AGI was achieved on a Zoom call

Shane: AGI was achieved on a Zoom meeting.

Brad: I know. I told Claude to, hey, produce an image and share it to the meeting and it said, I don't know how to do that. And before I could respond, the Grok Bot agent in the meeting told the Claude agent how to do it.

Shane: We have someone who Abhi and I have worked with in the past, we've known for a while, been working on a new thing, wants to share it to the world. Let's bring on Brad.

Brad: Hey guys.

Shane: And Brad, sometimes, you know, we see you in the comments occasionally. So you've watched the show many times, I imagine. Now you're on it. So would love to know. I mean, maybe a quick introduction to yourself, and then obviously we want to go into what you're building.

Brad: Yeah, just an old software guy at this point, been building apps and websites for the better part of 20 years, and worked with you guys in the past at Gatsby first. Well, I had a COVID stint with Amazon, and that was kind of like my last corporate thing. And they called me back to Palo Alto once they reversed their policy and I said, you know what, I think I'm gonna skip that one. And put my shingle out and do something for myself. So I've basically been consulting ever since, which, you know, with the rise of AI has increasingly just meant onboarding people onto AI and figuring out, you know, how to, really for small to medium businesses, how do they think about it and, you know, maybe build the occasional app for them and things for myself as well.

Shane: Yeah, I do imagine being in that world is different post AI. I mean all of our jobs are way different post AI. I hear some people saying we're just like slot machines hitting enter, you know, which is true to some extent and definitely true for some folks. But I do think it also opens the range of especially for people in teams that are not as invested in AI, what you can do is probably pretty impressive to them, right? If you can kind of unlock this new uncharted territory a bit.

Brad: Yeah, no doubt. It's tough, right? Because I came from a background that's, you know, software's hard, right? And now software is easy. So like I have to actually put my biases aside and be like, wait, software's easy. So how do I talk about it with these folks who might still think it's hard. So yeah, it's a new paradigm.

Shane: I guess here's my actual question. Is software actually easy now? Is it easy? Or does it just seem easy on the surface? Because it still feels to me that now you can get to 80% in 5% of the time, 2% of the time, but it still feels hard to get from that 80% to 100%, at least in my experience. How's it been for you?

Brad: I'll say it's easy for folks that are like tech- adjacent. And so in the past they might have had to go through someone like me or someone like you and be like, hey, we got Google auth. Can we add like a Microsoft auth to the login page? And, you know, we would like hem and haw and kind of wring our hands and be like, yeah, we could get that for you in like, you know, Sprint 16, maybe. Maybe next quarter. Whereas now, those kind of people that I'm working with, literally, if I give them access to the code, which I'm increasingly doing, they'll just vibe it into existence, right? And so when I say software is easy, it's really kind of the product side of me is like, forget about your roadmap. You don't need to push off that Microsoft integration out to next quarter. Like maybe we could get it done this afternoon. But it's still hard for like all the reasons I think you guys like to talk about, and I totally agree.

Abhi: It's so much easier to create software Especially things that are mapped from things that already exist, like adding auth or something. There's a lot of training and like what that should look like or be. Yeah. But we're struggling, I think software is still hard for creating net new things. Or even things that are not necessarily well trained on, like different languages than TypeScript, or, you know, complex backend problems. The models are super eager to please. You know, our team just went through this whole retrospective about we all agreed that we have AI psychosis, because we were trusting the models too much and we thought that we could literally do anything. So, one, it's still a people problem, but then we had to really face reality that no, the model was actually wrong like all the time. And we just trusted it.

Shane: Yeah, I think the taste in software engineering now is understanding what the model can probably do really well and when you can trust it. And then also understanding what the model might struggle with. And so you really need to still deeply understand those things. And I think that's a balance that we're all probably learning to strike.

Brad: Yeah, there's a, there's this

analogy I used to use back in the day: spreadsheets to space shuttles. So look, yeah, on one end of the spectrum, you know, a spreadsheet is arguably a piece of software, right? It's got macros, you could do cool stuff in there and get some automations done. On the other end, it's like we're launching a space shuttle. And I think that vibe coding and trusting a model kind of lives on maybe a similar spectrum where it's like, yeah, you could write the thrust propulsion gimbal algorithm using, you know, Claude it'd probably do a pretty good one-shot job for a space shuttle, but it's not good enough, right? But for the kind of lower-end-of-the-spectrum things, I'm definitely leaning in really hard and I'm not even looking at the code on an increasing number of things. So.

Abhi: Shane and I were like obviously in SF and we're on the bleeding edge and you know, we're part of the doomers or whatever. How are your clients? Are they AI-pilled? Are they cautious? Like what's the culture of that or the DNA of those people?

Brad: They're definitely AI-pilled, probably more so in the work context, less on the generating code side. So maybe they're they're generating proposals and things. It's just all ChatGPT or Claude like all day long. In some cases, for some clients, because I work in a lot of agriculture and small manufacturing spaces, they tend to be laggards, right? So they think about AI with kind of a default no position. And so it's sort of convincing them that, hey, this actually could be meaningful for your business. Let me, let me make the case for AI in addition to explaining this next feature that we need to build. So it's kind of a little bit of extra work trying to convince some clients that they, I'd not say that they need to be AI-pilled, but that, because I kind of believe that AI is almost just becoming like a feature, right? From a product perspective, it's just expected to be there. People want to talk to their software now. It's a little bit what are you even doing, right, if you're not talking about adding AI to your platform?

Shane: How many meat proxies you got?

Brad: How many meat proxies do I... I don't have any right now. I did at the beginning of the year. But I'm down to zero.

Shane: That's good. Yeah, well, because if you need a meat proxy, you just do it yourself, I suppose, right?

Brad: I do it myself and yeah, increasingly, I will say that I'm a part of a couple of different operations that have several meat sacks in chairs. So we're figuring that out.

Shane: Yeah. Well, you have also been building some stuff on your own as well, right? And so I know you kinda reached out a few weeks ago and... You showed me it and, you know, it looked pretty interesting and I thought why not just show the world? What's the preface?

Brad: The preface is this email right here. Have you ever seen an email like this? Yeah. So okay. So I find this to be a pretty interesting email in 2026. I'm on a paid account and this amount that my cloud account allows me is like less than a, I can't even buy an SD card that's that small, right? Like this is burrito levels of gigabytes, right? Like, so I kinda, like, I saw this enough and, you know, I was upgrading and it just was like, well, let me see how far I could get to maybe think about Zoom, a Zoom alternative a little differently that's a little more agent friendly as well. So, joining calls where people sort of roll their eyes when the meeting note taker joins the call, like a Fireflies or something. And I get a lot of that in my day to day. It's like, do we really need this to be here? I didn't expect this. It's kind of, my Zoom note taker's gonna take notes, is that okay? And so it's just kind of this rigmarole of maybe something that wasn't designed for the agentic world that we're living in. And so I just created something called Plum. And I'm in a meeting right now. Hopefully you can see my screen, but ignore the UX, man. Like we're early days. So there's some stuff here that I'm not loving, but maybe there's things I can point out. So one of them is right away that we've got a couple things here. There's people in the meeting, it's just me, but I also have like four agents in the room. Interesting thing is that these four, three of the four agents, right, Keeper, Quinn and Scout, are Mastra agents running, for this demo it might be locally, but they're running on a cloud VPS. And these are kind of personas that keep the meeting honest, right? Like, is there an agenda? What's going on in the meeting? Let me take some notes afterwards. But then an interesting thing is I've already got a slide bot in here that, you know, might be creating assets for me while I'm in the meeting. And then I've got another agent that wants to join. So Gilly is at the door and I'll just say admit. And I've got a bug right now. Gilly's joining as a human, thinks he's a meat sack, but he's not. He's just an agent. But at the end of the day, I kind of wanted to start thinking about agents and meetings a little differently, right? So one thing about this slide bot agent is that it's actually just a Grok Bot session. So if I pull this up, you can see that all I had to do to get it to join the meeting was say join this meeting. And what's interesting about this link, right, is if you go to the actual page, I thought your audience would think this was, you know, an interesting use case for WebMCP. Basically I'm telling any agent from any harness from anywhere in the world. If you hit this link, this is what I need you to do, and this is what's available to you, kind of consistent with the WebMCP protocol. You can do things like join the meeting, talk, raise your hand, leave the meeting, share assets. So basically this is kind of the one I'm trying to demo, but I might have a couple bugs. But ultimately, if I ask the slide bot in the meeting, hey, slide bot, make me a slide, right? The thing that's interesting to me, I don't know if anybody else thinks this is, but this is not inside Plum. This is not my harness. This is just talking to an agent over there on another agent platform. It's got its memory and its integrations and its API keys and things. So it's having a dialogue with me, what it should be about, the donut shop launch. And it has a link to Linear. So I can say go to Linear and, not that every donut shop tracks its projects in Linear, but they should. And so it'll go off... The AI-pilled ones.

Shane: The AI-pilled ones would track their donut launch in Linear.

Brad: So it's back here and you can see Grok Bot in this example is kind of working on it. And I just think it's kind of cool. It's sort of a yeah, you get a little indicator that it's thinking. And then eventually, hopefully if the demo gods allow, it'll produce something that we can put on screen. But you kind of get the gist, right? Like many agents from many platforms are gonna need to interact with humans, humans in the loop somewhere. And I think that might be a meeting platform. And so that's kind of the bet that I'm making.

Shane: That's super interesting. So it's no longer just agents as participants in a chat, but should agents be participants in a live meeting? Because I imagine it's not that interesting if it's just you in the meeting, right? But if you can imagine a team, and your team, they have their, you know, it's not just one agent. It might be there might be multiple agents. You know, we have a bunch of agents running around our Slack. You know, I can just think, we have one, Jan is our producer, he's a real human. You've maybe, you know, you've all kind of seen him before. Or at least heard of him. We talked about him. But we also have an agent named Vic that lives in our Slack and does a lot of producer-type tasks. I can imagine a meeting, like a pre-production meeting of this show, where Vic can also do some of these things that were maybe manual before, right? That a human did, but now, you know, we'd normally tag Vic in Slack, but could we just talk to Vic in a live meeting and assign our tasks. That is kind of interesting.

Brad: Yeah, that's the gist. And so I'll just end the meeting because I just showed you the screen and kind of the basic gist too is that I really wanted quick access to all the meetings that get recorded. I feel like for whatever reason these meeting platforms tuck away historical meetings. And so because these are stored locally, this is just scanning my local disk for meeting recordings. And I can generate AI notes. I can replay the meeting. But eventually I'm just going to add like a talk-to-your- meeting capability here. And then of course offline, I think Mastra agents could be given this context of all these meetings and do cool things with your entire kind of organizational meeting history.

Abhi: One thing I love about this is it could solve the loneliness problem with working with AI and a team. Yeah. Oftentimes people say they're siloed, and even the collaboration tools are you and someone else hopping on a Zoom, one person sharing their screen while they type and talk to an agent. And that's really not collaboration. Maybe it is, but I think the future would be you have engineers in a call expressing ideas, and the background agent could be a coding agent too that's just, you know, listening, it's making prototypes, it's showing you the tunneled URL, you're all playing with it, testing with it. Like that's really cool.

Brad: Yeah, thanks. You know, it's one of the first side hustles, to keep my mind sharp, keep, you know, whenever I have a technology I want to learn like WebMCP, I just build something now. This is kind of the first one where I was like, I gotta get it polished and get it out the door, and find somebody who's got marketing chops to put it in front of somebody because it could have some legs.

Shane: Yeah, I do really think one of the things that has to be figured out, and different teams have been approaching it, but I also see this like a live meeting as one potential use case for this. It's just also agents interacting with each other, not just humans, but in this case, humans are supervising or a part of the conversation. That's not just agents to agents. It's not just human in the loop. It is a combination of the two of them, right? Agents might talk to agents. They might talk to humans in the meeting. So the question to me is, will agents just be another participant like a human and have the same types of functionality? They can talk to each other, they can talk to us and, that's an interesting thought that I I don't think we've quite unpacked completely, but I do think there's something there we will figure out over the next, you know, months or you know, maybe even years.

Brad: Yeah, I totally agree. I mean, if you imagine you had a workspace that you were all sharing under Mastra and you walked in a kind of traditional work setting, and there was a meeting room off to the right, and every time anybody entered the meeting room, they were not allowed to see the whiteboard from the last meeting. They had to delete all of their prior knowledge of, you know, what happened outside the meeting room. Like I don't think that's the future. I think the future is agents and humans getting together with ongoing nonstop shared context with memory and some of the things you guys have been building in Mastra where there's just kind of collective jump- in, jump-out, like lightweight, right? Like, let's keep it all low touch.

Abhi: Yeah. To maintain a knowledge base. Today you have to have all these importers from Fireflies and every little knowledge source to just even make sense of it. But if your meeting software can create the knowledge base as it goes, that's like super powerful.

Shane: We have a few questions in the chat. Dev said, sorry, I might have missed it. Does the agent speak in the meeting or just listen?

Brad: Yeah, the agent raises its hand and it will speak if I grant it the stage. I don't wanna do that right now because it's a little like it kind of talks over and there's a little bit of latency that I'm trying to work through. But you know, I'd love ideas if folks have it. I've been testing the OpenAI TTS model and seeing if I can get things like hyper fast, or at, like down to the level of how the latest voice agents are working. Yeah, so they do raise their hand, but naturally, it does not work to have three, four, five agents in a room all with the mics hot. There's too much latency and they talk over each other.

Shane: Sebastian said, imagine a backlog grooming where you discuss the complexity of a bug and an agent raises its hand and points out something everybody missed, using a real voice.

Brad: That'd be tight. There's use cases like that I've been thinking of. And here's something that happened that kind of really pushed me over the edge and like I gotta get this out. When I had Grok Bot, Claude, and OpenAI can all join these meetings. They understand the WebMCP and they know what to do. Grok I told how to share assets to one of the meetings I was in. And I had Claude join and I told Claude to, hey, produce an image and share it to the meeting. And it said, I don't know how to do that. And before I could respond, the Grok Bot agent in the meeting added to the chat and told the Claude agent how to do it. And so it fixed itself in the meeting. And I was just like okay, there's something happening here. AGI.

Shane: AGI was achieved on a Zoom meeting.

Brad: I know. So I haven't let them run a meeting themselves yet, but you know, hey, maybe this afternoon.

Shane: There's definitely something here. You know, I think there will be time for us to figure out like what's the right interaction model. And it's very interesting to always see people that are starting to like poke at the edges of what the future could look like.

Abhi: Imagine not having to pay for like Jackbox games in a remote team, where you could just go into the meeting, your agent's just generating the game for you. 'Cause like, dude, Jackbox games, no offense to those fools, but we could one- shot those, you know what I mean? With Gemini or whatever. And you have to be on like a Google Meet and then everyone has to join on their phone. It's like, that's so lame. So there's like so many, I can think of like a thousand.

Shane: Bring in the icebreaker agent. We need to like do a little game, like get the team loosened up. Bring in the icebreaker agent. You do a little fun game, kick him out of the meeting.

Brad: Yeah. You guys are thinking of the fun things. My brain goes to the stuff that's executive, meaning: I'm in a meeting and somebody said that we're not going to make our Q3 numbers. And I think that we actually are. And so if I've got my CFO agent on the call, I could have that data right now. I don't need to go tabbing out of the meeting. Just invite CFO agent and say, hey, are we gonna make our Q3s? And the answer's right there. And so for me it's just like reduce friction in every level of the business, from fun things all the way to like, we're having a serious call and we actually need some real real data to come at me. Just take the friction away, right?

Shane: There's certainly a lot of different use cases. And I do know of many meetings we've been in where it's like, well, I need to consult the data, or I need to like wire up the dashboard or things like that where you are actually just... The data's just there. It just takes a while for us as humans to go get it. But if you could have an agent working on it in the background, maybe you can.

Brad: And why should you share those API keys with a new meeting platform? Yeah. Why does that meeting platform's agent get the right to go calling that data platform. It shouldn't. If you've already set up a Mastra agent, just invite it to the meeting. It doesn't need all that context.

Shane: Well, Brad, how can people, if they do want to be early, test this thing out, give you some feedback? What's the best way for them to go about doing it?

Brad: Yeah, you can go to useplum.app. There's a hella vibe-coded website there that's got a waitlist sign-up. So if you just throw your name on the waitlist, I'm polishing things. I've got all three clients built, but I'm testing on Mac right now. So as soon as I get this thing a little bit more polished, I'm gonna start releasing keys to people on the waitlist.

Shane: All right. So go check it out. Provide Brad some feedback. Be part of the early movement of actually interacting with your agents that you're building in a live meeting, not just in a single-player chat. Let's build collaborative agent teams. Thank you for coming on the show. We appreciate you coming on. Okay.

Brad: Thanks. See you guys.

Shane: Alright, see you, dude. Every week in AI, something insane happens.

Abhi: And there's so much drama. Every Monday we break it down live.

Shane: We do the news, we bring on guests building in the space, and we go deep into the stuff that actually matters. Agents Hour, every Monday, noon Pacific. Follow, don't miss it. Peace.