly around programmatic. So That's a big focus for us and and trying to normalize things that, you know, we can, you know, hot swap any model or any application that comes out and still be building on that great like tenancy base, you know. joseph hirsch: Okay, let's talk about building on. So what is it exactly that you have built or are building it on? I think obviously if you go to the name Unharnessed, if I was if I was Mafie, this would be this would be my talk track and I'd say, Have you built an agent harness? Although he could he could he could do it hit better himself. But what what have you built it on? What is the backbone? What is the technology? I think it has a name. I remember you telling me the name once. I think it's named like after a a Greek god this. Mike Treon: I'm not sure. Ali is the entire kind of operating system within PMG. I mean, it covers everything from planning to audiences, and you know, the big piece is around data normalization. it's all built on a tenancy architecture. And really, what we've done is kind of extended that architecture to an orchestration layer for the marketing or buy side and A lot of that's built on you know, agent core within Amazon for, you know, governance, knowledge base. we own some of the identity and gateway stuff natively within Ali, but you know, trying to basically the goal is how do we extend a agentic orchestration layer that sits between Ali, kind of a repository of you know data and applications for clients, and how we interact with a highly fragmented universe in programmatic. So, you know, from Ali, it's it's pretty simple to go one-to-one to other platforms. We have, you know, an MCP integration with Google and Meta and TikTok, where the walled gardens are a lot easier to orchestrate with natively from an application, whether you're, you know, launching campaigns or obviously pulling reporting. But when you're going into a programmatic ecosystem and we want to ensure that our Our methodology, our mantra as PMG is being like non-principle-based, kind of vendor agnostic and really focused on how do we make tech enable the best paths for the given application, allow for some choice, allow for testing, you know, things like audience discovery are never going to kind of slow down, right? We're always gonna have partners in there you want to evaluate, you want to query, you want to. aggregate and orchestrate how best to service the client. So we really wanted a system that was extensible in that way where we could keep adding and avoid this issue of you know overly complicated one to ones, right? Like any PMG, any PMG you know, user has an IDE on their desktop and can connect to an MCP in any platform. But that one to one relationship is not capture we're not capturing any of the workflows. We're not governing it. We're not building on what we had before. So really building kind of a middle where where those connections live, we can capture business logic, our workflows, but we can also capture around clients and start to aggregate all the context around clients that we have from briefs, plans, previous campaigns, decisions that other folks on the teams may have made. things that were picked up in a weekly meeting, all that being context that we can derive and and kind of put into whatever agent or whatever MCP or whatever call gets kind of generated out of that out of that layer. Moffie: That's great. You said something interesting. You said walled gardens are easier to integrate into your orchestration layer. a little triggering. So why is that? Mike Treon: Well, I think it's just you you you face less of the problems that need orchestrated, right? It's what you're you have a much simpler sample size in terms of these platforms where we can, you know, our teams are able to make a kind of multi-campaign social trafficking application pretty easily. Here's the specs from three to four platforms. Here's how I set up campaigns. It's a much easier problem to solve than thinking about how we orchestrate across. You know, I mean, currently inside PMG, we maintain about 90 separate DSP seats, right? That's advertiser-owned, PMG owned, orchestrated, you know, multiple DSPs, just that level of orchestration and understanding that takes a lot of nuance. That's not just as easy as what would be in these like social platforms. And I think we have a a heritage of working it very heavily. In search and social alongside Ali, our marketing OS, we haven't tackled it in programmatic because of how complicated, how fragmented it is, and how quickly it moves. But I think this agentic application gives us the ability to more, you know, normalize that, simplify that. And again, going back to like the hot swap is if there's something new, we're not changing out the entire framework to be able to put something in. And we don't have to redevelop any code base to add another MCP, incorporate some semantic layer. Those kind of things can be much easier. Moffie: Well, this gets to a question I have later on, but now I wanna dive into it. What can platforms in ad tech do to make it easier to work with PMG as you roll out your orchestration layer? Mike Treon: Yeah, I think I think the most beneficial relationships we're seeing now is like where we're having the conversation beyond just the proof point, you know. I think, you know, in the traditional buyer and seller agent, it it's it's great that we could transact today. We're definitely not there with with many partners because our workflows don't don't follow the here's a brief, please launch a campaign, right? The most simplistic of workflows, whether that's in a single DSP or we're talking to a publisher agent, we're trying to extend those to be much more much more indicative of our workflows, our structures, how many people are involved in that end-to-end. What does a planner want to see? What does maybe a client services person want to see? How do hands-on keyboard folks actually access all the planning information that was done when they're actually running or launching a campaign? So, you know, for the platforms, I think what's more interesting than not is like in working with them is, you know, how do we go up beyond just an MCP wrapper of APIs? What are the tools that, you know, do the two or three things that we can't do with APIs or we can't see inside a system? So this works well for things like, you know, proprietary data and insights, right? We're not, we're no longer fighting this battle of. I need all your of your of your data inside the agency walls to make a decision. No, I can actually, you know, query a platform and and give a natural language query and come back with ingestable insights, right? No longer like, don't worry about data, you guys control the governance. Or, you know, in the in the supply side, you know, the simplest one is, you know, deal management and deal troubleshooting, right? Is You're combining me three or four different systems. Is it set up right in the DSP? Did the publisher do their part? Is the SSP flowing? Those kind of like three step or three system things are very valuable in the MCP because it it allows me to say troubleshoot DLID XYZ. And and that logic's not really going to change. But you know, the the publisher or the you know, the sell side platforms have the benefit of being able to see those. three systems where where I don't have to build three separate ones. I think those are the things that we're starting to see. that's the incremental value versus just again, the we found the MCP wrappers on APIs, you know, fairly token heavy in terms of discovery and learning. It's just like anything well documented, aligned use cases that you can get implemented soon. tend to drive that and then, you know, being able to work with a partner and saying, hey, here's one thing I wish we were able to do and allow, you know, that partner to kind of build that in on their side to that spec. joseph hirsch: Alright, so you've said programmatic a Mike Treon: Okay. joseph hirsch: lot. I understand agentic pretty well. I've got a good grasp on workflows. What are you seeing on the transactional side with the agent? Are you active with an agentic transaction? How much have you spent with one? Do you see any legs here beyond like a novel approach to recreating a direct buy? Do you see it as a replacement for programmatic? Do you see new budgets from buyers who are requesting an agentick buy? Or is it something that you are doing on behalf of one of your clients because it's more efficient or more performant? Mike Treon: Yeah. Yeah. I don't think, you know, the the pure transaction activation piece is not anything more than kind of a proof point or a proof of concept that we've seen. Most of our testing is, you know, inside our buyer agent and simulated seller agent for us to understand, you know, what data are we going to want to send in those in those agentic conversations and then what comes back and how we use it. we probably have half a dozen to a dozen direct seller agent conversations that, you know, skew the line in terms of what they're gonna be capable of, whether it's buying, whether it's creating a PMP or a direct deal. So I don't think we've fully hashed out where that goes. And again, you know, my contention is trying to maintain that layer being really agile at this point is I I I think. You know, one-to-one transaction is probably the lowest common denominator, a proof point that you could easily achieve, but it isn't really representative of any workflows that we have today. the things that we're trying to bring forth is more discoverability, you know, upstream. Our, you know, our planning teams and our strategic planning and insights teams, they don't get the benefit of a lot of that visibility of, you know, reach and frequency or you know, pricing or, you know, package descriptions, some of those things that I think, you know, need to be a part of an ongoing like conversation around a campaign that may hit those agents on a number of times with a number of different personas. So I think long term, you know, we want to be able to leverage this framework to do research, you know, update, you know, and pass off direct buys and kind of like, you know, more of a transactional and and setup approach. And then I think, you know, discoverability audience side and like non-reserve side is, you know, how do I talk to publishers directly, supply shape exactly what I want, put it in a deal ID and execute it. I think there's a number of different ways that we may go about this that are a lot more complicated than, you know, show me ad products, I choose ad products, I send it back. and I think You know, we're we're gonna continue to see iteration on that the content of that conversation and the context of what's going back and forth between these agents to be able to do, you know, kind of a number of different use cases. But today we're not transacting. Everything we've done is really discovery, you know, show us what this means. Can we replace some of our, you know, pricing requests with this, right? Like, you know, I don't know how many times we get a you know, hundred row Excel that needs to be filled out with pricing, you know, it'd be great to be able to just query agents and pull that data automatically. I think there's a lot of applications like that that are just beyond the the one to one, here's a brief, I want a campaign kind of atmosphere. joseph hirsch: Yeah, the the research and pricing thing I think is particularly fascinating as it pertains to planning. You know, I Mike Treon: Mm-hmm. joseph hirsch: I see a world where, you know, when the the IAB recently came out with their framework for this kind of generalized RFP system, where I mean, if you look at what an RFP has been in the past and media proposals, there's no standardization here, right? And and so what it means is is that like You know, for a hundred agencies and a thousand brands or ten thousand, you know, brands, everyone's doing this differently. Everyone's trying to solve the same problem, which is like, what media should I buy? How much should I spend per line item? What audience should I target? Right. And I I look at that part of the funnel and I say, like, this there's a lot of value to unlock here because if you can plan media better, right, you can improve performance before you spend money. Right. And s so much so much of ad tech is about improving before or you know, like in flight improvements for I I mean that that's like the whole premise of Mike Treon: Yeah. joseph hirsch: of advertising technology is like improve as it's spending. Right. But Mike Treon: Yeah. I was gonna say I think and I think we have like we haven't such an unused level of context that we could be, you know, working with publishers to expose. It's like I kind of hate this idea of the campaign brief because like it's one it's like again, the lowest common denominator of a campaign when, you know, if you've been running with anybody for any given amount of time, we have direction on, you know, time of day and genres and You know, does app A work better than app B within a publisher, right? Especially these portfolio publishers. And I think like when looking at, you know, what the IAB did with the proposal object, it's like, okay, how do we, you know, get more context about what this actually is, see if we have historical context around performance, and then be able to go back and say, publisher, I want like 80% on this because historically it's performed better, right? And then It's on that publisher to like keep that context and say, hey, for this advertiser, we know this works better. You know, and I think that will start to level up all these things that I, you know, we continually rebuild all of our knowledge base every campaign, right? Like if a new person comes in and you got a new rep, like they're gonna be throwing the kitchen sink at you, not knowing, hey, here's the three things we've tried in the past. Here's what their result is, and that's why we're leaning towards this recommendation. I think. That's, you know, a lot more of the opportunity that is really, I think, interesting, you know, to your point, Joe, is like what what is the promise of actually having the technology do this for us or with us? Moffie: Yeah, differentiated data obviously makes agents and agent use cases way more effective. Are there like really interesting data points you're trying to get from your advertisers at scale to really differentiate your agent offering when you're doing planning, et cetera? I think that's really interesting. You could really give it a lot more context about a brand, but curious if you guys are doing anything there. Mike Treon: Yeah, I mean, you we're definitely, you know, on the outcome side, we really, you know, our the alignment of our technology is really meant to, you know, focus everything we do from a data perspective on how it's driving business outcomes and how we can test and learn our way into, you know, finding the signals, the proxies, the things that actually we know end up driving. business outcomes, whether it's you know in-store sales, credit card sales, and you know, correlating that data back to media and very granular media performance. So whether it's with pixels or Cappy like offline data or third party measurement, there's inevitably an output of you know business outcomes that align to you know programmatic tradable Aspects, whether it's a DLID, an audience segment, a, you know, an app or a time of day. And we're starting to use those, you know, not just in the way they've been used historically, which is in, you know, machine learning learning or algorithmic development, but saying, you know, how do we pose those as kind of natural language insights and then natural language inputs back to planning and activation, where, you know, again, they can be a gate or a check on. You know, any setup or just suggestions, right? Is just, you know, show me I'm making a choice, whether it's in platform or through some kind of campaign workflow, and just say, hey, just so you know, historically, this has performed five X better than this, right? Like save that now. Those are things that are just not possible from a human standpoint, or ones that we've tried to do and stopped from a let's build a knowledge base or a a wiki or a database, right? Like none of that really brings along all that context directly to where that activation or that decisioning or that kind of intelligence ends up happening. Moffie: Great. so we obviously have PMG, you have to distribute across all the channels, CTV, web, now like ChatGPT, Search Social. We just had this announcement, unified media platforms we talked to Eric Franchi about, where we have startups and even more kind of further along companies building these unified media platforms and agents that promise to distribute across all of these platforms. How does a larger agency like PMG look at these umps, these unified media platforms and how you play with them? Mike Treon: Yeah, yeah. I I definitely think, you know, early on, I think the initial take is there's there's a lot of value here. It's something that again, I think PMG sits in this like we're in like a really unique space of like, you know, big enough, but still independent and small enough. But, you know, being where we are, I think that's inevitably what we're building, right? We we want to be a unified. Technology layer for marketers to access these platforms. And a lot of, you know, even just the reporting side, we've done natively for, you know, 10 plus years on the Alley platform in terms of like just unifying that, building semantic layers, having a single cross-channel table of data that's accessible, accelerating, you know, a marketer's ability to share that data back to an MMM, have data more accessible, more granular. I can ask questions of it. I think, you know, where independent platforms are gonna it's gonna be hard to, you know, cold start the amount of like context and institutional knowledge that you want to build and gate around working with multiple platforms. I think that is that's the hard part is like we still believe that you know there is, particularly in CTV, a fragmented ecosystem that Is fragmented with value. There's a reason to go here and here. And I want to include YouTube and Amazon. How are we going to, you know, tie in the nuance of the targetability, the measurability of some platforms with others? how do we derive value out of supply path optimization? What curation brings? There's a lot of human element and media expertise, I think, that's still valuable there. That's not gonna be inherently built into. that system. and I'll be interested to see how that develops. And, you know, again, smart media operators could build on top of these things and make it easier. I think it's a bigger threat to some other software. But you know, I think that's kind of where we look at now is definitely it's the strategy we are after. I don't think, you know, we're not going to be licensing one probably directly, but we very much believe that that is a movement. that's going to be something that develops directly alongside with, you know, our own methodology and ethos around those platforms. Moffie: Very interesting. so let's talk about the team, like building out all of these agents and across all the platforms and you have this data differentiating on the advertiser side and you're connecting in everywhere. Like what does this team look like that's building all of this? Mike Treon: I mean, I think our goal here is that we're empowering kind of everybody to be on this team. And we do that by, you know, our technology team, which is, you know, for PMG, is our our product and technology is over 35% of the org. You know, we are very much a technology driven org at heart. And one of the things the team has done in in getting out ahead of governance and control and what We can actually do here was they built a white labeled IDE, you know, similar, it's you spin off VS Code or anti-gravity, but it sits on every person's desktop in the organization, regardless of who you are, right? And what they've built on top of that is a collaborative environment that sits above that, right? So skill building and management, right? I want to build a skill that references how we talk to this. you know, client, what their presentation looks like, everything from what's the PMG branding, how do I put it into a presentation, all those skills can be natively built into an environment where people can build and develop applications or you know processes around that. The other layer that we've opened is within Ali is an Ally app environment, right? So our team has basically templatized Container is, you know, a container that can live inside alley. So now you get the benefit of an entire logged in environment, client tenancy, access to data, you know, access to the native APIs that are built in. And these teams can go and build an app that does something, right? that all is a feeder organization for, you know, I think skills to app building. pure vibe coded applications to on the shelf native alley products. And I think, you know, applications are going to span that, but use cases are going to develop in every single point of that. So a lot of what we're doing is, you know, the barrier to entry for us to, you know, build and test something on a one to one basis, on a you know, multi-orchestrated basis is very low. So trying to find out what those things are that we can build on our side that really add value to these connections, again, beyond the anybody could get clawed code and put this in and have a one-to-one relationship, but like, how do we go beyond that and make it intrinsic in the organization? And that's really what happens here. everything that we've started on like the AliByer Cloud started as a vibe coded application, right? It was You know, we have a lot of knowledge about, you know, what we think is good buying strategy, good addressability strategy, all these things. Like, how would I orchestrate all of that? And it literally started as like, I may want to run a campaign in four platforms at once. How do I just set it up once and then go traffic it four times? And it's really grown beyond that to say, okay, what is the what is the orchestration that could that could happen there? And and along the way. It's really been, you know, hitting a problem, you know, every so often that acquire or that requires us to innovate or make decisions and say, you know, what are we going to do about this? Or, you know, how do we handle how do we handle different protocols? Or, you know, what if an agent changes something that we don't want to get before? Or, you know, how do we use audit logs from the DSPs to make better decisions going forward, right? Like this continual iterative nature. you know, I'll be the first to say is like, I don't know what done looks like, but to your point, like umps didn't exist last last month, you know, like things move so fast that I don't know if this will be like a process we're continually on, but I think we're doing a good job of building alongside. you know, media operators and building tools that work for them. But my number one goal is to figure out, you know, some of our great programmatic operators, like how do we capture their workflows better than a DSP does, right? Like we, you know, there's a lot of programmatic has, you know, ultimately been distilled down to what's in a DSP's UI. Right. If the UI tells me a red flag, then I, you know, I know what to do next. But yeah. Yeah. joseph hirsch: Alright, so let's let's but la last last question. So that that Memory, right? What have you done, if anything, to capture memory such that the best practices of your best trader are available to a new employee at PMG? Mike Treon: Yeah, I mean, I think what we're doing now is just the the number one thing that went into our again, tenancy based knowledge bases are anything with an audit log, right? Just pulling the entire audit log out and being able to say, you know, it answers, you know, probably 90% of the questions of like, why did my campaign do this? Right. The first place you have to look is like, okay, what changes were made? When did this start happening? I think that kind of context was was a big unlock for us. And then actually building workflows, right? Like designing a system that can know, you know, given A, what's next with B, how to actually like program that and daisy chain those kind of things to work together so that like the next thing, you know, okay, you did that, like the agent suggesting, okay, that is completed. What's next? I've been very focused on that, like double checking this and doing this and putting, you know, again, there's something to this, like two or three actions that always happen together, identifying them and accelerating them. So, like deal ID. I get a deal ID, the first thing you gotta traffic it. Then we gotta put it in our system locally. Then we want to assign it to a campaign. We may put it, we may put it in the deal list. We may put it into a certain bid level, right? All those things kind of always happen together in tandem. How do we make that a little like, you know, trafficking workflow that nobody ever has to make the mistake of trafficking a deal and not assigning it to a campaign or not setting the bid price properly? Moffie: Makes sense. so let's bring it back to Muse. It's really Muse is on the mind and these like proactive optimizations that agents are doing are really inspiring. Is there anything you guys are learning from that and something you want to incorporate into your agents, maybe going into Q four and in twenty twenty seven? Mike Treon: Yeah, I mean, I think the one thing that I think is really valuable that that Muse takes advantage of is, you know, taking complex systems and then making them natural language inputs and outputs. And I think the outputs have been very much ignored from our in our space, right? Like just in terms of optimization, you know, that one of the toughest questions to answer in a highly complex campaign and distributed systems is. What changed. And that's something agents are just very good at, right? Like, hey, heads up, this is different than it was last week. This is no longer in your campaign. Or, you know, hey, your campaign's delivering fine, but 80% of it's on one deal ID or one app. These proactive things that have context that we can say are things we don't want or things we want to be alerted to, building those things up and then being able to say, like, I I think summarizing changes that have been made, particularly valuable in the CTV space, like you know, what what why is delivery up this week? Well, we saw a huge spike in reality content this weekend, right? That, you know, being able to tie it to things that happen in the real world, you know, how spending patterns and content consumption patterns change are super insightful in tying those two things together. And providing that intelligence back to operators, back to client services, and back back to marketers who say, you know, like, how do I message this upwards? Right. That we had a slowdown this weekend. Well, it kind of seems like maybe everybody did because there was a shift in whatever those things are. I think that's what Muse is doing as I like sit here and look at alerts. You know, it's like, hey, just a just a heads up, one of your emails bounced, you know, like I don't know if I'd ever catch that or they would ever see it, but like. You whoever what you intended did not happen, right? FYI and an agent can see that maybe quicker. I think that's some of the benefit of that natural language coming out and then folks understanding what to do next is, you know, the intrinsic value we think we have from, you know, really smart media operators and and media services on top of that. Moffie: Joe, I think you're muted. joseph hirsch: Shoot a monkey. Muse reminded me that my son had a field trip this morning and he needed to go to school early. And if Muse hadn't reminded me, my son would have missed his field trip. And so this this seems like a weird forum for me to thank Mark, but like Thanks Mark. Mike Treon: Yeah. Moffie: It is a Mike Treon: That's great. Moffie: wild world we're getting into with agents. Mike, it's great to have you on another episode of Unharness. Is there anywhere you want folks to follow you? LinkedIn, are you an ex guy? What's going on? Mike Treon: yeah, I mean LinkedIn's great. Catch me in the real world, ad week. you know, I'm I'm around. but yeah, always happy to talk with you guys. Love this audience, love what we're doing and excited to be working on it, man. Moffie: Great to chat with you.