Latest / Unharnessed / Unharnessed Ep. 2: Don't Ship Your (Agent) Org Chart, with Adam Epstein (Gigi)

Transcript

Unharnessed Ep. 2: Don't Ship Your (Agent) Org Chart, with Adam Epstein (Gigi) · Unharnessed

• Moffie: so that brings us to another episode of Unharnessed with Jonathan Moffey and my co-host Joe Hirsch. We cover agentic. in ad tech and today we have Adam Epstein from Gigi, a buyer agent that works on the Amazon DSP. Welcome Adam. adam: Thanks for having me, appreciate it. Moffie: so Gigi. Gee G is a she and you say you hired Gigi. Is that a decision that was really about your go to market or did it emerge? adam: a little bit of both. initially higher GG was the strategic go to market decision. We help our customers, which are predominantly agencies, cut the operational work of managing the Amazon DSP by 80%. That's the promise that we have to our customers. And by hiring Gigi, they're effectively making a choice to enhance the team members that are currently in those seats, the media managers in those seats, and enable them to be 10x media managers so that they don't need to h hire endless people and linearly scale headcount as they scale customer and revenue growth. So So we enable our agency customers by hiring Gigi to enhance their customer base and enhance the level of service that they offer against their customers. Whether or not we call it a she kind of manifested naturally. We just started calling Gigi She a lot. And we just got excited about Gigi. We developed this personal relationship to this product over time. And it's really funny. Our customers would go to us and they would say, hey, we found it really weird how you guys refer to Gigi as a she. But then as we started to use the product more and more, we started to refer to Gigi as a she. And I think that there's this natural be since there's a personification of the product that we initially have in the hiring of it, there's this natural extension in which you actually view it as a team member. And I think I think it's helpful. I think it's allowed us to build relationships, deep relationships with our customers that we wouldn't be able to do so if it was some nondescript AI agent. joseph hirsch: So let's let's talk about the interface. It sounds like most of this is conversational. If you were to describe kind of the UI, maybe even the user experience of Gigi, would you say that the primary way that people interact with the software is through natural language? Or is there also a lot of kind of traditional UX for the management of media buying? adam: There is no traditional UX to the management of media buy within Gigi. It's entirely conversational. And it's entirely conversational in two ways. So the primary mechanism with which our customers interact with the product is through what we call tasks. You've now seen tasks proliferate across really any AI ad agent or any AI agent. And the way that we work is we sit down with our customers as part of our onboarding. So let's say you're agency X. You have a playbook that you've been executing of how you manage the Amazon TSP. How you build campaigns, how you how you optimize campaigns, how you report on campaigns, and those playbooks are historically written for humans constrained by human labor. And so what we do with our customers is we actually deconstruct those playbooks and say, like, what if you could rewrite these playbooks in a manner that's totally unconstrained by human labor? totally with AI always on and always executing using proprietary ML models as tools that it can use. And so we get them to execute these tasks. And so tasks would be a collection of campaigns against a given advertiser and Gigi would sort through all of the permutations of optimizations and opportunities and ways to better achieve the goal that the customer has told us in in the form of tasks. And then the the job of The media manager itself is no longer actually doing the work, but rather it's reviewing Gigi's work in a con chronological feed across all of the advertisers that they have. These are the things that I've told you to automate. These are the things that I need human approval on, accept, go at go edit it, take action against it. And so that's the primary mechanism that our customers interact with our product is tasks. The other mechanism is obviously chat. So there's a chat UI that people come into our home screen and chats are typically done when there's ad hoc tasks and actions that you want GG to take, whereas tasks itself are more routines that you would do on a defined cadence. So something like building an ad hoc campaign would be done in a chat. And building a campaign is a great example. So the other part of our product is how we get our customers to contribute to their GG harness. So one of the things that I like to say is with agentic AI, we no longer buy buy software, we hire software, we train it. We manage it and we onboard it just like we would team member. And it's our job to give our customers the tools to do exactly that. And so we do that by allowing them to customize their GG agent to the way in which their they work and the way in which they operate. So for campaign builds, we have something like agentic operating procedures. And so you guys have been working in ITEC long enough to know that agencies will literally have a very prescribed way of how they name campaigns, name line items, name audiences, or creatives. It's often a very robust, sophisticated. language and strategic ways in which they break them down across line items given the media it is, whether it's display or whether it's CTB and how you might do it based on inventory or audience, etc. Then each advertiser might have its own context about how they build campaigns. And so the reason why I say all of that is because we allow people to build campaigns in a single prompt. So you can go into Gigi and say, like, hey, I'd like to build a retargeting campaign against this product line and natural language. And GG will leverage all of that context and harness collaboration that the customer has provided so that Gigi can build those campaigns in a manner exactly the way the customer wants. And so getting back to your initial question, Joe, you UI and the UX is totally untraditional. it's it's completely a paradigm shifting way in which people interact with software for media buying, but the primary mechanism with which they do so is in building tasks, reviewing and accepting tasks, and then ad hoc chats. joseph hirsch: All right, what do you think is the difference between a task and an agent? Because I I look at the space and I I come I come from a a very similar angle, an operational background where, you know, I've analyzed many of what I think we universally understand as a playbook, right? And a a playbook adam: Sure. joseph hirsch: is somewhat routine in that, you know, you can follow it repetitively and God willing it gives you the same output every time you follow it. Would you say adam: Yeah. joseph hirsch: that certain certain tasks are agents in and of themselves, or do you think that the the repetition of a rote task is not an agencork workflow, it is something lesser, it is a task. adam: Yeah, I get what that question that you're asking, and I think it it really flows to how you build your product. But one of the things that I like to say is don't ship your agentic org chart. So one of the things that has always existed in building software products is as a bad thing is you ship your org chart, in which the product that you build is entirely dependent on the teams that were building it. And so you might have competing teams that ultimately build two of the exact same products in a enterprise SaaS solution. And it's like, where do these things fit? And I think one of the things that we have d done a disservice in the industry of building with agentic AI is that companies will sh try and ship their agentic org chart and say that we have this agentic orchestration in which you're now hiring 30 different sub agents to do hundred different things and you now get this team of many many agents but like no one wants a team of many many agents people just want their problem solved in the simplest way possible and so whether or not that there are sub-agents performing a specific task or whether we choose to have GG deterministically write code so that we are able to provide more consistent less compute and token intensive workflows on those same tasks shouldn't be Something that our customers see, we're not going to ship our org chart, but we do have sub-agents that will perform some tasks and we will have just a standard LLM perform another, or Gigi will write code to execute some of the others as well. but but again, our our perspective on this is that you're hiring Gigi to do the work and you just need to trust that Gigi is gonna do the work regardless of the engineering infrastructure behind that decision. joseph hirsch: Yeah, so I don't think the the market really has any understanding of a fundamental concept that you mentioned, the idea of writing code. but you know, my product does this, a lot of agentic products do this in the background. In real time you might compose kind of a a script, right? in order to execute something that, you know, theoretically is deterministic, right? You don't adam: Yes. joseph hirsch: need the agent inventing the way to do this. You need a Kind of a dumb a dumb answer to a smart problem. it is your client base ever exposed to code? Do you have to know how to use anything? Like, is it can you even see the code in the background? If there's a script in the back executing a routine task, does the user even know? adam: No, we totally obfuscate that to our customers. Again, the customer just knows that Gigi's performing the work that they assigned to her. joseph hirsch: Fascinating. Moffie: Yeah, you said something interesting. You said our buzzword harness a few times and this is unharnessed. And I believe you said your customers can have their own custom harness. Is there a lot of work and like upkeep for a customer of G G to kind of customize their harness? adam: So the short answer is actually yes. we say again that we cut the operational cost of managing this media channel by 80%. But in actually onboarding Gigi and orchestrating GG, we get we ensure that every single agency or brand that we onboard assigns a GG owner. And the singular success metric as to whether or not we are deployed successfully across the enterprise is the engagement of that owner. And GG is a non-trivial amount of incremental work for that owner to the benefit of the team, the company, and their clients. So that owner is responsible for collaborating with our team to build those tasks, to set up agentic operating procedures, to set up context for every single advertiser. And we're trying to make it easier. Like the feedback that we've gotten is actually too onerous of an onboarding for the GG owner, particularly for a scaled agency that's managing hundreds. brands across dozens of traders. And so we're trying to make that easier. But in building that harness for that individual customer and relying on it to iterate the output as we improve the product and iterate the performance so that each media manager is achieving exactly what they want to achieve with GG takes work. But the thing that we've done is we've actually gone to agency leadership and we say like hey we we need to start flipping The script on how to position this to the GG owner. Like this shouldn't be viewed as a burden of incremental work, but rather this person is now given the gift of building what's probably the most important skill that anyone in the workforce can have over the next decade, which is training, managing, and deploying an enterprise AI agent. And this person should be rewarded in every way possible. And so we do that on our end. Like we send the owner merch. we try and really celebrate our GG owners, but it's really incumbent on enterprise agency leadership to carve out incremental time for this person to own Gigi in the same way that they would a new hire. Like if you were to have a senior, this is what we say to our customers, we say like, hey, if you were to hire a senior media manager tomorrow and you want to ensure that person's success, the person that they would report to is likely spending five to 10% of their time onboarding that person to make sure that they are as successful as possible in that role. And w it's no different when the an agency chooses to hire Gigi as well. Moffie: I love that you admitted that there is some friction. I was just listening to the PM of Grokbot that is like taking the world by storm. It was on Lenny's podcast. It's like Tech Bro podcast. And it was talking about like how it grew so viral so quickly. And he talked about how they picked 200 customers and they did really tedious onboarding with those customers, some influencers, so that once they either liked or didn't like Grokbot, they would kind of like promote it on X. And then some at like enterprises. But there's real friction. These are engineers they're going to say switch from like codex and cloud cowork that is like already taking the world by storm. And there's some lock-in there and context. So they needed to go super deep and do things that aren't really scalable with their first customers to get them to onboard the GrockBot properly. And now it's like the hottest thing right now. So I really like that you actually admitted that it's, you know. some friction is actually good because it's a pretty big change to get them out of the like just original Amazon DSP UI. adam: Yeah, and and we're we're not turning our customers from one agent to another, right? We're actually getting them to build a new skill. And there it's there's a difference between going from codec to cloud code, but you're still effectively doing the same thing. It's just a different harness that you're working with every single day. Whereas for many of our customers and and many people in the enterprise today, they're still using LLMs as a glorified search engine, right? people don't know about agentic orchestration, people don't know about context concepts like tool calling or context building. many of them aren't the ones that are vibe coding apps and cloud code to make enterprise more efficient. So we're we're bringing them along on this journey. They're raising their hand and saying like I want to be the owner or they're getting assigned this tasks and being empowered to be that owner. But it it is a non-trivial amount of work. Moffi, I I don't think that we should call it friction. I don't I don't like the word friction. You asked if it is it is it work it's it's very very valuable work that drives a lot of value to the enterprise, not necessarily friction. joseph hirsch: Alright, let's go back to the onboarding Moffie: No. Good call. joseph hirsch: thing, because adam: Sure. joseph hirsch: I've I feel like onboarding is probably not the sexiest topic, but inevitably people fall into this change management camp because they're nervous about their job. adam: Yeah. joseph hirsch: I think people don't yet understand does this replace me? Does it adam: Yeah. joseph hirsch: you know, you mentioned the 10x operator, I'm using similar adam: Yeah. joseph hirsch: language. adam: Yeah. joseph hirsch: and do you I I'm curious, do you do you forward deploy people inside of the agency to teach them how to use it? How do you how do you ensure that the onboarding process doesn't fail? I mean, especially for the r recording of what you call tasks. I'm curious adam: Mm-hmm. joseph hirsch: how you are able to accurately transcribe certain ad operations, whether you're taking a video doing a screen recording, if you have an account manager or client services who are uniquely responsible for coaching someone in how to record a workflow and then you're taking that and you're showing them how to use the workflow plus natural language to create a task. I'm adam: Yeah. joseph hirsch: just I there's there's a lot of there's a lot here that I think adam: Yeah, for sure. So so there's a a mix of totally human in the loop onboarding and some a mix of automation. So whenever we onboard a new customer, we authenticate into their Amazon DSP seed and we're able to see all of the actions that they've taken across order and line item history. so we don't get context of why those actions are taken, but Gigi immediately understands the work that they do prior to Gigi. And so we we built an onboarding agent, not to ship our Internal agentec org chart, but we built an onboarding agent that monitors all of this and is allowing our account management team to build out tasks. additionally, we sit down with our customers and we understand the decks that they prepare for their clients. So we see the way that they report the metrics that matter for them. we we understand these are what GG is uniquely good at. So we'll sit with them and say, like, hey, we notice that you're not doing these things, but we know that Gigi is uniquely. good at it and most of our customers see value against this. So would you like these tax built into your GG agent? And that's how it manifests itself. And then ultimately the GG owner will give a thumbs up to the task, but ultimately it's up to the media manager to review the output of the tasks. And it's a bug if the media manager is not accepting Gigi's work. Like there's no reason for a media manager to not accept Gigi's work because we've we've I already demonstrated that these are objectively good advertising decisions based on your best practices that are unique to you all. And so if you're uncomfortable taking accepting Gigi's work, then we need to work. with you to rewrite the task in a manner that the output is exactly the way that you want. And so that's a constant iteration. and so so so that's an interesting thing. But getting back to onboarding, you know, every every company I think is all obviously in in a search for product market fit. And I was listening to a podcast from some venture investor recently that was saying like product market fit isn't a matter of stickiness, it's not a matter of revenue, it's not a matter of slope and logos. it's whether you can scientifically say X number of users perform Y action that shows immediate time to value within this time period. And so for us, we actually thought that task acceptance and task acceptance rate was that product market fit signal for our customers. but we were wrong. And so we do a lot of the work of building the tasks. And it's so low lift for a media manager to just review Gigi's work and accept. And so I just said that the two primary mechanisms with which our customers interact with our product is task and chat. And so our hypothesis that task acceptance was a signal of product market fit was actually wrong. What was a signal of product market fit was when people assigned Gigi work in a task and GG took action. And the reason why that's different. Is because the customers needed to come up with the ingenuity of writing the ad hoc prompt for GG to take action and then accept Gigi's work rather than our team being the one to say, this was your best practice, just review and accept. And so it wasn't until the customer was able to seed themselves with the actions that they want Gigi to take, write the prompt themselves, and then accept actions against that work where we we now use that as our signal for product market fit. So it's really interesting like Moffie: Love it. adam: like we we have to balance the white glove onboarding with I don't want to use the word friction but but balancing the white glove onboarding while also enabling our customers to have be an author in the output of the agent so that they can become comfortable and see the magic of accepting the work of the agent. Moffie: So it's obvious you've gone super deep with your agent on top of the Amazon DSP and you've blogged about this approach of going really, really deep, making your agent the best possible agent on top of Amazon. talk to me more about that. As a moat versus we know there's a lot of startups that also go very wide and claim or, you know, work omni channel across multiple platforms. Talk more about that. adam: Sure. So so we started on Amazon for a few reasons. across our 12-person team, we have over 30 plus years of Amazon ad experience. Amazon DSP has a lot of natural industry tailwinds. There's not really any competition. There's a lot of third-party tech on Amazon-sponsored ads. No one really touches the Amazon DSP. So that's why we started here. And we launched this product literally a year ago, September 2025. And admittedly, we had some customer churn in those first few months. And if you think about what Gigi's capabilities are, Gigi's capability At the lowest level, at the like simplest, most rote level, are the tools, the tool calling orchestration that she has access to within Amazon AdTech. And just by being a nascent company a year ago, the capabilities that Gigi had were rather limited. And we actually had a PDF. We would say, like, hey, these are Gigi's capabilities right now. These are the things that Gigi can't do. And we're trying to close that gap. But just as an FYI, you can't assign Gigi this work yet. And We had churn. Our our customers had to have this mental bandwidth of saying, like, hey, can Gigi do this? Can Gigi do that? And then ultimately they were just like, Why don't I just do the work? Cause I'm not really sure if Gigi can actually do the things that I'm hoping to hire her for. And and it wasn't until that we got rid of the PDF, and it wasn't until our customers said to us, I can assign Gigi all of the work that I would do a media manager, were we actually rewarded with customer revenue. And so we closed that gap and Q1 of this year. And if you look at our managed spend and growth trajectory, it's it's as the charts that you would want to see. And so just very naturally, we heard from our customers that in order for them to hire Gigi, in order for them to reward Gigi and the Gigi team with their revenue and their adoption of our product, we needed to go what we call a mile deep in the Amazon DSP and cover every possible use case within Amazon ad tech that they would want a media manager to cover on their own. And we're now only Moffie: Would you have adam: Only a year plus in, a year post-launch. so we are building for a year and then a year post-launch that we think that we're at a level of product maturation on Amazon DSP that we can start to expand to new media channels. Moffie: I'm just curious, would do you think you would have gotten to the place where Gigi can do the full job by shipping maybe too soon where it couldn't do the full job and learning, or you should you have just built it all and then tried to ship it? adam: I don't know. I I I think that you should always ship as fast as you possibly can and demonstrate value as possible as fast as you can. Look, we we had some churn, but we have good relationship with those customers and some of them have come back as of like Q two of this year because they still had the same problem that they wanted Gigi to solve just in Q three twenty twenty-five, the promise of Gigi didn't meet the reality of Gigi. And now that we're able to get there, we're able to grow with our customers. And it's not like we didn't have any customers before, we just went through a cup a little bit of churn, but like you guys have built companies before, but companies haven't gone through the trough of sorrow immediately after launch and with has customer churn. Like that learning is invaluable. and so we got through it. It was dark. churn sucks. It really is painful. but we got through it and we're doing well now. joseph hirsch: Quickly, what do you what do you see on the performance side? Is Gigi improving the return on ad spend for the end advertiser? Is the ROI as a byproduct for the agency client the primary benefactor? Or is the agency the benefactor because of the productivity boost or the efficiency gain from less heads doing more work? adam: to answer your question directly, Gigi improves advertising performance almost immediately across every single advertiser that it touches. And Gigi doesn't do that because there's some black box AI model, but it's actually a lot simpler than that. Again, if we go back to this, you would have our customers with a playbook of this is how you optimize Amazon DSP campaigns. And that playbook would be seldomly executed by humans, constrained by human labor and the arithmetic of a human. And so now we can go to our customers. And say Gigi's always on and always executing and using ML models. And so we actually worked with a publication in the industry called StrataBly. And so StrataBly identified that 73% of brands are not optimizing their Amazon DSP campaigns on a cadence quicker than monthly, if at joseph hirsch: Mm. adam: all. It's just set it and forget it, like spend budget, hopefully get the best results. And our agencies are now able to optimize daily. So let's just use like one of those optimizations as an example. So running an inventory report, seeing what domains or apps are driving outcomes, or rather, what domains or apps are driving a lot of impressions but zero outcomes, and then blocking those domains or apps is like the simplest way to show Gigi's effect. And so now people are running inventory reports on whatever parameters they set, X number of impressions without Y number of outcomes, and are able to do that daily and automate those exclusions. And Gigi runs a report once a week that can say, hey, brand, We just blocked X number of dollars and wasted spend by blocking these domains this week. That's a really, really quick example of the many, many optimizations that GG is now able to execute for our customers. So it's time savings, yes, it's immediate performance improvements, but also it's an elevated level of client service, right? Like that brand was not getting that report weekly from their agency report before. And every agency knows that that weekly client caller, that weekly client deck is a theatrical. Moment for them to demonstrate expertise and value. And Gigi now enables them to tell a story of these are the changes that I'm making, these are the effect of those changes and why I made them every single week in a way that they weren't able to do so before. And so not only is the brand getting a better performance, but they're happier with their agency and investing more. So one of the agencies that we had in our pilot, the the media managers actually had to build a case for Gigi to leadership at the end of the pilot. And so the the the most demonstrable aspect of the case was they said this brand increased their monthly budget by $300,000. mid-pilot because of the elevated level of service and improved performance that GG brought. And so we we charge that agency $30,000 a month to use our product. And so immediately from that single brand alone, GG payback itself in fees. And so performance benefits are definitely one of the things that Gigi brings all of our customers immediately. joseph hirsch: Great. That's that's what everybody wants to hear. I think performance seems to be I I think the money follows the performance. And I I think your anecdote exhibits it very well. I think people increasingly are chasing ROI in a way that they haven't been able to in the past. I mean, there are automatic ROI generating products in market, whether it's through Meta or Google, and I think the open internet or ad tech in general is is always left answering the question, how are you going to provide the same performance that is available in a walled garden on the open internet? And I think that the startups that can answer that question are uniquely positioned to benefit. And I think the startups that can't are kind of just not. so it's it's always I think it's always critical to understand what is what is gonna move the needle For the return of the end dollar. Like is this dollar going to return twenty-two percent more as a byproduct? And if the answer is yes, I think you're in a good spot. adam: Yeah. go ahead. No, that's it. Moffie: Adam, can we go ahead. I was gonna say can we break some news here? Are you looking at some point expanding beyond Amazon with G G? adam: Yeah, we're we're like I don't know if this is news, but we like if you think about the people that use our product today, it's programmatic media teams at enterprise agencies and brands. And so our customers have effectively said to us, Hey, like I can't make B P N L bending decisions if you're just on one media channel that this team operates in. So very naturally we say, Hey, where would you like us to go next? And you guys can get take guesses of where these programmatic teams log into and the surface areas with which they operate. And so that's the the the TAM expansion that we're going after right now is these budgets, programmatic media budgets, and these teams, these teams are are the places that we'll quickly expand to after Amazon. Moffie: All right, let's get a little nerdy about that then. So when you make Gigi omnichannel and you expand beyond Amazon, are you looking to adopt a protocol like an ad CP? Are you in to add CP? Are you down with ad CP at all? Or is it something where you can just plug into another platform's MCP server or APIs and execute omnichannel spend? adam: Yeah. I I don't I don't know where you guys are at on the ad C P spectrum. Joe, I imagine that you're much heavier leaned in than I am, but our our customers are not asking about ad CP. like when I just told you we ask our customers where what surface areas they want us to live in. surface areas that they want us to live in are other enterprise DSPs that their teams are buying a lot of media against. And so we will leverage the tool set that's afforded to us by those DSPs and those media channels to provide the best possible GG experience. Whether that be joseph hirsch: Yeah, I think adam: go ahead. joseph hirsch: I think this is critical. The i it's very similar to the question that I'd asked about does the user ever see the code? And and I think the user is passionately unaware and uncaring of the how. And I think if you burden the user with the how, they just reject you. Because the the peer groups that we're operating with are not trying to educate the user on a new way of connecting an agent to an ad platform, the ad operator, the media buyer, the trader, they couldn't care less. They don't care about standards. They just care about is my job easier and am I improving performance? And and I think to i I Moffia, I'm I'm equally interested in the nitty-gritty of of how someone might be executing agentically across an ad platform, but I think for the most part the the client itself doesn't care. They they're it's not their responsibility to know. And even if it was, why their their opinion I feel like is kind of like, well, you you can satisfy me, so my opinion is satisfied. Moffie: Love it. so Adam, I've discovered you because your weekly blog post goes super deep on things like a harness, where it's really easy to tell if someone really knows that they're talking about building agents in ad tech when you write it down weekly. So I was super excited to have you on. Tell us more about the journey of writing that weekly as a founder and what's next. adam: Sure. So at the beginning of this calendar year, our marketing lead said to me we need to do more fanber led marketing. And she said, why don't you start a weekly newsletter and start posting on TikTok? So I started writing a weekly newsletter and I started post posting on TikTok three times a week. And I would go to industry events and everyone at the industry events would say to me, I'm really enjoying your newsletter. Like I I really learn a lot about your business. I learn about a lot about the industry. And so that was great. And then I would go to my daughter's swim class, and some parents would say, I see that you're posting on TikTok. And no one would say in the industry that I'm seeing your TikTok. So obviously the audience of one was better than the other. And so I kept on writing the newsletter. And so basically the newsletter is just unfiltered thoughts of how we think about building our business, how we think about building products in Agentec AI. And I sit down every Monday at from seven p.m. EST to 8 p.m. EST and write a post. It's on Cherry called Cherry Picked. So GG, our logo is a cherry. It's called Cherry Picked. I think it's cherrypick.gg co.ai. And we have a a weekly subscription with a almost 50% open rate. Moffi, you enjoy it. Other folks in industry enjoy it as well. I think 25% of our readers are actually Amazon.com email addresses so that's an interesting one as well but I take a lot of pride in it. it's definitely not AI slop and it's been a non-trivial amount of work from the beginning of January to now to write that weekly newsletter. Moffie: Tease us what's queued up next. adam: So the next one is kind of what you just mentioned. the the title of the next post is a post called Don't Wait for the Protocol. And so I was in LA last week chatting with an agency exec, in which I mentioned that we built a browser use GG agent. So for some of the actions that our customers want GG to take, there isn't an API or MCP available to take those actions. And we can't say to our customers, hey, like We need to wait for the API. The API doesn't exist. They just, again, mile deep on Amazon DSP, they just want to assign Gigi the work. They don't really care how the work gets done so long as it gets done. And so we built a browser use agents that can press buttons within Amazon's ad tech to execute actions on behalf of Gigi. And that agency exec was like, hey, like, why even invest in that if everything's gonna be MCP enabled in two to three years? And and I said, like, I don't I don't have the the ability to wait two to three years to Value to our customers. Like we may not exist in two, three years if we did that. And so you see that across many agencies that are choosing to go through their build phase in which they're building their own buyer agents and waiting on MCP protocols of these ad platforms to exist, for the capabilities of these platforms to exist so that their buyer agents immediately add value and deliver the promise that GG adds to our customers today. And it's in the inverse of that happened with this agency exec. He's like, what you just said is exactly why I decided. to hire GG and Q2 because I realized that I would have to wait two to three years for whatever our internal tech would build to live on these platforms when I can solve the problems that Gigi's solving for our teams, our clients, and our company right now. And that provides a really great playbook of how I think agencies should collaborate with someone like us. and the the the next decision that that agency exec made is he said, like, hey, how can I use the signals that I have access to across many media channels, across lots of client contexts, emails, calls, et cetera, to build a central intelligence layer that I can then use to enrich all of the agents that we work with. And so instead he made decisions to build that. which we're now collaborating with so that Gigi's able to enrich the signals that Gigi has access to to make better decision on behalf of their clients. And so, th that's my call to the industry is just don't wait for the MCP. It's also very It's very I'm very biased in saying that because I'm basically saying don't wait for the MCP, hire Gigi right now, but I believe that to be true. Like we we we're adding value to our customers immediately. and anyone that's gonna be waiting for an Amazon MCP to have all of Gigi's capabilities or a Google MCP to have Gigi's future capabilities in the next six months is gonna be waiting for a really long time. Moffie: Yeah, waiting for an MCP two to three years with how fast agents are moving is insane. So I think you're making the right choice. this was great, Adam. where can people find and follow you and G G? adam: Sure, GGcode.ai, I'm on LinkedIn and X, Adam Epstein, A Epstein underscore, I think my handle is on X. but pr pretty pretty online guy, so you you can find me anytime. Moffie: Good stuff. All right. This has been another episode of Unharnessed. We will be back next week with more AgenTick and AdTech.