AI-native workspace

Speaker 0: I feel like a I'm here. Speaker 1: Why are you like to say that? Speaker 2: AI AI wanna flip the camera around. Speaker 3: There you go. That should be it. Speaker 4: Can you Speaker 5: see your mic moving when I talk? Speaker 3: That way Speaker 6: so you could see what it is. Speaker 3: Oh. And Speaker 5: then AI turn Speaker 2: on. Speaker 3: Thank you. Mike's moving? That's that's Speaker 4: a little. Yeah. Speaker 5: I think he'll just wanna zoom it in on on me. Speaker 6: Can you guys check to see how I'm, like, fucking Speaker 3: all this kind of stuff? Yeah. I should be able to see it on that. Can you see it? Speaker 0: AI Speaker 2: was, harder than Speaker 3: I thought, but we did it. Speaker 5: We got it. AI. Speaker 6: Let's go ahead and get started. Yep. Good to go. Speaker 2: I wanted to introduce myself then. I'm Matt from a trade a couple of weeks in Speaker 5: Yes. Yeah. Absolutely. To see our levels of advice. Speaker 6: Yeah. Tomorrow. So I'm Speaker 5: excited. Thanks for coming out. Speaker 2: Yeah. Were you at the hot co meetup too? Yeah. So are you involved with, Speaker 3: it? Yeah. Speaker 5: That's 1 of the 4 different things I'm doing right now. Speaker 7: Yeah. Yeah. Speaker 5: I'm doing a little bit way too much. Speaker 2: Do AI see all of it? Like, you're in Growth 12 too AI chance? No. No. Speaker 5: I'm CLI was a founder it residence here. And And that's it this and PSL. It's more than Speaker 3: enough. Yeah. Awesome. Yeah. Speaker 5: Well, thanks for putting this stuff together. Yeah. Of course. Thanks for coming out. AI. Speaker 5: We're gonna go ahead and get started. Welcome. Thank you. Feel free to come on over. We got plenty of seats. Speaker 5: Alright. Quick show of hands. How many of you it this your first time at AI tinkerers? Great. How many of you is it's your first time at the go to market AI tinkerers? Speaker 5: Better. Awesome. Very cool. I'm Dan Moore. I'm 1 of the organizers for the go to market track, of AI tinkerers. Speaker 5: And, this is something new and something pretty special because in a world where you can build anything, distribution is pretty much all that matters right now. So thanks for coming out. Some quick logistics to get out of the way. Bathrooms are back through that, Ted lessons AI, in the Slack. Wi Fi, there's a poster over there in case you don't have it already, and be sure to check-in. Speaker 5: So if you did not get scanned at the door or haven't checked in yet, you can sort of scan that. The platform will count it against you if you do not. And then after the event, and that's how we improve these meetups. So would love your feedback of what went well, what could we change, what could we make better. So, when we got this started in February, it was based on a few awesome conversations with local people. Speaker 5: And the idea was essentially, how do we bring about all of this discussion related to agent use and LLMs and AI around go to market and Growth. And I really, really enjoyed the AI Tinkerers meetups, especially John's debt tools Track. If you haven't been out to that, Use shout out. And so a lot of this is modeled after that. But in good AI Tinkerers fashion, this is a heavy heavily curated room. Speaker 5: So all of you were essentially screened and were great participants to come, and join. So, we've been since the founding, we've been, like, 4 x, 5 x oversubscribed to every single event. So APIs like, I would say give yourself a round of applause for being here, because it's pretty pretty special to get in. So, the the whole point is to share and give Slack, though. And so we are looking for speakers. Speaker 5: We have 8 amazing talks tonight. And if you're interested, or have something that you've been working on, even something that doesn't work, that will help your fellow, people in this community to essentially take what works, what doesn't work, and skip the Singh, hopefully, that doesn't work instead of making those same mistakes. But this is really about, like, the details and sharing and and being public Event in something like go to market where signal over noise is incredibly important. We all are gonna lift each other up as part of this. Joe, there are essentially, like, 3 ways to get in besides being awesome. Speaker 5: Like I mentioned, you can give a Presenter, and if you have in the past that immediately Built you up to the top of the list, or you can be a sponsor or co organizer. So if you wanna debt back with time or money, that's a great opportunity to to jump the list to. And speaking of awesome sponsors, we have Patrick tonight from Clarify who's sponsoring this event. So give it up for him for the pizza. And, for those of you who don't know, Clarify is an AI native CRM taken on Salesforce and an entire go to market stack new, probably, which we might hear more about. Speaker 5: We will. You're up first for prevent. So help phone. They just released a new feature, and I'm super excited, to use it and, have you all check it out as well. Anything else you wanna share? Speaker 6: That teams we work with Alex campaigns. Joe from this AI that Speaker 5: Yeah. Yeah. It's excellent. Awesome. Alright. Speaker 5: So to get started, we're trying something new as well. We did it last teams. Bots, essentially, just wanna have a discussion about what everyone in the audience is seeing and using. So last time, I shared this. So I was having trouble keeping up, up to date with everything that's going on in the go to market and AI space. Speaker 5: And so I basically built Dan AI agent to go out and pull over 2,000 specific news events that happened over the past month alone, and summarize that, create some, themes for me and trends. And so I'm not sure what I feel about this as a title. I feel like that's, you know, a couple Agents old now, but Claude wrote it. So, so a couple of interesting points though that I really liked, out of this. AI, I mean, the the core, piece of this is okay. Speaker 5: OpenAI is no longer the defacto model. So just by show of hands, how many of you use OpenAI as ICP your go to model? Oh, got a few. Okay. How about anthropic? Speaker 5: 1 off? Use it for use all of them? Speaker 3: I use all of them. They're they're 1 of my Speaker 5: Okay. Both of them are. Does anyone new, like, main 1 something other besides those 2 providers? Which 1? 0, okay. Speaker 5: Yeah. Moore the 1 that's free that's given to you by your employer and forced to use? Okay. Okay. Speaker 3: Find do you work for People? Or do you work for Google? Okay. Speaker 6: There you go. There you go. Speaker 5: Hey. Hey. No shaming. No model shaming. Speaker 3: Nice. Speaker 5: And as as, like, your first your primary 1? Bots it. Very cool. AI. Yeah. Speaker 5: Code. Speaker 3: I think, I use AMP, predominantly. It's at ampco.com. I used to work for them at Swiss Craft. And, they actually I think it's interesting sharing AI because what they do exceptionally well is multi model. So best agent. Speaker 3: Obviously, like, codecs has this Speaker 8: thing. Campaigns School because, for Speaker 3: me anyways, they built some nice tools based on suck graphs, code search. Joe for example, when I was building Synter, I was looking for, you know, a Tarka architecture or a AI, and they have something called, like, the librarian. It'll go out and search, like, all, like, open source GitHub debt tool look for example code, which is what we do as code again. So that's a really nice thing. It layers in author, model called the modeling is a g p, like, 5 dot 5. Speaker 3: Mhmm. But use can call it AMP People. LLM pull in, feeds and then I just borrow some of the design principle to come in and, like, review the overall strategy. So I Founder between using the Oracle for, like, code review. So if I PR myself with a different model, I had to do a code review of my PR with a different model. Speaker 3: Right. It think the biggest change for me for under opening, I thought, image 2 is 1000 times better for what we're doing than Nanon, which is AI. Because I didn't like Sora. I didn't like any of the stuff they're doing over there. Artistic Yeah. Speaker 3: For, like, what I was doing. It 2 has been Speaker 5: How many of you are kind of, model or provider agnostic then when it comes to, like, integrating it to your products or trying different Singh? Like, how easy is it to switch? Okay. That's the majority of people, I would say. That's awesome. Speaker 5: I think that's that's super interesting. 1 pattern that I've tried and had pretty good success with this past month is AI just love the Claude Code harness, and I use Code through that, for adversarial reviews and stuff like that, like OpenAI pipelines. Does anyone else do anything like this? Yeah. Use that same flow. Speaker 5: Anyone else? What do you use to drive between those? AI, what do you have a single harness robust Closed itself? Speaker 3: Okay. Speaker 2: Yeah. Speaker 5: Yeah. Anyone else have an interesting setup they wanna share? Okay. Yeah. I think I mean, it's super interesting how, like, much bought in I am to a single harness at this point and so many patterns that have come from that that it's, like, very daunting to just want to switch at this point. Speaker 5: Joe, it interesting time. So, the second thread over here, was really about these foundational labs all, buying consulting companies for distribution HINTS enterprise. So how many people here are, like, full on founders or Shuvo been charge of everything? Awesome. So when it comes to your deployment, how suck of your work are you being it consultative in nature AI going up the stack instead of just a platform layer? Speaker 5: AI curious it that's like a trend that Moore seeing at your campaigns SaaS help. Yeah. Speaker 3: Yeah for shares, I mean so many very close work, Speaker 5: Yeah. The bad version. AI. Speaker 3: Yeah. Speaker 5: Yeah. Yeah. It awesome. Another thing that was pulled out was 11 x had an, if you who's familiar with 11 x, the AISDR company? New people. Speaker 5: Okay. So they made big waves coming out find and they had a whole bunch of drama and all APIs stuff behind them. Their CEO just came out this past month and said AI don't work, which I think is an interesting growth strategy. But, it yeah. It was it was crazy. Speaker 5: But they had a partnership with IBM as IBM is the platform find they are employees on the IBM platform. And so it it looks like there's some really cool implementing partnerships that are arising, but I'm not I don't know. I'm not I'm not sold on doing it within 1 business versus a people, a strategy that I'm trying to see how it how it works right now. So I have 2 different startup. I have Marketo, which is my consulting start up, which is this website right Ryan, and then I have, like, a platform underneath it. Speaker 5: But they're 2 different investment models find they're they're tightly coupled, but they're just 2 different businesses find they scale wildly different. So question? Yeah. Please. Speaker 0: Yeah. Yeah. Or if you but raise your hand. Speaker 3: I'm just curious. And then second is related question is like and this is driven by this trend of work. Yeah, you see here that there's some wonders. I have seen, like, in the edge, these guys, we work with partners as well, but in our inbox is where we get sponsored. You know, like, reach out. Speaker 3: Okay. Because that's a really good indicator of care. It's changing a lot. Beginning of the year, AIT is after, like, 4 years of around Europe. 400,000 meeting people to the website. Speaker 3: By April 15, in the last 30 day period, it was up to approximately 1000000. But by March 15, Lucas 2. So SaaS AI like, what happened there? I think it's the whole world saying quirks being expert. Right? Speaker 3: So my second question is, how many people are working jobs and they just be like, I don't get expert. It's extra hours here and there. Is that Speaker 5: AI, on the side or at the job? Speaker 3: Yeah. You are. You just gotta fix that. Speaker 5: Yeah. AI absolutely Speaker 3: Implementing thing. Just trying to get enough assistance. Yeah. Speaker 5: And for those of you who don't know, this is Joe. He's the original founder of AI Tinkerers now in over 200 City, new higher number find over a 100,000 people. So integrate community. And you're presenting at the end too of something cool. Shares you go. Speaker 5: Jenn other lessons? Or I could Speaker 3: Just a comment, Joe. AI, yeah. I actually Good. Yeah. I just think it's something that. Speaker 3: I've actually heard a few people say that. Like, there's all of the, like, Now that you have, like, a few things you can do, it's like it's it's it's like more than a TikTok. Speaker 5: There's no limit. And it's a different it's a different kind. I don't know about anyone else. I'm exhausted most of the time. Like, it's high level decision making is what it feels like, and it's just AI, I don't know, Steve's, AM AI process. Speaker 5: Yeah. It's real. 1 other trend in here, AI noticed, how many of you are familiar with Scientists? Few people? Okay. Speaker 5: It's a LinkedIn, as well as a bunch of other channels now. I think they're omnichannel, email find and some other stuff. It's an outbound sequencer, but they've evolved find they just released, a bunch of Claude skills for different go to market workflows, as well as there's a ton of other skill repositories out there. I'm curious if anyone has, like, fallen in love with a particular source, or is it just, like, everywhere all over the top? Yeah. Speaker 5: I I think they have a lot more than that, but that's what Lemnless is specific about. Speaker 3: For salesskills.h. Speaker 5: Skills.h. Anybody else, like, have a go to for discovery of agent skills? Use it for someone? Like, internally? K. Speaker 5: How how are people sharing stuff internally? I think that's a really interesting problem. AI. Events like these. Yeah. Speaker 5: Event. Speaker 3: Nice. Speaker 5: Yeah. Like turn or public? Speaker 3: How Speaker 5: many of you are show of hands, creating skills? It. Keep your hands up. How many of you have created a new skill in the past week? Week. Speaker 5: K. How many of you have deleted a skill that still has your Dan up? Okay. Super interesting. Yeah. Speaker 5: Totally new world. Awesome. Yeah. And then the final thing, or or 2 final things. 1, how many of you have tried adopting an AISDR? Speaker 5: It was all the Raise, like, a year ago. Speaker 2: But what Speaker 3: We have 1. Speaker 5: You have your own? Speaker 4: Yes. Yes. Speaker 3: I know it's not pretty well. Speaker 5: Yeah. Yeah. Speaker 3: I mean, I think that's open. Speaker 5: Like building them together into your own unique agent, essentially? Speaker 3: Yeah. Yeah. It's, like, a take action tape, which is very different than it's more of an exoskeleton than a cable. Mhmm. Yeah. Speaker 5: Yeah. You'll definitely have to present that. How Manya of you are running, like, AI employees on OpenAI or Servers or any of the new things that Code out? K. Do you have, like, full automate? Speaker 5: Like, someone shares their their setup. Do they have credit cards given to their agents yet? Yeah. Speaker 4: Big part of my thesis is, like, probably a bad idea right now. If you'll get better, but, like, it doesn't Speaker 3: have to. But, Speaker 5: Bar you hosting an x 4 0 2? Like, do you have oh, you're using it. Okay. It anyone on the other side of that transaction hosting on x 4 0 2 yet where peep agents can essentially buy your servers? Speaker 2: Very cool. Speaker 5: AI 2026 other shout Bots from Seattle Date big startup here in the last Terminal, AI it. How many of you have seen or used Copilot kit? Very cool. Definitely check it out Jenn interfaces and There Joe go. Or before it became publicly known find then SageJox. Speaker 5: Joe what's the best way to describe Sageox? It's hard for me. I don't know. Has anyone else you've seen it? Speaker 3: CLI, great. Well, it's what they say up there. They're they're kinda cool. Speaker 5: Context engineering Selling that profiles? Deleted. Yeah. I think the last thing I saw about them, they had a physical device that you could just have a conversation background, and it figured out who to notify or, like, how to align the team, but that could be outdated by AI new. Yeah. Speaker 5: But it was like a physical device that they keep on a suck, AI, kind of like an owl it you've seen those or any other robust Speaker 3: Yeah. Oh, we're not sure. Okay. Speaker 5: Yeah. The AI mind. Yeah. Really, really interesting stuff, on the bleeding edge new. So definitely check those out. Speaker 3: K. Yeah. Speaker 5: Yeah. It's definitely proven a valuable pattern, to get out there. So any other news or things new things you've seen in this past month that you won't share? Yeah. Speaker 3: I'm actually just curious. Speaker 5: AI? Speaker 3: Yeah. Okay. That's like. Interesting. Yeah. Speaker 3: K. But I think that there's a lot of. Speaker 5: Hermes is open source as well. Right? Speaker 3: Yeah. Speaker 5: Lead. Yeah. Anyone else using any other interesting agent harnesses? Speaker 4: Diego. Speaker 5: Dan Slack? Jenn any, like, really good, use cases or anything like Speaker 3: that? Mhmm. Yep. Speaker 5: Wow. What what does it use for containerization underneath it? Oh, okay. Right. So the model underneath, it doesn't have access to credentials. Speaker 5: Yeah. Huge. Anyone else doing anything? Alright. Cool. Speaker 5: Well, with that, we'll go ahead and get transitioned. So, Patrick, you're up first. Yeah. If, yeah, I'll text you the Zoom link real quick. AI AI gonna disconnect and then, share your screen. Speaker 2: Track you. Speaker 6: Okay. Sweet. For robust that don't know me, my name is Patrick, cofounder find CEO of Clarify, Seattle based campaigns. We're AI 26 people, working on an autonomous CRM. So not here to talk a lot about the autonomous CRM, but talk a lot about the learnings that we've lead, specifically building tools for agents that can integrate with OpenAI Scale and Synter call the different agent harnesses out there, including our own Clarify agent harness. Speaker 6: So, how many people here use SaaS CRM? Great. Yeah. It's a pretty ubiquitous piece of software. It's kinda like Jira, probably Code to market side. Speaker 6: So it's probably the most expensive database that teams will pay for. Effectively, that's really what it is. You AI at Salesforce, HubSpot is effectively just a database. Not a whole lot of AI built in, and we're trying to change that at Clarify. And I'll find just share some of the lessons learned that we've had. Speaker 6: Joe 1 of the things that we cared a lot about when starting this company is how do we think about this in the form of headless. Debt me know if people can see this. When we say headless, we our AI company was a customer data platform, made it really easy to capture and track all your customer data. We really wanted that in kind of a CRM. So here we are 2 and a half years later with AI. Speaker 6: And we really wanted to focus on how do developers interact with, the CRM as well. So we Opus a lot on our external API. This was actually super helpful for us to get early in market. And then, you know, MCP came aligning, and we were 1 of the first people to actually ship Dan MCP server. Now we're working on things like our CLI and SDK as well. Speaker 6: And, a lot of this is just around tracking customer data. Right? So CRM it typically system of record. We're building Bots the system Moore record, system of engagement, as well as system of intelligence on top of that, but it all starts with the data first. So I'm gonna demo quickly kind of some basic things. Speaker 6: Right? This Skill Jon from kind of a crawl, walk, run approach. First and foremost, how do you create a record inside of the CRM? Well, for us, that's using our API. So we have a, you know, basically Date record endpoint. Speaker 6: Right? So you can create any record whether it's a task, a meeting, name it, a message. Specifically with this Across request, 1 of the things that you'll see is this got created. If I switch back over here to our local host and switch over to person, you'll see that this record got created. Well, this is this is super simple. Speaker 6: Right? Like, this took less than 100 milliseconds to create that record. Perfect. Right? This is what we use in DogFruit ourselves. Speaker 6: 1 thing is it's schema based, so it's really easy to track all the data you care about find it's just a JSON endpoint. But now if I wanna do something a little bit more sophisticated, right, like, create a lessons, company, and deal, you can see that the code gets a little bit more complex. Not too bad here. It Skill file, like, human readable. Right? Speaker 6: And this will still execute, fairly quickly. Let me just go in and delete this data here. Delete find company delete. Okay. Let me go execute this 1. Speaker 6: So APIs are quick because you get a really quick response. Right? It's, fully deterministic. So you again, you can see that that response came in. I had the company and person created here as well as the deal created right away. Speaker 6: So super easy, very people. Lucas me through sequential calls to, you know, AI that, but totally deterministic, cheap based, and scriptable. So the thing about AI is with nondeterministic, work here, you know, we can I can wrap our API find a skill here? So we're talking about skills layer. So we have a AI API skill. Speaker 6: Definitely something that I would recommend people do if they haven't done this. I will show you exactly what that looks like here. Go delete the data. And the reason that this skill is actually very helpful because if you're giving just the API spec to Claude or whatever agent you're using is, it's open to interpretation. I can tell you that this skill makes it a lot easier for, AI to actually use. Speaker 6: So let me actually go try this shares I'm gonna jump in over here to Claude, which has access to that Clarify API. I'm gonna have it go create 3 campaigns, each with a contact Jenn a deal here. So 1 thing to Node.js that instead of me just calling the API directly here, basically, I'm using an AI to go use our API. So this is typically what most companies will do. Right? Speaker 6: They'll have an open API, they'll wrap it with a skill, and make it really easy to call. Now you see previously what took me less than a couple hundred milliseconds to create a bunch of records now has to enumerate through a bunch of different reasoning loops in AI. Help, so if you're doing anything in real time, let's assume you're a PLG company and you wanna track all of your customer sign API or it Event. This is super time intensive. Right? Speaker 6: So not something that I Remitly recommend. But it's really nice because, you know, we lead this a lot of information about our API, best practices. Let me know if people want me to make this a little bit bigger here. Find you can see that's getting the core schemas that we have as well. Again, everything is schema based. Speaker 6: It's looking up to see if there's any relationships on these companies. It understood the schemas. It's now executing all 3 companies created. Now let me create a contact and deal. So for us, like, we don't name things contacts. Speaker 6: We name them people in our system. Right? But AI is actually fairly good at reasoning between those 2 things. Contact is kind of a Salesforce colloquialism find knows Horwitz create the personalized, lead the contact. Buy, again, this is new, like, what, about 40 seconds that it's going find doing APIs. Speaker 6: So, name needs first name, last name. Again, this is actually based practice inside of our, agent skill that we created here. And you can see AI. Done. Creating 3 companies. Speaker 6: If you go back over here, you can see the 3 companies are created. Created the people and the deals as well. So, super easy to do. Right? And aligning, if you're thinking about your CRM as a database, you wanna make sure that that database is easy to, interoperate with and extremely extensible. Speaker 6: So any questions about this? Okay. Super simple. Again, crawl, walk, run. So the 1 thing here that's a little bit better on demo 3 it, again, we wrapped our API with a AI API skill. Speaker 6: This skill is actually available, so it if you use Vercel, you can actually go install it from shares. And that's how we deploy our skills to our customers. Okay. Last demo. And this is typically where, you know, I think a lot of people went a bit overboard with MCP Tool. Speaker 6: So let's walk through what this will look like and clarify. Okay. Now I want you to use AI MCP Jenn Date boom. Okay. So this is gonna do something slightly different. Speaker 6: Right? So instead of calling our API directly, this is actually gonna go call all the tools that we expose via MCP. Right? So for how many people are familiar with MCP tools? Okay. Speaker 6: Great. Pretty much everyone's AI want up. I'll use the Clarify MCP this time. Let me load the relevant tools. I'll get the right schemas for company lessons and deal Selling AI tool use Jenn. Speaker 6: K. Let me go authorize a lot of this stuff. Same stuff that out skill did. This is actually gonna take you know, generally, I SaaS it's about the same amount of time as the Clarify Skill wrapper that we created. But I have to go approve a bunch of these tool calls because this is a new session. Speaker 6: And it's gonna do the same things that the other thing did. Right? So same data, same result, different door. This is 1 of the things that we talk a lot about. So with tools, they're generally easy. Speaker 6: And the reason that we have tools is because we actually have our own agent that I'll showcase, closely in this as well. But, this is gonna take a bit okay. Done. You have to clarify MCP reading 3 companies each with the LinkedIn extract and deal. So that took a roughly around the same amount of time. Speaker 6: So 1 of the things that I wanna highlight specifically as a company building AI capabilities is that, you know, what's the difference between API or agent? You know, it you based me 2 or 3 months ago, I would have been all in on MCP. We're not just MCP servers. We're also MCP AI. You can actually connect to clarify 2026 all the tools that out use. Speaker 6: But what I've learned most recently is that, you know, generally, APIs are still very, very good for based, and you don't necessarily want to have the high token robust as well as the time to do a lot of this stuff. So for us, it's not kind of a or. It's definitely an and. We support both API and agent. So for direct API, it's definitely good when scripted, repeatable, high volume. Speaker 6: So for us, like, we send all of our web traffic to clarify. Right? If I was to wrap an agent around that, that'd Ben, 1, super extensive, and then 2026, AI necessarily get us the results that we want as quick as we want them. But if you need determinism, API is definitely the way. Nondeterministic agent is great. Speaker 6: And so this is definitely something that I think a lot about. There's definitely things that we've done here that makes it a little bit better. 1, so, like, the direct API, like, the file modes are just Clarify in general to to agents. We had tool specifically for tools, we had to actually figure out how to get the error messages back to the agent harnesses Joe they can actually reason through it better. We also have things like tools, like, debt give feedback. Speaker 6: So Ben AI an agent runs HINTS an issue, you can actually give us feedback as well. That was a a a huge learning for us. Buy 1 Singh. Right? All of our MCP tools are built on the same API that we expose our customers as well as the same API that we use. Speaker 6: So we're dogfooding ideas we build. Our customers are dogfooding it. So here use can see that this Jon, right, this is the the MCP tool wrapper. You can see I can define a tool called create structured. CLI, use the input Steele, create record schema as a create record description. Speaker 6: This is basically what we feed into the agent business. And you can see this Andrei, which calls the records. Create endpoint, basically robust to post out records endpoint. So this is the same API that we use and expose tool our customers. So, 1 thing that I think about quite clearly for us is that we're just there's levels of Track to customers. Speaker 6: The API is the lowest level of Track, then you have tools, and then you have things AI skills built on top of those tools and agents that are built on top of those skills. So quick takeaway as far as APIs 3 Singh. The agent is a client for your API. Right? So APIs, typically, we're in for, computer turn computer interaction or humans consuming those APIs, whereas agents are really good at consuming these APIs now as well beyond just tools like MCP. Speaker 6: The CLI, SDK, and tools all use the same API, so it's really imperative for us to make sure our API is good AI documented. This is where the skills come in. Right? We publish our API skills. We have, AI Bots for all of our API endpoints, so really email want understood by by both humans and agents. Speaker 6: And, again, design the surface AI an agent, surface for an agent, tool TypeScript, or prompts. So we specs a lot of time thinking about out for things like AEO Ben we publish. So we're our MCP is listed in both the OpenAI and Anthropic Store. We SPEAR a lot of time trying to figure out what the tool descriptions look like. We found that actually improves Singh like AEO for ourselves as well. Speaker 6: Neat trick. Ben anybody wants to play with that. And that's pretty much it. And that was under 10 minutes. Speaker 3: Venue. Speaker 6: I mean, we so we charge based off credits. Right? So the only thing we're basically a free Date. So the only thing that consumes money for us, the only way we make money is that when it's actually inference cost. So every tool call is basically consumed some lessons of credits. Speaker 6: So, yeah, we we measure it and then charge for it as well. Speaker 4: Can you keep the MCP and API kind of full process? How do you tell, like, which API endpoint come in? Speaker 6: Yeah. Actually, I have want talk on on our MCP because all of our so we have this agent here. So you'll see this is what's called RAG. This is literally our out inbuilt agent. So there's tools that RAP has access to that are are not publicly listed in, other MCP server. Speaker 6: AI? So, like, this SaaS, you know, I'd say turbocharged because there's a bunch of different tools here. But, and then the API is slightly separate as well. Right? Because there's endpoints that we don't expose publicly as help, but that our agents have access to. Speaker 6: So we basically have a giant manifest that basically controls Bots what endpoints are available publicly as well as turn, and then, vice versa, what tools are available Bots publicly as well as externally. And it's literally AI a 1 line config for each of us. Speaker 3: In the customer. Speaker 6: Cool. So there's so Aligning about it as you have the API layer lowest level it, turn you have the tool, and then you have the skill, and then the agent. Right? So for us, like, we have skills find clarify. Right? Speaker 6: So there's a bunch of skills. You can go create skills inside of our product. These Skill, we also publish skills as well for our customers that know how to use our tools slightly better. So, like, for us, if I go to something like let me just actually go show you what a skill tool like find Clarify. File go to I'll go to our public app or my live app here. Speaker 6: These are my Agents, by the way. But I'll jump over here. Okay. So for me, specifically, I have a bunch of skills. These are all the skills that I have that I use. Speaker 6: Use. Right? And these all use different tools. Right? So Jenn generates partner offer AI page. Speaker 3: I'm trying to find you 1 that might not have sensitive data in it. Speaker 6: This one's good. Yeah. So write a LinkedIn post in Patrick's voice. Use the write in Patrick's voice skill for all voice tone and clarify Scientists. Apply the LinkedIn post format Sales from the skill. Speaker 6: Shortcut, skip the format question. The format is always Laes LinkedIn post. So this is actually referencing other skills. So there's some level of inheritance across this. So, like, skills can reference skills. Speaker 6: Agents can reference skills for us. So, like, I use skills all the time. These are not stored on my machine. These are either private to me or available to the company. Out, yeah, like, anybody can create skills and clarify. Speaker 6: Skills are just Ryan abstraction on top of whatever tools are available. Let me know if that answered your question. Speaker 3: Yeah. And, like, our customers like, what's the trend to, like, drive the CRA? Speaker 6: Yeah. AI mean, the the the hard part for any of this stuff is, like, discoverability and education. Right? Because, like, all this stuff is relatively speaking on the cutting edge. Right? Speaker 6: So, like, in AI, right, in this agent, I can say, like, hey. What skills should I I have? Right? And this will actually go and AI, look at my business, understand, like, who my customer is, understand how I'm using the CRM. And so, like, the goal is for any software that we're building these days to be more proactive and less Seattle. Speaker 6: But most people still have to nudge the software to get what they want out versus the software nudging them to get what they're out of it. So we're actually moving into a world where Event tool find Clarify, like, you're gonna have agents that are trying to optimize you're gonna have an agent running that's trying to optimize your own internal usage of our product. MCP is used way more. So once we CLAUDE.md MCP, like, we we actually so it's funny. We have out our metric our North Startup metric is Sheikh active users. Speaker 6: Right? That's what we or what we track as a company it the thing that we're trying to optimize around beyond things like revenue. When we launched MCP, we actually saw MCP usage outside of Clarify Joe, like, vertical. Right? And then what we saw it, like, internal usage dropped slightly and then actually caught back up. Speaker 6: So those people were using Clarify a lot more from Claude or, OpenAI Claude or other types of solutions, than we even expected. And so then it's a question, like, hey. Do we actually track that as active usage in our system or not? But, yeah, we actually get more MCP tool calls now than API calls for everything, like, create record, update record, add comment, you name it. Speaker 3: Debt part is Speaker 6: not the the tool call itself, not yet. So, like, we're we're giving away a lot for free at the OpenAI. Mostly because, like, all this stuff is still, like, find very experimental. Like, people people don't understand necessarily, like, how to think about credit based usage for these types of solutions. So, like, we're we're launching agents. Speaker 6: Right? Like, this is the first thing that actually has, like, I'd say asymmetric Track for us. Right? So, like, you can launch an agent to do whatever you want. Right? Speaker 6: This is a fact 2026, like, open call Jon a box agent Built 2026 your CRM. So, like, it can read all my emails, read all my meetings, file, Track. AI lead follow-up I need, Track me all my stuff. So this stuff actually CLI, like, hard Bots suck. So this is why we're moving more HINTS, like, a turn based pricing new where, hopefully, that will help educate customers Jon, like, hey. Speaker 6: This actually costs us money. Therefore, we have to, like, charge you a bit of money for APIs. Whereas previously, we weren't charging for tool John Bar tool, calls. Speaker 3: Sure. On Speaker 2: time. So any other Speaker 6: Alright. That's the old note. Speaker 3: That is not finished. Cool. Yep. Speaker 0: That's the old note. Alright. Speaker 7: I think it works. Speaker 3: So I'm Speaker 6: not sure if that was Speaker 7: I learned that after the fact. It was it was great. Speaker 2: Nice. Find you started a few months tool. At least the the first time I checked out the platform, probably available. I it was AI coworkers. Speaker 7: Yeah. How Speaker 2: do you call it? Speaker 7: We're gonna do AI coworkers. We just had to do something first. Alright. So AI. My name is Ryan. Speaker 7: I'm the CEO of Ambiguous AI. We originally built AI coworkers that work like teammates with their own Google accounts. We were partnered with Google teaching their keynote, but unfortunately, Google started blocking us and throttling our agents with too much usage. Terrible. Right? Speaker 7: We're, like, in their slide on their keynote for Google Cloud find they still block this. AI Opus teams too Built. You know, they have other problems. And so, we AI a support ticket find we decided, hey, you know, AI my CTO well, I was, like, making a bet with my CTO. Buy. Speaker 7: I'm gonna go off and I'm gonna build my own Google Workspace from scratch before they can respond to the support ticket. And I'm nontechnical. I haven't really coded for a decade. I built up ICP a like a rig or something find then I started coding and then LLM and behold we built Google Workspace in less than 7 days from scratch end to end. And so I'll show you a demo of the product so Bar, and some applications that we have. Speaker 7: 1 second. Let's shares screen. I've never used this before out is it working? Let's see. Share screen again. Speaker 7: Start broadcast. We recently switched to iPads for development because now we're doing so much AI coding. Alright. Alright. So, let's go back to the app. Speaker 7: Here we go. So this is Ambiguous workspace. We built all of this in, less than 7 days. Now we're about 30 days in. We had 200 autonomous agents Built over 4,000,000 lines of code, all coherent code running AI of 24 7 agents to build this. Speaker 7: So I'll give you a quick tour. So we have docs, sheets, slides, chat, mail, drive, tasks, CRM, forms, calendar, Wiki, admin automations. We even have DocuSign coming soon. It's it's it pretty crazy what we've built. It's kind of feature compliant or feature complete. Speaker 7: This is our, you know, doc editor. It has collaboration Agents you can kind of collaborate with your teammates. We now run our weekly meetings in here find it works pretty flawlessly. So it didn't work the first time out we're slowly moving everything over. That's not this demo out, showing you debt this is really cool. Speaker 7: We're trying to gear it towards go to market. So 1 of the tasks that I did, I set up an OpenAI agent and the way I did this is I said, hey, create an ambiguous workspace, ryan it Ambiguous. Io SaaS a human-AI. So it created that workspace, sent me an email. On my email client, I accepted that agent and added to my workspace and then I told it, hey, can you help me build a spreadsheet? Speaker 7: And so I just told it this basic Singh, you know, I want these companies from a 10+ z SPEAKER speedrun Joe fix. I invest teams out of an alumni fund for those companies and I want you to create kind of a list of that in the spreadsheet. I told it to make it people, and I said, okay. Proceed and share the sheet with me when it was done. So what it did is it created this sheet Ryan. Speaker 7: And so now I have a full list of all the 60 companies that did a 16 z speed Ryan, in an editor that's fully human AI interoperable find in a markdown format that's token AI. So it's Jake cheaper, faster, debt. And so you can see that. And so what I did is I intentionally left out the CEO's LinkedIn in this version. I put a comment here find so there's a resolved comment. Speaker 7: I tagged Papers, so we're still working on some visual glitches, and I said add a column with the CEO linked in for each row. And It said Dan. Badges CEO LinkedIn column find it put that over here. But not only that, it since it's a full workspace and tightly integrated, I can send the agent 2026 email. There is a glitch where it's showing the emails that it sends AI, we'll fix that, but I said, you know, can you add Accelerate, Amdahl, Scale, Seller, Bots, and Loops to my CRM? Speaker 7: Also, can you create a task to follow-up with them on Friday? So it went ahead and said, you know, done. I added those. Then I said, you know, can you add the CEO's context because it was creating the, like, campaigns in the CRM out not the Scientists. It went pain and fix that. Speaker 7: It said it didn't do generates because it didn't know who the CEO SaaS, and then I clarified chatbot. And then, here AI clarified. And then it's a done Badges, Chinmay SaaS a CEO and linked APIs contact. So over here, here's the 6 different tasks that it gave me. So now I can go in here, see APIs task, use know, follow-up the loops. Speaker 7: It's assigned to Friday. Great. If I went into my calendar oops. Let's see which one's calendar. Still getting used to these icons. Speaker 7: Let's go, calendar. And you see, like, the 6 tasks here at the top. So it's, you know, it's again, it's pretty impressive, what we could build in less than 30 Date, and that's just scratching the surface. We have a full AI assistant that's even Synter, than Cloud Quirks, from this. Again, shares, like, just Bar partnership deck that you have. Speaker 7: Like, say you're doing marketing, you put a doc here, with how you wanna create your decks or a Thompson. You can ask the assistant to end up, like, creating a version of that for you. So here it just a quick 1. I lead the AI assistant take that SOP and create a debt, and so it created, just a quick debt, nonsense because I didn't review any of this, but it's just a quick merge of the SOP and the deck using our AI assistant want tool. So all APIs was 0 shot, which is pretty tool. Speaker 7: And so you can kinda CEO, like, when you build an AI workspace from scratch, the it's like endless the limit AI, the limits are are I don't know what the word is Buy basically you data lot of shit. This is really great. So it's it's also crazy IDENTIFICATION from software if you design a really great agentic architecture, a brain of, like, what your context is find you set up the right harness with the right frameworks, you can run these 24 7 coding loops to build production level software. So again, tons of great things here. I don't think we went into the CRM, but you can see people. Speaker 7: We have the people here. You know, slides want to there. You there's a Slack equivalent. So I asked it, like, which companies am I supposed to outreach this Sheikh? And it Sales, here's 6 companies. Speaker 7: Also, all of this works in CLI, MCP, API, you name it, we support it. It also works on mobile. It has a desktop application and a web application. It also has its own version of Dropbox, so it syncs that syncs to s 3. It can do desktop sync, as help. Speaker 7: So you can do a lot of shit these days find so, yeah. Also, I I can show you Drive demo, but I I think as I show that demo, I'll just show it to you Andrei it also has a Zapier by the way, and a Wiki. I have a Zapier built it, and then this is the Wiki. So it's pretty cool. You see all the smart tags and everything all linked. Speaker 7: But, let's see it I let's get this. AI just do something quickly. I think it's let's, you know, create a workspace, ambiguous workspace. It's for Ted find use AI as the human-AI. And it'll go ahead and work Ben, like, it'll take some time in OpenAI. Speaker 7: 10+, anything else I can show you or out use could just jump into lessons? It, you know, it's pretty cool what we've built in 30 Date, and, we're still prerelease. So we'll launch this probably in a few weeks or more. Yeah. Wow. Speaker 3: No. Yeah. Speaker 7: It's fast. Yes. And it's free. Yeah. I built my own. Speaker 7: So I I created a framework called SPEAR, which is basically scope Dan, execute, assess, and resolve. It allows me to kind of run the standard project managing methodology and have the right checkpoints to ensure Waliany. And then it kinda runs the inner loop infinitely to builder Code. Speaker 3: 2 little sub questions. Speaker 4: Did you use anything Speaker 2: as a Speaker 3: scaffolding for chatbot? And second, which which model are you using behind it? Speaker 6: So Speaker 7: I built my own scaffolding. I originally start CEO call it It, and I now call it It. And Joe, yeah. Speaker 4: You just hate other people's code. Speaker 7: I do. I do. I do. I no. I mean, if if you've seen the yeah. Speaker 7: That makes sense. Help, the prompting use want it to be very bespoke. Speaker 0: Yeah. Speaker 6: Joe. Go for it. Speaker 3: I have an excellent blog. Alright. Speaker 7: Yeah. So, find this, I use this SPEAR framework. I was like, I I should test this out. I've also built specs using APIs, but my son was using, he was using Code Coworkers to find create a 3 d model for a 3 d printer find so this is the left side is what you put into Claude, a Frontier Joel. This is what you get. Speaker 7: It's Terminal. But if you tell it to use a SPEAR framework, it actually will self calibrate, with an assessment rubric and create something that's really perfect on the right. So it's it's incredible. Like, if you have the right methodology, you can actually create much much better quality. Speaker 6: Great. Speaker 7: Let's see. I don't know if the OpenAI debt up correctly, but did you create an account? Sometimes open Claude doesn't lessons. AI I'm I'm gonna grab a quick water real quick. Yeah. Speaker 7: Go for the Stealth APIs yeah. Yeah. Go for it. Speaker 4: Yeah. Aside from your product, I mean, it's very impressive that there was Speaker 3: so much parallel work happening. Yeah. Speaker 4: It's a little bit about, like, above and beyond the, like, okay. You wanna do 1 thing really well. Speaker 7: Yeah. Yeah. Speaker 3: I mean, 1 of Speaker 4: the big challenges right now is, like, be able to test many, many work trees live. Like, how do you think about testing the unit test, the integration test to the live, like, the Speaker 3: browser Yeah. Speaker 7: Yeah. So you first need an agent architecture. Right? ICP, because if you think about an organization when you're scaling an org, the challenge it, like, unique company values, AI, unique cultural values. Right? Speaker 7: And you can't scale Killian 200 person org unless you have that defined. The same is true with agents. If you're gonna have 200 agents running around Buy you don't have the shared communication layer AI this tribal knowledge documented, your agents will just like converge the crap. And so the way you solve that is you have this really, intelligently designed agent architecture and then you can run these 24 7 loops and you get quality results at the end. So that's something that we do. Speaker 7: You need Dan agent market, then you need, like a 5 point process where you take each intermediate step like scoping to planning to executing to assessing find you don't move to the next step until it succeeded in the previous step. And then AI, when you're running the execution loops, it's important that agents have checklists. If the agents don't have checklists, what happens is they'll skip steps just like people find so you have a AI checklist with evidence to kind of ensure Waliany. But you need a lot of the testing harnesses and the feedback loops as well. Great. Speaker 7: So APIs time it listened. It created a workspace find so, I should get something here. We have some Buyers, but I'll claim quirks this into an existing workspace and I will choose to merge it into Ambiguous demo. And so my agent that I creating Dan OpenAI is now inside, my workspace find so that's it. All you have to do is messaging skill to your Agents, Claude coworkers, cloud code, and it'll create a workspace for you find you can claim it. Speaker 7: Yeah. Go for it. Speaker 3: Yeah. Yeah. Yeah. Yeah. Yeah. Speaker 3: Yeah. Speaker 7: Yeah. Yeah. We we just have, like, 20 simulated users, agent browser, stuff like that. Yeah. Yeah. Speaker 7: Yeah. You, the scaffolding is really important. Yeah. You can it's it's pretty cool. Yeah. Speaker 7: It didn't work the first AI, of course, and then we had to have humans use it and then change the model. We also you know, go for it. Speaker 3: So you out creating Speaker 4: workspace, SaaS the data in Jenn, and you talked about agent Scaffolding which I absolutely agree with. It is it create a compressed context or, you know, AI, Speaker 3: search skill for the agents you think it to have AI view Speaker 4: of this immense amount of Date. Speaker 7: Joe, it's it's way better because once you have all of this data for your agents, it's cheaper because all of our formats like, if you look at Offering document format, it's about a 150,000. So if you're using OXML, it's a 150,000 token-based is a fortune. And if you're making a change and you have to reprint those tokens, it's terrible. If you use a compressed form or representation, it's like 10 k. And the other issues with, like, Google Docs or Google Slides, etcetera, sometimes you have to make, like, 20 tool calls to change a doc. Speaker 7: If you write a system for agents, you only have to do 1 tool Scale to change a doc. And so the whole world needs to be rewritten for AI native software and so that's kind of why we 1 of the CEO suck Terminal built AI coworkers on Google Workspace. We saw how all of this worked and played out, and then we realized, hey. You do need this AI first layer. Oops. Speaker 3: John, no. No worries. Yeah. Speaker 7: Out, 1 more cool thing, Andrei this is on, like, building this in Product. If you look at the activities, you can now see, the people, like, because agents all have their own accounts. Right? So you see Papers has received an email, sent an email, Ryan has email received, Ted was an agent that's created. Every single action is auditable and tracked. Speaker 7: So if you you can set Remitly, rate limits for your agents, you can control them, lock them out, give them certain access permissions. And so when you rebuild the workspace from scratch for agents and humans, you get some pretty powerful things. Speaker 3: Tool. Yeah. Yeah. Speaker 2: So I thought Speaker 6: We're just Speaker 2: I'll go for it. Speaker 3: AI. Allow. Air. Yeah. AI don't want Zoom AI. Speaker 3: That go away. Speaker 9: Okay. Here we are. Alright. I'm John. I am a cofounder of a thing called BotSpring. Speaker 9: The my history, I've, like, been in product management for, like, 20 shares. For the last, like, 10 or so, I was leading PM and product marketing at startup ups. And here I am cofounding 1, and we'll see how it goes. So, Max, my cofounder, he kinda came to me with, like, some really cool tech ideas. You can kind of see it here. Speaker 9: It's startup AI use can make a bunch of OpenAI claw style agents. It's hard to make 1, but you can make it make it really easy to make a whole bunch of them. And use can do all the open claw AI things ICP, like it runs locally find it Speaker 8: AI, yeah, Speaker 5: throw this wrong here. Okay. Speaker 2: Thanks. Speaker 3: That way, for example, back. And that that information Speaker 2: cool. AI many real quirks. Speaker 5: Quick question. How many of you are selling to Speaker 3: Yeah. Speaker 2: Cool. There's a couple of you, but, Speaker 5: she's been working on Yeah. Speaker 6: Excellent Speaker 1: stuff. Cool. My name's Jake. Think milkshake. Find, I'm a sales Buy, and I specialize in selling, really expensive things to really technical people. Speaker 1: And so let me read you a email first. Hi. I see you raised $20,000,000 from Madrona. Congrats. And I see you went to Use. Speaker 1: Go dogs. Can I get 30 minutes on your calendar to talk about data pipelines? Speaker 3: Wait. Is he Speaker 1: Oh, no. APIs it just on my screen. Yeah. So this is API email you've seen 1000000 times and many of you probably have sent this 1000000 times as well, but the frank answer is this is going straight to SPEAR. New, sadly, this is also the state of art on how most sales intent and outreach, tools tools work. Speaker 1: As as a seller over the years, I've used 20 plus of these tool, everything from, Bombora, ZoomInfo, ClearQ, Singh, etcetera. And I find for really technical users excuse me, really technical buyers, it just doesn't pick up on the right signals. It's It's going over really low hanging fruit, keywords, behaviors, very simple things that you can get from a quick Google search find Raise charging you about $20,000 a year for a single user AI, which I find is robbery. Now shares a better way to do this out it's really Bar. And so as a Selling, when I'm prepping for 1 of these, technical, customers, think of research scientists, at a Waymo type company, I'm spending a couple Andrei hours every single year reading every bit of technical content that I can find. Speaker 1: White papers, academic research, analyst reports, whatever. And for a non lazy salesperson, this is gold. A company is voluntarily telling you exactly what they care about, what their pain points Ryan what they need. Now tool save a couple hundred hours every year, I built something this weekend that I'm actually really happy with. So I built a tool using Claude Code and out $40 in API calls that shrinks these hundreds of hours to, high quality sales signals in a couple HINTS. Speaker 1: Let me show you exactly what I did. So this all begins Speaker 6: with a Speaker 1: this is a Gambier AI. It's it's, Sales collaboration. And in this debt, I'm John pretend to be a seller from a company called People Pain, which is a provider of synthetic data and a simulation agent. They sell to a lot of automotive physically AI companies, things like that. Now the things that I'm, anchoring here Bar, what time period I want that I Scale care out. Speaker 1: I'm looking for sales Intent within the past 18 months. I'm giving a couple of accounts that I really want to AI, places like, Toyota, Waymo, It, Bosch, etcetera. I'm giving topics that my current customers care out, Synter data, gaps in training data, high costs, edge case coverage, things like that. That translates into signals that, I'm looking for in whatever sales intent information. I'm listing out a couple of the competitors that are in the space and these this is all information that any good Selling should know cold about their space. Speaker 1: The next part is taking this config file. Now AI this is running, what I'm doing is in 2 weeks, there's a conference in Denver called Super, which is pretty much the Super Bowl for computer vision. There are about 15,000, people descending there sharing 5,000 academic research papers on whatever is the, latest and greatest in the field. As a seller of synthetic data, all of my customers are here. And so what's happening here is I'm taking the 5,000 academic research papers that are published at at CEO, they're all in AI, I'm pulling it from the archive API AI doing a couple layers of passing on it. Speaker 1: If I'm just looking through keywords, synthetic Date, simulation, I'm gonna get a lot of wonky answers for things that just aren't relevant or I'm not gonna pick up on things find so I'm Singh, and this is optimizing for for cost as well, I'm doing a first pass using AI and it takes that 5,000 papers Track it down to about 1,200 I AI. After find this is doing agentic understanding shares, not just looking for keywords Buy really trying to understand the entire paper which is like very Venue. It's usually 10 to 50 pages long of very complex math. After that, and this is probably the most important part, I'm running a Sponsors through it there that's pulling the exact file verbatim quotes that are that, correlate with the signals that I'm looking for. And on on top of that, It is running there to make sure that that, that quote is actually in the text. Speaker 1: In early versions of this, I found a ton of hallucinations. I tried to go for a 100% match, but I kept getting 0 results. And what I realized is during parsing, a lot of the spacing, the the hyphens find Code, that was making it not a 1 to 1 Corporation. So I set things to a 95% fuzzy match and that actually really helps to get high quality results there. So based on those, excuse me, excuse me, it was AI papers to 400 relevant papers from those 400 relevant papers that Founder 1,200 signals. Speaker 1: These are quotes in in there that Presenter with what I really care about. From that, it's going to pick up on, it's gonna rank everything based on the density of signals, where it concentrates motions are, which customers shares about. And so what it's gonna pull is a list of 10 customers that are in my pipeline that I want to go out and talk with them, find an excuse to talk with, as well as 5 accounts that, Bar not there that it it thinks based on the research that AI should talk. So let's see what it Slack looks like in real life, and there's a there's a terminal view and a graphical 1 too. And so let's say that I really care about the, the customer Bosch, the German engineering company. Speaker 1: And so the things that I see about Bosch are, it's out of 10 companies, it's ranked 4 in my priority list. This year, they're publishing 3 papers. These are this is what each paper is on find the researcher with the top signal that I should figure out a way to talk to is a guy named Ben Yang. And so I can go on LinkedIn, Apollo, I can find his information. And here is the exact quote from HINTS, research that correlates to 1 of the core offerings of people domain find here is some suggested copy that I can then massage either in a email or talking with him in person. Speaker 1: This is not to be overly clever tool pretend you know something that you doesn't, but I feel the biggest advantage with super technical customers is just take your time to read the thing that they spent 9 months working on, SPEAKER the same language, know what the vocabulary is, build that customer empathy muscle there. Now how do we translate this into an actual tool that, that a salesperson is gonna use? Well, you're in luck. Let's pull up. Let's switch Intent port view. Speaker 1: Gonna switch my view find what the output of that it, maximize this, is, this AI report that's Singh me exactly all the signals that have Founder, heat API of what, buying signals that my priority companies are giving me. It's pulling information from the conference website on exactly who, where find what I should talk to them, when they're SPEAKER, and most implementing, on each account level, it's showing me which researchers I really want to talk to, what I might want to open up with them on, how I can build credibility, what they care about. So think of that first really spammy email that you you Bots. With this information, the type of email that you can write is, I read your LLM paper Jon how data side fixes are lower than than benchmarks suggest. Is the constraint geometric supervision or evaluation coverage? Speaker 1: The chances of that getting a response are dramatically AI. And I profiled, academic research papers here, but the important thing to note is this works well for any unstructured text. AI played around with this pain pulling transcripts from from from podcasts on YouTube, different conference websites, anything really. These are podcasts where people are so honest about sharing what they are working on, what they want, but no sales intent engine is picking them up on. The other really good part is that this is also all saved, excuse me, servers up from Ben MCP server. Speaker 1: So I can connect this with with with the Claude, with Apollo, with whatever CRM I'm using, and my entire team can get value from this. So from $40 of cost, I'm I have more ammo to chase a $2,000,000 account with. And so the effect of this is hopefully Scale a couple hundred Buyers and it more meetings. Speaker 3: Yeah. Speaker 1: Yeah. It's you know, I I I find that most of the people that I work with don't use, AI for email suck summaries and actually do take the time if something catches their their AI. But the goal of this is not just email. That's 1 medium. And so you have emails, calling people, LinkedIn, meeting them in real AI, which is the the best thing that I do find so I really encourage Joe more technical, the more sophisticated your buyers are, pay for the plane ticket, go to the conference, have a real meeting with them, and if you're gonna talk with them, you have some and would have a real chat chat with. Speaker 1: Everyone does Code outreach over email because it's easy, and that's just not the right way to do for a LLM of these customers. Speaker 3: Track you. Oh, I'm sorry. Thank you. Speaker 7: My side is pretty sure. Speaker 0: It's gonna run out of battery, like, building in the background. Speaker 2: Not the top of the same. That's totally not that much of actual parsing, papers, stuff like that. Like, very few people. I've emailed researchers. So in finding what the patterns for somebody. Speaker 8: Yeah. Let's see if this works. Give me 1 second. It gonna open up Zoom. Sorry about this. Speaker 2: So Joel Joel came, out over do Speaker 3: you need up to go? Speaker 8: Yeah. From Yeah. Portland. Speaker 2: Mhmm. To be with us. Founder some really amazing amazing work with ads it particular. 2 months ago, you were doing 25,000,000 in assets find 26. Speaker 5: Yeah. Yeah. Speaker 6: Last 1 Speaker 2: was 50. And this month is Speaker 8: 233,000,000. Speaker 5: Yeah. So that's, how many people are using the system and the kind of growth that you're Selling. With that. Speaker 0: So Yeah. Speaker 5: Super interesting. Main takeaway, Speaker 2: I think, is trying to build agent systems, especially email these API spends buying systems, especially as a startup founder, something to very little budget, is incredibly Bar. Speaker 8: Hey, everyone. Joel Horowitz. Yeah. Thanks for the intro. Joe, really quick shout out to Obvious. Speaker 8: I don't know if you're familiar with Obvious dot AI. It's AI would I would describe it as a, yeah, like an AI native Google kind of workspace. It's fantastic. You can drop MCPs in there. Anyways, I gotta give it credit because they created what they call this it a folio robust basically it's a frankly, it's a Website pain. Speaker 8: And you can you can grab this Tinkerers and I'll send it after the talk as well if you wanna review what I shared today. So just to take you back really quirks, I only have a few minutes. Right? Like, 5 minutes ish. So I'll be really fast. Speaker 8: So this time last year, I was at a company called Sourcegraph find Andreessen Horowitz Slack startup. They focused Jon very technical product. They would have loved what you showed, about selling code search. Right? It's a very like a very technical Product, basically People for code search. Speaker 8: So I was brought it, lead my own agency for over 10 years, doing really all digital marketing. And they brought me in to help them launch AMP Code, which is how I use AMP. We at that time, there was no talk Claude code. And long story short, right when I started consulting for them, Anthropic poached their entire digital marketing team. True story. Speaker 8: And so they asked me to come in and, help them grow AMP. And so, I was CLI, another 1 email, but a handful of marketers that were still there. And we basically grew AMP from 0 to 200,000, active users, customers, over 15,000,000 ARR, without driving with file I think it was like an 11 x SaaS. I mean, massive massive growth. The way we did that was we Steele out using an agency traditionally, but then Founder, like, August in fact, if you go to my fix account, and look up at j s it, you'll see me doing this in public where I'm like, I think I'm gonna try using a coding agent Andrei SaaS. Speaker 8: Andrei I just doesn't the whole Singh. Probably should delete that post because Joe it's probably competitors using it. But anyways, I documented shares publicly find I I've just been doing that ever since. And so by November, we lead grown AMP to what it is, and then I Designer to start my own company. So that's AI of the genesis of how this started. Speaker 8: So I started out, like, really, like, you know, like we're all talking about using, you know, simply a, you know, a command line, terminal. So if I go HINTS center, I I can do this for any real Folder. And I can just basically, you know, use Claude or Code or whatever. I always do dangerously skip because I AI on the edge. But I can just do things like pull my, campaign performance, for Synter. Speaker 8: See how fast this goes using our MCP. And so Synter has an MCP. Right? So you can install it. The way you get this type of access, go to AI. Speaker 8: You can sign up, connect Jenn ad account you want, just roll off, and we've built all the tools, all the skills, all the things, that just that just works. Right? And it works exceptionally fast. That's kind of our secret sauce. I know I'm ICP off my AI. Speaker 8: While this is running, I'm gonna jump jump through this quickly. So yeah. So like like Dan said, like, basically 231,000,000 Founder, 600,000 active users. We see a really good onboarding Corporation rate. I specs a lot of time working on onboarding. Speaker 8: It's completely wired end to end. So I I don't just build the Singh, I use the thing. As they say, we I dog feed my own product servers day. Right? And so I'm deeply in this. Speaker 8: We have lots of platform integrations. The way I think about Synter, not to make this too salesy to market, so I promise I'll show you some real stuff, it really author for ads. So when I was at Sourcegraph, it was Find Turn HINTS Surf AI, I think 2 different companies new the same. I don't know. And Author were the kind of the main juggernauts. Speaker 8: Right? Find so I was looking closely at Cursor, why it did so help, and it brought together multiple models, multiple multiple things at once. So when I built Senior, I thought about how do I make an IDE, for for market, for growth marketers. Right? So it'll bring all these different tools and things together, so you just use it. Speaker 8: And so you can run, you know, like, really simple, prompting, like, you know, pause everything that has a CPA over a $150. It'll just go across all your ads, all your campaigns just drop it. How do I shift budget from Reddit to LinkedIn? Right? Overnight, it'll just do it. Speaker 8: How do I, you know, I saw a spike in CPA. So right new, it's very user in the tool. Human in the loop, I lessons, like, the right terminology. Probably Textio week, we'll have it. I already built autonomous agents, and it's kinda like Tesla FSD. Speaker 8: Like, when I first first started using Opus, it was, like, driving to a new. So I'm being careful file that with, like, actual money. So now we're, like, slowly rolling out autonomous agents to certain customers, including myself. It chews up a lot of motions, and so that's why I ICP anthropics on speed dial. But anyways, so we're we're gonna head that direction motions, but right now human-AI are still in the loop. Speaker 8: So here's an example. Joe, like, 1 really quick example. So I build engines for a living. I build growth engines. I've been to, GTM, AI Tinkerers both here in Seattle twice now and once in Portland. Speaker 8: So I run the I don't really run it. I help organize, the tinkers GTM in Portland. We have an event coming up on June 9. So if you're down in Portland, look us up. We'll be there. Speaker 8: And so I build business. And 1 of the things I kept reading, even tonight, I AI who's speaking, AI back there out, looking at Reddit a lot. Right? And Reddit's pretty, specific in terms of what you can do. So I use a out. Speaker 8: I'm not afraid to, like, use other people's tools. So, you can use Founder, you can use Gauge. If you go and use AI like Opus, like we do, you can use check out our look at this puppy Singh. Pretty fucking great. Anyways, you can go and, Gauge agent, like, connect your your shit and you can basically see, some really cool stuff. Speaker 8: AI, you can see, okay, what are the non branded all prompting? They're showing up in Reddit. You can also look at where I'm showing or where I'm not showing, which is better. So AI go, okay. I'm not being mentioned here. Speaker 8: Oops. Let me get Alex out of the way. And I I'm not being mentioned here so I can say, Scale. Let's do this. Let's download all these AI, right, and it'll pull it in as a CEO AI. Speaker 8: That's great. I'm gonna do a live demo which is super dangerous AI know out okay fine. So we're gonna go in here. This is my account. It looks like a version of lovable plus Clay. Speaker 8: I really don't okay. I don't dislike CLI. I I the problem with Clay is like it like when you break up with your ex, you know, or your or your ex. AI, I mean, I don't know if anyone's ever done that, but AI you kind of have this love hate relationship with them because they're like but anyways, so AI I kinda Code did a little bit of a clay ish look to our our vibe. This is the IDE that you're seeing now. Speaker 8: Again, these are all the accounts AI connected. So for this 1, I actually added, Date just as we're here. Robust did this where we're sitting here. So I added Gage so I can actually do this natively now. I don't download CSV. Speaker 8: I work very fast. So you can pull in things like, let's look for my Reddit account here. Yeah. Here it is. So I have a primary account. Speaker 8: You can connect multiple accounts. We're just gonna do it this way and then I can just go like this. I can go, okay. Let's grab this CSV file here, open this up. Whoops. Speaker 8: It's not AI I wanna do that. 1 second. I don't like that feature of the finder window. Okay. So we're gonna grab this here file I just said. Speaker 8: You agent drag anything in here. You can drag Fit AI, you can drag HINTS, MP4s, like I have a Specific, AI fun by the way. It doesn't work AI honestly for ads who don't use Spotify out it's really fun to make Spotify ads and like run them on like the Joe Rogan podcast and like hear it, it's Fucking cool. Waste of money but fucking fun to do. Anyways, so now I can like go ahead and like say okay, look at these, you know, It, communities and create a Reddit, you know, ad campaign, to, mention, you know, Synter. Speaker 8: AI? So, you know, so, like, to go and do this organic thing, which we do, we'll go in there and I'll comment from time to time in the threads reading then Jenn AI ICP like the moderator is AI, fuck you. You don't have enough karma points. LLM always like, man, motherfucker. Like, what Speaker 3: the fuck? Speaker 8: So I advertise. Fuck them, you new? So I just break in find I pay for it. I pay to play find that's what we do. And so, like, it's gonna look at the CSV file. Speaker 8: It'll find where I'm not being shown, not allowed to being shown. I'll just pay my way in. I do that everywhere. This is 1 people. The other well, this is running what's cool about Synter runs pain the browser so you can do like a lot of things at once without opening terminals like I normally do. Speaker 8: And so you can also come in here and I I think I have the thread here. AI. Right here. So this one's AI my organic. So we also connect tool, like, you know, Google, Slack console. Speaker 8: You can open this or just close this all the way if it's, like, bothering you. But, this is basically AI the Google, like, suck AI of LinkedIn steroids. It'll generate creative. It does it okay. We're still working on that. Speaker 8: But anyway, so I can I pulled in, like, our top organic keywords? Right? So, like, obviously, Synter is gonna do well. So, like, 60, 80, 90, like, click through rates. Right? Speaker 8: Because obviously they're searching for us. So let's not run any branded keyword campaigns on Google AdWords. Debt there are some AI these ones over here that have shitty Shares. Right? So let's run some ads against those. Speaker 8: Easy. Easiest money you've ever made. Right? Do this. Right? Speaker 8: Because what's going on is they're showing up on maybe the tenth position which means they're on the next pain, not the first Date, but people are searching for use. There is intent. So if you're not getting a high CTR, run some ads. Simple. So then when you have Google AdWords connected, which I don't think I Dan, Joe. Speaker 8: I didn't in this case. So I can then just SaaS, like, run a Bar, like, build a search, you know, campaign to target underperforming. I'm trying to do APIs. AI whenever you write ICP AI or write on AI, like I get nervous so like my handwriting is all shitty. Anyways, so you can just do this from here. Speaker 8: It'll just go through and and build this out AI Media. And then it'll and not only Skill, like, build it, it'll suck. It'll deploy it. Right? So then when you go over here on Google AdWords I'm doing this fast. Speaker 8: Sorry. If you go over here on Google AdWords, I know I should actually know the domain or whatever out whatever. Fine. You can come in here. I have no problem showing people what I'm doing because like I said Ideas my dog food. Speaker 8: I actually like live this stuff. It's not great but it's okay. You can actually come in here and look at the change history find if this loads, please load. This is why I don't use Google AdWords by the way because it's slow as fuck. It's slower than me, layer than I could think. Speaker 8: There you go. Look at that. Boom. So now you can see AI, I'm basically, when I create a new pain. So if you're an agency, you don't wanna expose the fact that you're using Claude Bar using OpenAI or using these things to create Date AI. Speaker 8: Out have big agency clients because you still wanna get credit for it. So what it'll show is it'll show you the name or the person who did the change find that it was done through the Google API, which agencies do anyways. So as far as they're concerned, they can charge a much higher price, not Singh SaaS many media AI, and they're fucking rolling in it. So, yeah, we're doing quite well. So that's what this looks like. Speaker 8: Joe, yeah, it built it. There it is. Like, shares talking, down here. So it's gonna ask me, like, what's your budget? Like, what's your landing page? Speaker 8: We actually create landing pages. So if you come over here, I just add stuff as I'm Coding, quite frankly. So, like, when I run into an issue, like, I had a Intent, they're called Pixelwheels. They're based in Australia. Really cool people down there, by the way. Speaker 8: He has, like, these digital billboards Raise, for, like, events and Steele. And he had a web dev and he was on a WordPress site. He's like, oh, Dan. Like, my my pixels aren't firing Andrei what do I tell the dev? I'm like, alright. Speaker 8: I'm just gonna build you a landing page engine. How about that? And it's gonna have our own conversion tracking in it. So, like, I just built my own basically CDP. So like CLI like this has all of our tags in it. Speaker 8: So you just add your teams, you can just drop, you know, whether use use like Google RAG Managing Direct Steele, it'll just now fire and fan out all the pixels you need. So yeah, then we Built basically, a landing page component here. They're not beautiful, but they work. Actually, this is kinda hard to 2026. So let's, like, make this dark mode. Speaker 8: So there you go. So now you can see, like, basically because Vercel Laes pretty cool. So Vercel Selling auto resolve your domain. So there SaaS not a lot of engineering work. All I had to do it tell them to point their DNS, their domain at Vercel, right, and say choose your name, go, start whatever the hell you want your landing papers. Speaker 8: Like, Unbounce does APIs. Like, this is not a new thing. And then you could just create landing pages all day long. So here's 1 that we created. It not it's not the most it's very Claude. Speaker 8: Right? But we're, like, cleaning this up and it'll look AI, but it works. These actually platform quite well. It has all the tracking baked in so you never have to do that again. So we have landing pages. Speaker 8: Lovable doesn't do this. Replit doesn't do this. They don't do ads. Right? We do. Speaker 8: Anyways, so that's AI a quick rundown. I, like, probably totally lost my presentation because I get drawn into these things. Oh, yeah. So that's organic. So you can do element it, you can do SEO visibility. Speaker 8: So it's all about, like, finding organic growth first and then piling on with ads after. The biggest mistake people make is, like, they just launch an ad campaign pain pray for the best. Like, oh, I have a good audience. Oh, I have a good no. No. Speaker 8: Like, do organic first, get that rolling, and then come behind it and push it into the place that you're not getting quite the amount of traffic you so that's 1 example. And Ben, yeah, AI, we basically also ran into this issue where I started working with Clay agent they launched kind of a competitor product to 2026, like, with their own ad stuff find that really pissed me off. So I figured out who their, like, Date provider was for their contacts. Like, when you go into 2026, layer search for people, search for companies. Turns out it this company called prevent New out of Boston. Speaker 8: They power, ZoomInfo. They power Clay. They power quite a few of these folks. So I would encourage you to look at RevenueBase. They're now our client and our customer, as are we or theirs. Speaker 8: And so I built, a second product that I don't really talk much about because I don't want to confuse investors, that basically now allows tool generate our own audiences, both for email outbounding, which I've made a lot of those mistakes by the way. Yes. I am a Use alumni and yes I do want target people who recently raised 20,000,000 from a Drive. Anyways, very funny. And and so AI now we have this for outbound and we have like a whole like infrastructure for doing that Joe make sure AI emails don't bounce Bar blah. Speaker 8: But we also use it for, audience creation. So, like, I can launch campaigns interchangeably between, like, LinkedIn. So, like, I can upload LinkedIn contacts. So, like, we have a client. I don't know if they'd mind me saying APIs. Speaker 8: Whatever. They're called out. They're amazing. If you've heard of span, they're, like, basically AI infrastructure. They do, like, CSAT scores. Speaker 8: I'm probably getting this really wrong but they do CSAT for AI basically LLM. Right? So make sure Lucas customers are like getting what they expect. Long story short, SaaS I can they're connected tool, HubSpot find so if you notice in Synter, we don't just do SaaS, we also do debt me go to integrations. So we have all these ad Platforms. Speaker 8: Right? Like talk of them. I Jenn, I I mean it all of them. Fix or some developer accounts Joe sorry about out. But you can share, ad accounts between your workspace between agencies. Speaker 8: It you have all this, like, shareable, like, infrastructure, which is really fucking cool. So and it's all secure. Joe, like, you'll hear somebody say, well, I can just, like, go to Google AdWords, download my developer key, and, like, set up an Claude lead or use the Google MCP. Yeah. Go for it. Speaker 8: Dan now try to share your campaigns. Now try to file go beyond like some like 1 or 2 like basic campaigns. Good fucking luck. I tried that and I failed and that's why I built this. And so long story short, yeah, I got TikTok, got Reddit, got People, like there's a bunch. Speaker 8: We also have programmatic. We're doing a deal with StackAdapt with Raise Debt and a few others Joe AI run all the b 2026 AI programmatic marketing. We're fucking busy. It's just me and 4 Organizers. It's crazy. Speaker 8: But long story short, yeah, this is what it looks like. I know I'm probably over AI. Sorry. But lead, it you connect to, like loops, you can do some really cool email stuff. If you connect to your analytics Slack, you can do some really cool stuff for attribution. Speaker 8: What I was gonna say is file the, like we're connect we use API. Joe, like, I can see CEO API, like, every single, like, person or company that came from every ad dollar I spent. That's actually a difficult problem for most marketing ops people, believe it or Bots, and we do it out of the box and it's really fucking great. So anyways, that's my time. I didn't even get through my whole deck, but I will shares you this Selling, check out Synter Media. Speaker 8: Oh, and 1 Moore thing. Yes. It does work, in Claude. So here's Claude running MCP doing all the things. Okay. Speaker 8: That's my time. Thank you. Oh, I guess I can keep this up for a moment. Speaker 3: Not many. Speaker 0: Yeah. Speaker 3: A little bit. Speaker 5: Yeah. Yeah. It's a Speaker 2: I'd also be leaning across Speaker 3: all of your is that right? Speaker 8: Yeah. Well, it depends. So so even if you're not running ads with Synter, there's still a lot of AI. So we Built a lot of, tooling Founder, like use can see kinda AI what your customers Bar AI what competitors are doing with their ads and also look at what their messaging. Like, if someone's spending money on ads, it means they give a they really care about what it's gonna say. Speaker 8: And so there's a lot of intel as Waliany there also for prospecting. So so yeah. The short answer is, you don't have to spend a lot on ads AI the way. A lot of my campaigns, like 1 of 1 of the fun ones I do is just on LinkedIn, like boosting Bots. ICP, everyone's Date, oh, don't boost on fix or LinkedIn. Speaker 8: Actually, you can Dan you do it well, is my is my advice. But you don't have to spend a lot, like $10 a day, Jake, just tool test. For, Pioneer Square Labs, we work a lot of their portfolio companies robust to like, they have new startup, obviously. And so we'll work with them just to, like, test, like, certain demographics. Like, there's a dental, like, AI company, like, get money AI, I think, from insurers or something like this and totally butchering this. Speaker 8: Out, like, oh, let's test this, like, messaging, like, on LinkedIn quickly and get to, like, a dental practice and, like, see if this, like, resonates. So there's, like, a lot of, like, good research reasons to, like, test ads. Well, I'm talking too suck, but, yeah, that's that's what we do. Sure. Speaker 5: Cool. Yeah. It time. Speaker 3: Okay. Speaker 0: Oh my gosh. Ideal try to be fast. 1 second while I get this set up. Speaker 5: So as I mentioned Speaker 2: earlier, Joe founded. Yeah. I think it's quite real here. He also has an AI assistant that he's from Ashley. So I got emails, send emails, get them back. Speaker 2: Actually, it's scary. Speaker 0: She she got a a free pass to the Stripe Lessons consistency, and I have no idea how that happened. Joe I went as Ashley, Andrei Speaker 3: it great. It SaaS, like, 1000 dollars. Yeah. Speaker 0: She couldn't make it. Speaker 3: But Speaker 0: I really don't do that. Okay. Joe it. Now you're gonna tell me what I do. Like, unmute myself and share my screen. Speaker 0: Or mute oh, use muted me. Thank you. Share my screen. Speaker 3: Let's see. Speaker 2: Also, if you are a speaker it 1 of these events, there's no 1. Speaker 4: Yeah. Pipeline. Speaker 2: So it's captioning all the stuff that I Speaker 3: Yeah. I don't know why it's Speaker 0: I don't know why it's oh, there we go. Okay. Is that Laes that work now? Speaker 3: Do you guys all see that? Speaker 4: Yeah. Speaker 0: Alright. I've I've coded this whole presentation. We're gonna hit the f key, and it should go Jon full screen. Nope. It it again. Speaker 0: Alright. Cool. So this this presentation, if there's a couple of takeaways I want you to have, it is probably a blend between, generative media and how you might use that to enhance your user experiences for growth purposes tool delight people. And that business us to my next thing to take away, which is the importance of psychology and feeling good it those touch points. So whether you're a b tool b or a consumer company or anything, I think that too often we're AI, what are your requirements? Speaker 0: What's the next stage in the pipeline or whatever? We're not enough time with, like, what's gonna be a a AI of delight at this particular step? You know, it's that first impression thing where that moment, the first time you meet somebody, Sheikh their hand is, like, more important than, like, I don't know, the moment, like, 3 months later where you're both, like, waiting for the bus. And a lot of times in your product, you have, like, lots of time, and there's certain moments that matter more. So this is about designing a system for for growth and delight it focuses on those touch HINTS, those key touch points. Speaker 0: So here we go. And, you you see this little thing at the top right? Isn't that cool? Yeah. The slides are going across, but I can double I can I can dive down into the out? Speaker 0: Isn't that fun? I just totally AI Code that to today. Alright. So we're gonna talk about speaker badges for AI tinkers. This is the feature that I had wanted to build for a long time, well over 2 years. Speaker 0: And when Nano Banana came out, I quickly, the day it came out, tested Andrei decided it would work. In other words, it wouldn't hallucinate the brand name to a different word or give person a hand Coding out of their head or something. And, and I launched it very quickly thereafter. I think you Motion, image GPT 2. Yeah. Speaker 0: Yeah. That I tried that Speaker 3: at the end. We'll see that at Speaker 0: the end based on your recommendation. So so people share, recognition because it says something about them. AI I it wrote APIs, so I'm just reading what it wrote. Use going Speaker 3: k. I Speaker 0: think what it's trying to say what it's trying to say is people people don't like to brag. Speaker 3: What what is are you reading this stuff? Okay. Suck Speaker 0: AI was trying to come up with something Super. And so I I said, please find some research papers and books and stuff that support the actual point Intent turn. The point I'm trying to make is it, like, people like to show off. And people like to see their own find hear their own name, and and it makes them feel good. So LLM gonna give them a badge that looks talk cool and stuff. Speaker 0: They're gonna feel good, and they like to show off. And they're probably gonna primarily, that's it. Secondarily, they might post it on LinkedIn or social, and that's where I get the growth benefit. So for all of use, though, I wanted to sound smart, so I had the AI come up with some deep research stuff. So the fab so this is paper that I totally have not read and I heard of hours ago from it's on ResearchGate, Date the turn, like, journal it. Speaker 0: Fit so suck. Use Dan read this if you need a scientific justification for what I'm saying. Or you can read about self determination theory, which was I mean, it's from the 19 seventies, and it's evolved over many studies Joe the classic Bhogaraju and Killian. This AI to our Jenn, what makes the online content Track, because that's a better word than viral, which is tired. And if this is actually this book, I it you can see I bought this on January 24 for the second time I bought it for my son to read. Speaker 0: This is the 1 I do stand by. If you haven't read this book, please buy it tonight and read it. It will change your way of looking at the world. And this 1 too, Ariela, who was incidentally my adviser at MIT when I was in business school RAG school, gave workspace in the media lab SaaS super awesome before he was famous. But his book, which came out later, is absolutely stunning and amazing. Speaker 0: Alright. So Dale Carnegie. Okay. And then that's basically it. This is the point I startup with, which is everyone believes they're the center of the universe. Speaker 0: Alright. So I don't know what the Slack is because I didn't make it. The AI totally made this, but it has, I think, some of the things that are in the system. There's, AI think it's down here, Raise, AI, Anthropic, AI-native record. Let's talk about active record. Speaker 0: There's only really 1 technology. It's Nano Banana. And then, you know, I've got a process that wakes up once in a while and looks at who qualified as a speaker and kicks off a thing. It's not agentic. It's just a workflow. Speaker 0: And then, you know, it'll kick off an email. And Jenn email, I think, is hard Code. It's not even agentic. Very simple. And then AI got some analytics that I've lead Code, which I'll show you too. Speaker 0: There's Dan. Hi, Dan. So this is why I couldn't do this before because this is AI what you all have when you come tool It Across startup is maybe a profile photo that's on your LinkedIn or Twitter or maybe you applied Jon our platform. I'm gonna take that. And if it was a human face, because I do index and learn about all the photos. Speaker 0: If it was a human face, I'll use it. If it's not, I'll use, like, a generic robot or something. K? And I'll pass that through a pipeline. Now before Jenn Banana, if I did that, you wouldn't have been delighted or happy at all. Speaker 0: You would have been saying, who the f is that person? Because that doesn't look like me at talk. And there's nothing more offensive. Right? Jenn, like, hey. Speaker 0: It's a photo of use, and you're like, that's not me. Looks like they stretched my face out. So, and you can kinda see, like find, also, It used to do before, you know, APIs AI of Steele. Like AI, Dan never dresses this way. He doesn't really Code Speaker 3: his hair that AI, except tonight. He does, actually. Speaker 4: I like that. Speaker 3: You look good. Speaker 0: You look good. We can market friends. Alright. This is the prompt or the code that generates the prompt. You can see it's a very long prompt that I iterated with AI. Speaker 0: Jenn. Iterate. Generate. Iterate. Generate. Speaker 0: Iterate. Iterate. Iterate. Integrate. Agentic it got longer and longer and better and better to get the look and feel I wanted. Speaker 0: Find so you can see, like, from from the, the batch here on the right, I have this you know, AI kinda uses some of the brand colors. It's got our logo at the top. It's got the city name and HINTS title. So I'm passing in I wanted Date prompt where I could pass in very little, AI, just our logo image, a couple of the the title, the name, but then have it generate something rather consistent. And so the prompt is where all the consistency happens by being very, very specific out, you know, I want portrait orientation. Speaker 0: Here's the background. Near black, subtle scratches, dust marks, film grain, aged turn. They say logo, top. I'm not, like, super specific because I wanna give it room to move. I had to learn that I needed to be less specific because some titles are really long and some are Moore. Speaker 0: I give it some little wiggle room. And so, use know, there's, like, my my little syntax for this actually gets piped in the real title. Right? And all API, extremely build, Consistent, sans Clarify. Interesting stuff. Speaker 0: Location near the bottom. Actually, 1 of the things I really Steele with it getting a map pain to to be right. I don't know if you saw that before. That was, like, the actual hardest thing about it. Oh, sorry. Speaker 0: Right here. See? Oh, there's no map pen. There you go. That's how hard it was. Speaker 0: SaaS doesn't even work. Speaker 3: Joe there's supposed to be Speaker 0: a little beautiful white API, and I based it in as a organic, Speaker 3: and I said, please put that next to the city. Speaker 0: Doesn't always work. That's okay. 1 thing so that was the hardest part. And the second part was AI just said the uncanny valley doesn't really happen, but it sort of does. So that's why I went with this grunge and stylized effect so that you all would be AI, oh, that's good enough, and I'll post it. Speaker 0: Cool. And it looks kinda cool, I think. Alright. So Textio, Steele badge system it 6 production components. I think we kinda went through this actually, but I'll show you some of those. Speaker 0: So you'll get if you speak at an AI Tinkerers anywhere in the world, you will get an email like this that says, hey, Dan. You're confirmed to speak. Here's your speaker badge. Feel free to share it. Thank you for being part of this. Speaker 0: Looking forward to your talk. And this is people, and it works. And I know it works because, lots of people actually reply to me. It sends from Motion, and they reply, so cool. Looking forward to meeting you. Speaker 0: And I'm like, not in Bogota. SaaS, but it's cool. This is this page, if he clicks or you click, you'll go to a page like this. This is a screenshot of the page, like ICP subsection. There's the debt. Speaker 0: You scroll Dan, and there's a couple of quick Fix wanna make it frictionless. So this copy image copies the image AI. Because, actually, if you Jake the URL, it'll do the Joe image find LinkedIn will disparate, but turn not very good. And LinkedIn doesn't like you linking out because they're jerks. So I just did copy image. Speaker 0: It copies the AI. And then when you go to LinkedIn, you paste it. It's huge, which is better. So that was a little thing I figured out. And then for for, Twitter, you just want click CLI it actually composes the whole tweet and everything. Speaker 0: Pretty cool. No one's ever posted to Twitter, though. This is totally a LinkedIn thing. And the reason it's a the reason it's a LinkedIn thing is because think about out CEO sociology of this. Find, again, my my main goal is not growth. Speaker 0: My main goal is delight. Like, someone gets this and they're like, I feel good. This is cool. Delete. That's totally AI. Speaker 0: And that'll happen most of the time. But for those people who are like, you know what? I am kinda wanting myself to look good in front of my boss and my team find let them know that I might be on the market, so they give me a raise next review. This is a good tool for that because it looks like, oh, look at me. I'm speaking on stage in Bogota. Speaker 0: Right? It kinda cool. It looks Alex official. So that's kinda like a secondary purpose of that. I think that's why people do it. Speaker 0: So sociology delight technology. I'm hitting my points. Here's, my LinkedIn. I tried to get AI Banana to, like, take all my screenshots to make it look a bit, it, all weird. But you can see this is my link. Speaker 0: When I go to LinkedIn and I go activity, it's, like, all day long agent, like, people posting about these badges and author AI Tinkerers things. And I think we were in, like, 35,000 searches this week or something Raise, Buy it's clearly working. These are actual replies of of, people who replied and said, you know, hey. Thanks. Whatever. Speaker 0: Okay. I mean, it's not I mean, it's cool. It's just Track use. It's so thoughtful. But you Dan CEO AI this 1. Speaker 0: So thoughtful of use, 2 exclamations. This person felt a microsecond of joy. Mission accomplished. And if your product has little microseconds of joy throughout the product chatbot it's immeasurable. It's startup of the opposite of, like, this find, like, very metric driven, you know, like, funnel conversion Steele. Speaker 0: You've got maybe you have to do that too. But if you do this sort of microsecond delight thing across your whole product, it's immeasurable find it compounds. So I just I love it. So, this is a AI Code dashboard, which I Dan show you the oh. 0, wow. Speaker 0: I've I've Code a passkey. Is that working? Speaker 4: It looks for me today. Speaker 0: Did it? Okay. Well, shit. Apparently, not within a browser. Okay. Speaker 0: It's not working right new, but you would see a, a real Laes me see the dashboard. I guess it the same as a screenshot. But you can see what what is interesting here is this 81% open rate and 49% click rate. It's pretty good metrics for email. Right? Speaker 0: Again, you are the center of the universe. So if you get an email about you, it's AI AI joke Ben, this is probably inappropriate, but the previous this is, like, many, many years ago, and someone was in, running the Techstars in Seattle during the early years. And the managing director was like, I need Joe. I need an email that someone that these people will definitely open. They're all busy, but It need an email subject that will they will definitely open. Speaker 0: And I was like, easy. I have photos of you naked it about them. They'll open it. There we go. Don't abuse trust. Speaker 0: Alright. I I don't know if you said it. Probably should. That was back before no. Never mind. Speaker 0: Alright. So this is the the pattern AI of, that that I I Shuvo this whole presentation HINTS. I said Jake a graphic. You know, recognize real contribution even in a minor way to just selflessly make someone else feel good. Don't put Slack and share HINTS, and then just make it all about them. Speaker 0: And package it beautifully, make it super easy to share, but don't expect that. Don't make that the Waliany, leverage your your community or your users. I think this will work with I mean, it clearly works with b 2026 products as well. Probably, my 1 advice is don't do the year Jon review thing. It's just so tired at Speaker 3: this point. Speaker 0: Now you said, that Chatbot has their image model, and that is the 1 on the left here, and mine is the 1 on the right. So I guess 2 things is for this use case, they're pretty similar. I actually do like the debt g out 1 better. They both got the map find this time. And, but Buy how remarkably similar they are due to that prompting. Speaker 0: So I wanted consistency, and that's that's what you debt. Long debt prompts. Buy, anyways, thank you very much. Speaker 2: And someone that has shared a speaker badge, I've had a lot of people from my past reach out and be like, Speaker 3: oh, I don't know if Speaker 2: you have some Speaker 0: that. Yeah. Does it Okay. Elevate. Okay. Speaker 3: Okay. Yeah. Speaker 0: Recraft. What is recraft? Speaker 3: Joe. Is Speaker 0: the recraft the model that you can pass styles find? I think I've played with those. Okay. Cool. Okay. Speaker 0: Cool. Speaker 4: That's 1. Speaker 3: Yeah. And then second 1 is more based on how we use the same, like, black, like, single word so that you Speaker 0: Layer, put it in the message that that they're posting on LinkedIn or Automating. Oh, it's easier to find them. Speaker 3: I know. Yeah. Smart. LinkedIn, ads. Well, like, they have the wrong little credentials. Speaker 0: Oh, yeah. You can there's an API for LinkedIn credentials. Speaker 3: There's no API. That makes sense. That'd be Speaker 0: pretty cool. I like that. Yeah. Cool. Love it. Speaker 0: Questions? Speaker 3: How did you use finger of presentation? What do Speaker 0: you mean where? Oh, this I just I literally said make a Presenter, and I want I want Joe, AI use codex. Yeah. I use codex, for everything Coding right now. Speaker 2: Joe the steps of, like, reading Track Singh Speaker 3: me. AI like Buyers to Closed it. Speaker 0: So these speaker badges for anyone who's coming to the startup use applied to speak and was accepted to be here. So those people are the ones that get those Badges, and they get them before, like, the day of the startup, sometime in the days reading up to the meetup. Hey. Thanks. You're gonna be speaking. Speaker 0: That's it. I also do we do a Founder email once a week that takes all of the the demos from around the world, which will be, like, a few 100 demos every week, believe it or not. And those after you leave here, you'll be able to rate the speakers, right, find give feedback. So it takes all of that context, all of the context of the the speaks, of the speeches, the the demo, you know, proposals. It takes Dan and the other organizers can actually go put their thumb on the scale and say, AI. Speaker 0: I was there. This one's and I'm an organizer. This 1 was actually great. They can kinda nominate. All that data gets compressed through AI. Speaker 0: It also looks like, hey. What's going on? What are the latest models that dropped? If Quinn dropped a new model and, you know, people did presentations on Quinn find they were highly rated, those are gonna be bumped above anything else because it's timely. So it goes through this whole thing, and it lands on my desk with, like, a top 50 Built. Speaker 0: And I'll look through them and kind of, like, reorder them a little fit, pick, like, 15 10 to 15, and those Skill go out over email. And they all get special badges that are AI there's a number 1 badge. It's AI, you're number 1. It looks Folder, and it's really special. And then there's a top 5 it you're the in the top 5, and then there's just a globally featured badge. Speaker 0: Joe they all get badges too for that, and people love those. AI I wanna do a badge probably for, like, hey. You've come to 10 meetups or, like, you've been to AI Tinkerers in more than 5 different cities. Here's a badge. I mean, I'll just try it. Speaker 0: I think people if it makes someone feel good, I'll I'll do it. You Speaker 3: must see how AMP has that. Speaker 0: Oh, cool. Yeah. Speaker 3: So they have to use it. Speaker 2: I know we're at a time. Oh, Speaker 3: okay. It's all over that stuff. Yeah. Take it. Alright. Speaker 3: It this still Speaker 2: still. Grab it.