Kimi K2.5’s Swarm of Agents, Claude Goes Vertical, and AI Data Centers in Space | [Sidecar Sync Episode 120]

Summary:

In this high-octane episode of Sidecar Sync, Amith and Mallory cover an ambitious trio of AI developments with massive implications for associations. They dive into Kimi K2.5, a Chinese open-source model built for multimodal agent swarms that rival GPT-5.2 at a fraction of the cost. Then, they explore Claude’s new domain-specific plugins for Cowork and what it means for associations when Big AI moves into vertical markets like legal and finance. Finally, they unpack Elon Musk’s latest megamerger: SpaceX and xAI joining forces to launch AI data centers into orbit. Whether it’s AI agents that run teams of themselves or compute infrastructure leaving Earth altogether, this episode challenges assumptions and encourages leaders to rethink what’s possible.

Timestamps:

👥Provide comprehensive AI education for your team

https://learn.sidecar.ai/teams

📅 Register for digitalNow 2026:

https://digitalnow.sidecar.ai/digitalnow

🤖 Join the AI Mastermind:

https://sidecar.ai/association-ai-mas...

🎀 Use code AIPOD50 for $50 off your Association AI Professional (AAiP) certification

https://learn.sidecar.ai/

📕 Download ‘Ascend 3rd Edition: Unlocking the Power of AI for Associations’ for FREE

https://sidecar.ai/ai

🛠 AI Tools and Resources Mentioned in This Episode:

Kimi K2.5 ➔ https://moonshot-ai.com

Claude Cowork ➔ https://www.anthropic.com

OpenRouter ➔ https://openrouter.ai

Open Source AI Data Platform ➔ https://memberjunction.org

👍 Please Like & Subscribe!

https://www.linkedin.com/company/sidecar-global

https://twitter.com/sidecarglobal

https://www.youtube.com/@SidecarSync

Follow Sidecar on LinkedIn

⚙️ Other Resources from Sidecar: 

• Sidecar Blog
• Sidecar Community
• digitalNow Conference
• Upcoming Webinars and Events
• Association AI Mastermind Group

More about Your Hosts:

Amith Nagarajan is the Chairman of Blue Cypress 🔗 https://BlueCypress.io, a family of purpose-driven companies and proud practitioners of Conscious Capitalism. The Blue Cypress companies focus on helping associations, non-profits, and other purpose-driven organizations achieve long-term success. Amith is also an active early-stage investor in B2B SaaS companies. He’s had the good fortune of nearly three decades of success as an entrepreneur and enjoys helping others in their journey.

📣 Follow Amith on LinkedIn:
https://linkedin.com/amithnagarajan

Mallory Mejias is passionate about creating opportunities for association professionals to learn, grow, and better serve their members using artificial intelligence. She enjoys blending creativity and innovation to produce fresh, meaningful content for the association space.

📣 Follow Mallory on Linkedin:
https://linkedin.com/mallorymejias

Read the Transcript

🤖 Please note this transcript was generated using (you guessed it) AI, so please excuse any errors 🤖

[00:00:00:14 - 00:00:09:17]
Amith
 Welcome to the Sidecar Sync Podcast, your home for all things innovation, artificial intelligence and associations.

[00:00:09:17 - 00:00:26:01]
Amith
My name is Amith Nagarajan.

[00:00:26:01 - 00:00:27:23]
Mallory
 And my name is Mallory Mejias.

[00:00:27:23 - 00:00:41:21]
Amith
 And we are your hosts. And as always, we have an awesome array of topics for you prepared at that intersection of associations and AI. Really excited to get into that Mallory. First of all, how are you doing today?

[00:00:41:21 - 00:01:07:24]
Mallory
 I'm doing pretty well, Amith. I'm fresh off of a week in Mexico, which got extended a little longer than I had hoped. Well, I say that, but it was Mexico. So not a terrible place to get stuck with all the winter storms, but now I'm back. I'm still in the home buying process. We're actually set to close this week. So fingers crossed all goes well, but yeah, it's been a pretty busy front for me in Atlanta, but everything's going well. What about you, Amith?

[00:01:07:24 - 00:01:12:01]
Amith
 Well, congratulations to you guys on the first home purchase. That is exciting.

[00:01:13:02 - 00:01:41:06]
Amith
 That is really exciting. And yeah, I don't think anyone probably felt too bad for you for being stranded in paradise for an extra couple of days, but it was kind of a mess last week, travel wise with the winter storm heading south when cold weather is heading to the southern part of the United States. It tends not to be a good situation. It hit New Orleans a little bit, but mostly Texas and Alabama got hit pretty hard, parts of Florida, I think as well. And of course Atlanta and that whole area.

[00:01:41:06 - 00:01:56:24]
Mallory
 Tennessee, yeah. I mean, overall, thankfully it didn't hit Atlanta too hard, but I think people just weren't sure. And then ice on the roads, planes flying in, cars driving. I think it was probably canceled preemptively the flight, but I made it home. We're all safe, so that's good.

[00:01:58:00 - 00:03:10:10]
Amith
 Yeah, it wasn't great timing for us at Blue Cypress because we had a hackathon last week, all week. We had six days and we had 14 people in a house in Florida. It was on the beach and we enjoyed that, but it was also about 14 hours a day or so of software development. It was super fun. And unfortunately it resulted in some delays for some folks. One individual coming from the West Coast was two and a half days delayed. And he had to go to the airport multiple times, sit around for four to six hours each time before he was able to finally get on a plane and fly here. And it's one of those things that you think, well, hopefully airlines will get better. Maybe with AI or maybe some leadership changes, but this individual was waiting for an American Airlines flight and Americans like, well, it's about 30 minutes delayed. Oh, it's an hour delayed. Oh, it's 15 minutes delayed. And they just kept pushing it back just a tiny bit at a time. And then ultimately between six and eight hours later or whatever, they'd cancel the flight. I don't know if that's ever happened to you, but it's unfortunately happened to me too many times with United, with American, with other airlines. But I think it's better for them to just say, listen, we don't think this flight's gonna be getting off the ground today. So you probably should just go home. That's bad enough, but it's a lot worse to get that like incremental feed of the bad news, right?

[00:03:10:10 - 00:03:11:01]
Mallory
 Yeah.

[00:03:11:01 - 00:03:39:06]
Amith
 So anyway, this guy was a trooper and he kept working virtually while everyone else was already in Florida and eventually got down there. And we cranked out a lot of good stuff. Our theme Mallory was audio. We focused the entire week for all the developers on audio. Well, not literally the entire week. The first half of it had to be audio. And after that, people kind of chose their own projects, but it was extremely, extremely interesting, super fun. And we got a lot of great stuff done.

[00:03:40:08 - 00:03:51:06]
Mallory
 14 hours of software development a day. You said that's super fun. We might have some listeners thinking, eh, I don't know about that, but what came out of it? I'm assuming you have some good outcomes about 14 hours a day of software development.

[00:03:51:06 - 00:04:47:05]
Amith
 Well, if you have a bunch of people that all believe in what you're doing and are excited by it, and it's super interesting work and you're working really hard, sure, but you're collaborating in interesting ways, you're face to face, which these days, most of us are not face to face all the time. Although we just opened a new office down here in New Orleans, which is awesome. We've got a bunch of people in the office every day now. It's not mandatory. We're still a fully remote company, but people just love coming in and having a collaborative space. But any event, having these different developers come together across the whole enterprise, we've got eight different subsidiaries, and then we also have partner companies that we have non-controlling or minority positions in that we've made investments in over the years. And so different companies all came together. It's really this hodgepodge of ideas, and that made it super, super interesting. Because when you see different people experimenting with different concepts, you see, you just get a lot of energy out of that if you're into this stuff.

[00:04:48:07 - 00:05:05:24]
Amith
 So it was super fun, and the days were long, but the enthusiasm and the energy is what fueled everyone through it. And we got out on the beach a couple times. We threw the football around. We had a bonfire one night, which was fun. So you get a bunch of programmers on the beach when the weather's cold and you light a big fire and you hope everything turns out all right. That's pretty much what we did.

[00:05:05:24 - 00:05:13:11]
Mallory
 Well, that sounds like a blast. Can we get any sneak peeks of perhaps any audio agents that might be coming to our audience soon?

[00:05:13:11 - 00:05:38:04]
Amith
 Yes. So for Sidecar, Sidecar was actually one of the guinea pig companies that we said, hey, what if we could produce an audio agent for Sidecar that was this incredible agent that lived on the Sidecar website that could help people in their journey, wherever they're at. We have a ton of people come to the Sidecar website that are trying to learn about our certification, the AAIP or Association AI Professional Certification.

[00:05:39:04 - 00:05:42:05]
Amith
 We have a lot of people going through that process now, which is super exciting.

[00:05:43:06 - 00:08:02:24]
Amith
 Way more people come to the website all the time looking for information and not getting it probably because just like all websites, we have funnel challenges with getting people to the right information and the right way for them. And so the audio modality we thought would be a helpful thing to have available. So we built out an audio agent that you can go to. It's not yet live on the Sidecar site, but what we probably will do, Mallory, if you're game is we'll preview it on this pod in the next couple of weeks. On games. So we will invite this new agent, which has yet to be named, but I'm sure a new name will be assigned to it. This agent will join us on the call and we'll have a chat with this agent all about the AAIP certification and all the other great things that it'll know about. Anyway, the agent, what it does, it's kind of like having a Zoom call actually. You don't see an agent, at least not at this point. You don't see kind of a cartoon or whatever of the AI, but you hear the agent. And the reason I call it like a Zoom call is the agent can kind of share its screen. So let's say that you want to know about, hey, what are the requirements to become certified? Well, if it's, sure the agent can tell you that verbally, but wouldn't it be great if while the agent's describing that it could also pop up a slide that showed you all the requirements to be a certified AAIP and questions that follow up on that. And what about pricing? What about this? What about that? So the kinds of questions people typically have, certainly at the front end of their journey, which is what I'm describing, but at any point in time, someone says, well, I've just completed this course. What's the best next course I can take? And the agent will have awareness of who you are if you're logged in. If you're not logged in, it won't, but that agent will go live in the next few weeks. And it is very much an experiment, so we will label it as such. Audio AI technology is both extraordinarily impressive, but also very, very early. So we at Sidecar obviously want to live in this so-called glass house of AI adoption, where we want to put stuff out there early. We obviously want people to have a great experiment or experience, but we want people to realize, it's also for now an experiment, just like everything that we do. We all have to be willing to go out there and see what happens. So we're going to tell people, hey, if you have problems with this, please let us know, and we'll see how it goes. But I'm super pumped about it. And that in turn, of course, will likely become a product within the broader Blue Cypress ecosystem at some point, but we're going to make sure that it actually works, and we're going to make sure that it works really, really well for Sidecar. And if it works well for Sidecar, it probably will work well for a lot of people.

[00:08:02:24 - 00:09:06:09]
Mallory
 I love it. Everybody knows who listens to this pod, we at Sidecar, we like to be beginning pigs and to refine and to optimize until we can create something incredible, hopefully for the association community. So we're looking forward to having, oh, I know it's going to have a good name, whatever it is, so we'll keep you posted on that on the podcast. Moving into today's topics, we've got some meaty ones. First, we're talking about ChemEK 2.5 and the rise of agent swarms. Then we're talking about Claude's new domain specific plugins and what it means when big AI comes for your industry. And then finally, we've got to talk about the hot news item, the SpaceX XAI merger, and Elon Musk's vision for AI data centers in space. So starting off with ChemEK 2.5, this is the latest AI model for Moonshot AI, a Chinese company founded in 2023. It was released on January 27th, 2026, and it's an open source model with around 1 trillion total parameters, but only 32 billion are active at any given time, which makes it efficient to run while still being extremely capable.

[00:09:07:12 - 00:09:23:07]
Mallory
 Why does this matter? Well, this continues the trend. We've been tracking of powerful open source models coming out of China following DeepSeek and others. These models are free to use, modify, and build on, which changes the calculus for organizations deciding whether to build custom AI solutions.

[00:09:24:08 - 00:09:51:09]
Mallory
 So let's talk about this agent swarm concept. I really like the name. Instead of a single AI agent working through tasks step by step, ChemEK 2.5 can break complex work into pieces and spawn up to 100 specialized subagents that work in parallel. The model creates these roles and subtasks on the fly. They're not pre-programmed. Moonshot claims this cuts execution time by up to 4.5x for complex workflows.

[00:09:52:13 - 00:10:08:20]
Mallory
 ChemEK 2.5 is multimodal from the ground up, trained on 15 trillion tokens, mixing text and visual data together so it natively understands images and video. You can actually upload a screen recording of a website and it can recreate the interface and clean working code.

[00:10:09:21 - 00:10:40:04]
Mallory
 Now in terms of benchmarks, Moonshot compared K2.5 against GPT 5.2 and Claude Opus 4.5 across more than two dozen benchmarks. Their headline claim is that K2.5 scored a 50.2% on humanity's last exam, one of the industry's most difficult evaluations, the highest score among models tested while costing 76% less to run than Claude Opus 4.5. It's a performance plus efficiency argument, we match or beat you and we're way cheaper.

[00:10:41:07 - 00:10:59:10]
Mallory
 Pricing is aggressive, API access is about 60 cents per million input tokens and about $2.50 per million output tokens, significantly cheaper than US frontier models. Amith, man, there's a lot to unpack for K2.5. Where do you wanna start? What was your initial thoughts when you saw this drop?

[00:10:59:10 - 00:11:14:08]
Amith
 Well, I jumped online, I went to the Moonshot AI website and I checked it out, just had a conversation with it and it was a very impressive model. Granted, I didn't go super deep. One thing we have to keep in mind is where you run your AI models can matter depending on the sensitivity of the data that you're providing.

[00:11:15:10 - 00:13:12:05]
Amith
 So the current availability of this model, it is available on a handful of US based inference providers, Fireworks AI and a few others. You can go to OpenRouter if you're looking for a place to find out like where different models are available to run or so-called inference providers. You can go to OpenRouter, which is an AI platform that actually allows you to very easily use a wide variety of different models across providers. But you can see where these models are run. Moonshot, I'm not sure if they have data centers in the US or exclusively in China, but you need to keep in mind that thing we've talked about before about the difference between an inference provider and model developer and where you put your data does matter. So just be thoughtful about that. That being said, I would say that this is truly an extraordinary advancement. It is very much on par from all the benchmarks I've seen. We have not yet run it through our own internal testing across various Blue Cypress products like Skip and Betty and so forth, but I expect it to perform quite well. Kimmy K2, there was an update to a release in September of last year that we did run through some of our most complex use cases like with Skip, and it performed extremely well. It's up there with like the Opus 4.5 caliber almost, even at that point. So 2.5 is a natural jump. I expect it to be really impressive. So really pumped about it. I think that the trend line around non-US based models and specifically Chinese open source models, we got a lot more. GLM 4.7 came out recently. I'm sure there will be a GLM 5 that is incredibly good. DeepSeek is poised to drop their next major model technology very soon and on and on and on. There will be more and more of these things. The QWEN series is still on QWEN 3. That's from Alibaba, which is no slouch in developing this stuff. I'm sure there'll be a QWEN 4 series. That is incredibly impressive too. So all this is very exciting. It just means more choice, and it means better results for way less money. And usually these models are also faster.

[00:13:14:01 - 00:13:30:22]
Mallory
 I want to talk about the agent swarm concept, because we've talked about on the podcast before this framework where you have perhaps a lead AI agent and then under it sub-agents that go out and do different tasks and then they come together. Is that concept of an AI agent spawning up its own sub-agents, is that new? Is that novel?

[00:13:32:04 - 00:15:02:18]
Amith
 It's not a new concept. It is new to my knowledge at the model level. So in an AI system like Cloud Code, for example, which is extremely popular, Cloud Code has been actually since very early versions of it last year, spawning sub-versions of Cloud Code to solve specific pieces of a problem, or in some cases, to solve a problem in parallel multiple times and then choose the best answer. So Cloud Code has been able to do that. OpenAI's Codex tool, I believe, has been able to do that. And actually preceding those tools, even going back to Google's first deep research tool, that's exactly the way it worked. It would spawn a whole bunch of sub-agents that would go off and research bits and pieces of your request. So if you say, I want to study the best TVs for my house, and you do that through the deep research mode-- and this is now available across all the major AI tools, Chatchee, PT, Cloud, and Gemini-- they'll go out. And first of all, they'll look at the problem. And they'll say, well, I need to research what are the top brands. And then they'll say, OK, well, the top brands are Samsung and Sony and blah, blah, blah. And then it will go off and research each of those individually with sub-agents. In fact, we do that with member junction within our research agent, which is part of our open source platform. It does literally the exact same thing. So that concept has existed for a while in what I'd call AI systems. But the novel thing here is that the model itself is doing it, which means that in the model, if it's capable of intelligently spawning essentially sub-tasks,

[00:15:03:19 - 00:15:48:04]
Amith
 that's actually quite an interesting development, because that potentially makes the model just inherently more flexible. It's also generalized. In the case of the examples I provided you, those were all pre-thought-of paths, meaning that the person or people who built the system said, oh, Cloud Code, when it makes sense, you can spawn sub-tasks that do these kinds of things. Whereas it seems as though the model architecture in K2.5 is capable of generalizing, coming up with new types of specialized sub-agents, and that I believe is a novel aspect of it. This is an area that I don't have extreme depth in, so there could be other people doing things in parallel that are similar to this. But it's certainly from a mainstream perspective the first time I've seen this in a model architecture.

[00:15:49:16 - 00:16:20:02]
Mallory
 I want to talk a little bit about this visual to code idea in the demo from the founder of Moonshot AI. He talks about a screen recording of a website, giving that to K2.5, and it basically being able to replicate that with code. Now, that sounds great, and perhaps other coding models do this already. But, Ami, for you as someone who is very deep in the trenches of AI code generation, is it that easy that an association leader can just perhaps record a website that they really like and then replicate something like that at the snap of their fingers, or is it more complex?

[00:16:21:03 - 00:17:46:12]
Amith
 Well, I think it's a useful capability. So there's two sides to conversations with both humans and AI. There's inputs and outputs, and there's different kinds of inputs and outputs that we can all have. So there's the ability for a model, in this case, to accept inputs that are text, as well as to accept inputs that are image, or in this case, image or video, which are essentially the same thing, but video obviously is a series of images, possibly that has audio, although I'm not sure if they're actually listening to the audio. But the visual side of it is super interesting, because the depth of information that you get from a single image, and certainly from a video, is quite substantial. And so if you want a model to be able to truly understand a problem you're trying to solve, if you can provide an image along with text, that can be really interesting. And so a common use case for this in coding is if you're working with Cloud Code or Codex, you might take a screenshot of an application you're working on and say, here's something that's not right. There's a bug, and here's what it looks like. And by providing that screenshot along with the description, possibly some other technical information, those AI tools are dramatically more effective at solving the problem. And that would be true for you and I, right? If I said, hey, Mallory, can we work on making this blog better, and I gave you an example of another blog that I like better for whatever reason, that would be pretty helpful to you, I would think, as an input, right?

[00:17:47:19 - 00:20:30:22]
Amith
 So it's the same kind of idea. And so the essence of it is, is by being able to reason in parallel over both image and text, those additional input modalities just can help the model become smarter. Now, KimmyK2.5 emits or outputs text tokens. So it's generating text, which is actually very similar to a lot of other models that are out there that can take in images, but only output text. So K2.5 isn't an image generator for clarity, but it can take in images to understand the world better. And that's a very powerful capability. One of the things I would add to that is, in the context of doing any kind of work that we're talking about, of any kind, it could be marketing work, it could be coding, of course, this is helpful. It's also really powerful for these systems to become smarter about images, because that's one of the ways they can actually start to use our computers more effectively. So we've talked about computer use on this pod. For listeners who aren't super familiar with that, the basic idea is that you can kind of hand over the reins to a web browser, or if you want your entire computer to an AI and say, I have a task I want you to accomplish. I want you to go into my AMS, log in, I want you to go to a search feature, search for this member, pull up their latest information, export it to an Excel file, take that Excel file, and then I want you to turn it into a PDF in this format, or something like that, which is a multi-step process. And traditionally, that would require either an API if it was available, or if you'd want to try to automate that with classical tools, it would be an extremely involved engineering effort to try to make that work, and it would be brittle, because the minute the developer of that AMS changes something in the screen, it would probably break that classical automation. So what the AI model does is basically what you and I do, Mallory, the AI would say, hey, I've been asked to solve this problem, and I'm going to go figure it out. So we'll go to the AMS website. It'll log in, and then it'll take a screenshot. It'll say, OK, what does this thing look like? Oh, it looks like this and this and this, and oh, there's a search-- it looks like there's a search box at the top. OK, let me click in that. Let me type in a search phrase. Let me hit Enter. And then now let's take another screenshot after a second or so, because it might take a second or two for the results to be displayed. Oh, let me click on this result, and then take another screenshot. So this is called a-- it's essentially like a language vision action model, where it takes a combination of language and images, and then it produces actions, which then you actually go execute the actions and then give it another image. But the long story short is the reason that's interesting is, as models become more inherently fluent in imagery, they become better and better partners for doing computer automation. And that opens up the door to a lot of powerful things.

[00:20:32:01 - 00:20:39:12]
Mallory
 That sounds incredibly inefficient. I don't know why in my mind. The screenshot click, the screenshot search bar screenshot-- what's the next step of that?

[00:20:40:18 - 00:20:42:16]
Amith
 Doing it a lot faster is the next step.

[00:20:42:16 - 00:20:46:07]
Mallory
 OK. It's still going to be a screenshot. Will that be the process?

[00:20:46:07 - 00:22:09:08]
Amith
 I don't know that it will be forever. I think something that's more inherently fluid would be way, way better and more efficient. But in effect, the computer has to know what's on the screen. It has to find-- when I say the computer, I mean the AI. It has to know what's on the screen somehow, and then it has to reason over, OK, this is what I see. What do I do next? It's just the way our brains work. We're just working at a much, much higher, faster pace, because we're seeing the images at 60 or 100 images per second, and then we're able to very quickly make a decision on what we want to do. And so the AI systems are very, very slow compared to us. It might take even Claude 4.5 Haiku, which is quite good at this. And you can try it out. They have a Chrome plugin that does literally what I'm describing. It takes it, I don't know, somewhere in the neighborhood of 2 to 5 seconds for each turn, for each decision. So if you were to do that yourself, it would just take you forever to get anything done. But remember, we're on an AI curve, and so this stuff is going to get faster and better at a ridiculous pace. And so while it's maybe impractical for real time usage right now, if you gave it a task and it took it an hour or two to go solve the problem, that you might have been able to solve in 10 minutes, but it took the computer an hour, two hours, three hours, you don't really care, because you get back the results and it took you no time. So there's still value now, but when these things can go at our pace, that will, of course, be better. And very soon thereafter, they'll go to pace that we can't comprehend.

[00:22:11:06 - 00:22:15:24]
Mallory
 And we've got to be ready for that. That's why when things are slow, we're talking about them, we're thinking ahead and we're planning.

[00:22:17:04 - 00:22:23:03]
Mallory
 Moving to topic two for today, we want to talk about Cloud Co-Work plugins and Big AI going vertical.

[00:22:24:04 - 00:22:42:02]
Mallory
 On January 30th, Anthropic, the company behind Cloud, which we were just speaking about, released plugins for Co-Work, their enterprise AI product. These are domain-specific plugins designed to make Cloud work like an expert in particular fields, legal, finance, sales, marketing, product management, and even biology research.

[00:22:43:07 - 00:23:26:06]
Mallory
 Co-Work was launched in January and it's Anthropic's AI collaborator that runs on your local machine, can access your files directly and coordinate multi-step tasks. Think of it as Cloud that can actually do work rather than just advise on it. So it can create spreadsheets with formulas, PowerPoint decks, formatted documents, and more. Now plugins bundle together skills, connectors to external tools like CRMs or document systems, slash commands, and subagents. For example, the sales plugin can connect to your CRM and knowledge base, learns your sales process, and provides commands for prospect, research, and follow-ups. The legal plugin handles document review and contract analysis, and the finance plugin builds models and tracks metrics.

[00:23:27:07 - 00:23:39:04]
Mallory
 Now, there was an impact when they released these plugins. Shares of Relics, owner of LexisNexis, Thompson Reuters, owner of Westlaw, and LegalZoom got hammered in pre-market trading.

[00:23:40:10 - 00:24:21:13]
Mallory
 These plugins, as a note, are also open source and customizable, which I thought was interesting. They are file-based on GitHub and designed to be modified. You can swap connectors to point at your own tools, add your company's terminology and processes, and build entirely new plugins, no code or infrastructure required. The bigger question here I think that we're going to talk about is, do we see a dedicated legal product line next? This appears to be the beginning of big AI moving systematically into vertical markets, which could be a very interesting conversation for you as association. So Amith, what is your take on a legal plugin for Clod? Is this just a cool add-on feature? Is this something associations really need to be watching?

[00:24:21:13 - 00:24:28:17]
Amith
 I think co-work, broadly speaking, is a very big deal, and plugins are a very natural extension of it, super smart.

[00:24:29:19 - 00:25:59:16]
Amith
 So it's interesting because Clod code, before we talk about co-work, it's worth maybe venturing into Clod code for just a second. So Clod code has taken the world by storm from a coder's perspective. But actually, a ton of non-programmers have been using Clod code as a general purpose agent for quite some time. In fact, here at Blue Cypress, our chief operating officer is not a developer. He's a very tech savvy individual, but he's not a programmer. He's been using Clod code for months and months and months to do a lot of the things that people are talking about using in co-work. It is a little bit harder to use. If you're not very computer savvy, it's kind of clunky, and it's not the easiest thing to understand. Even to install it takes a bit of work, but it is extraordinarily powerful. And so because of the power of Clod code, people who are not software developers have been using it to automate all sorts of workflows. And so co-work is essentially Clod code for the masses. It allows anybody to be able to run a genteck work in the desktop very, very easily. And these plugins essentially are a combination of a few things. First, they're pre-built instructions. So if you have a plugin for common tasks you're going to do, you can bundle into that plugin a bunch of what are called slash commands, which are basically shortcuts. So imagine commands like in the world of sales or customer support, very common things that you need to do. Research a prospect or look into a complaint. Or in the case of association work, perhaps there's something like search for a meeting venue.

[00:26:00:17 - 00:27:18:06]
Amith
 Lots of common tasks. You can put these slash commands essentially are instructions pre-built that tell the AI a bunch of context about what you're trying to solve for. And they can also include directions to use certain skills. And a skill is kind of the consumer version of describing what we've talked about here before, MCP servers. And MCP server is the technical term. A skill essentially is instructions combined with MCP servers. Because like for example, if I say, hey, look into this particular order from this customer and it's just text, the AI has no ability to actually connect to my CRM and find out what the order status is. But the skill would typically package both the context of how to do the work and then also access to your Salesforce system or your Microsoft CRM system or whatever you're using. So that's the idea is that you can have these plugins that combine a variety of skills and connections to tools. And the net result is actually I think quite extraordinary in terms of its impact. This is going to replace first of all, a lot of manual work, which is exciting. And then through these connectors, these MCP servers, potentially replace the need for a lot of custom software. So it's quite exciting. It empowers just regular office productivity workers.

[00:27:19:10 - 00:27:47:01]
Amith
 And this is the trend line we've been talking about for years here is that these AI tools basically level the playing field, allowing anyone to build. And that's what we're doing here. We're talking about building things. It's just, Claude has essentially or Anthropic has wrapped that power with a fairly user friendly interface. So I'm sure all the labs will have something like this very soon. But this concept of basically building your own agents that do stuff like this is it's gonna become normal. It's just gonna come to the expectation very, very soon.

[00:27:47:01 - 00:28:11:07]
Mallory
 Amith, from an entrepreneur perspective, if Claude Co-Work had just dropped an association plugin that helped associations do what they do, but better, would you be concerned? And I'm thinking particularly legal software, obviously we saw the market share or the shares drop in price. So what, I guess from a software developer or product developer perspective, what do you think about that?

[00:28:11:07 - 00:28:31:02]
Amith
 I think it's amazing. I think that if you look at it from the viewpoint of a new platform has emerged or is emerging, and you have the opportunity to build tools that can help people solve new problems and solve old problems in new ways, you should look at that as an incredible opportunity. It's like the emergence of the internet or the emergence of mobile computing. It's a new surface area that developers,

[00:28:32:06 - 00:31:46:06]
Amith
 whether they're classical software developers or just people with ideas, builders would probably be the better term I should use. These builders can simply create things. And these builders include people within the association who just have ideas and don't know how to do it, but now they have a tool that can do it for them. And it includes the typical independent software vendors in the vertical who want to build solutions that will help their customers solve problems. So I think it's extremely exciting. I think it will absolutely crush a lot of kind of edge case or more like fringe fluff type tools that people use because these AIs are gonna be able to create a lot of those peripheral features themselves. And I think the core automation that people are gonna need in their CRM, in their AMS, in core systems that handle key business logic, you'll probably want that in some kind of a product that is maintained and upgraded and looked after essentially by professionals, at least for I would think the foreseeable future. But all that's up for grabs at this point because a motivated organization that has the right skill set internally could build many of the things they've traditionally relied on external parties for with these types of tools, which a combination of cloud code, cloud co-work. Our perspective is that the most critically important thing for associations to do is to get their data house in order. And we've said that a lot because you have your data in all of these distributed places. You have it in an AMS, you have it in a CRM, you have it in an LMS, all these different places and probably more spreadsheets than you can count. Your data governance is a real challenge. And so you're not gonna solve necessarily that problem anytime soon, even with AI. But what you can solve for is bringing that data together into unified location so that you can operate against it in a kind of sensible way. So rather than building your AI workloads to try to get cloud to connect to 50 different systems, we would recommend that you should unify that data into some kind of a data platform and then have the AI tools work against that data platform because then you have security, you have some level of data governance against it and you also have the ability to really run with this just like we're talking about here. In the case of Member Junction, which is the open source platform we maintain for the community, it's totally free. The Member Junction data platform does all the stuff we're talking about and it has built in MCP server capabilities so if you have a hundred different tables of data and as well as unstructured sources plugged in to Member Junction, if you wanna connect that directly to your cloud co-work, it's literally as easy as just a couple of things that you tell cloud and off you are, you're connected, you authenticate obviously and then you're in and then cloud co-work can interact with all of your data across the organization and you take advantage of all this stuff. So you see the tools that are talked about in the mainstream like HubSpot and Salesforce and many organizations have those two particular tools but most of the software that runs the business of the association are much more proprietary older systems and these are things that might house the most important data so if you get that data over to a modern data platform of any type and then connect that data platform with these AI tools, that's when the opportunity really will explode.

[00:31:47:19 - 00:32:25:19]
Mallory
 I wanna divert just a little bit, go on a tangent because we've been talking about data platforms since I started working with the Greater Blue Cypress family that was in 2022 and at least from my perspective, I feel like it has been a bit slower to be adopted in the association community and maybe that's picked up recently, you would know better than me Amith but why do you think that is that an AI data platform being so essential, like if you want to do any profound deep AI work within your organization, member facing work, having all of that in place is essential yet it seems like organizations are a bit slower to adopt it, why do you think that is?

[00:32:26:20 - 00:33:18:20]
Amith
 There's two things I'd share in response to that. First, everything moves slow in the association universe, it just does and associations are the first to admit this and associations by the way are not alone in that respect. Lots of organizations, if they've been around for a long time or if they have a lot of different constituents or stakeholders, things move slowly. I do think that many associations are doing their very best to move much more rapidly than probably I've ever seen them do in adopting the technology but there's that piece, it just takes time for momentum to build and for people to run these experiments. Even the associations that are more forward looking still are held back in terms of their speed by a lot of governance challenges, a lot of other things. They have tons of really creaky old, difficult to integrate custom systems and systems that are purchased systems that have been customized. There's a lot of pain around that that slows things down.

[00:33:19:22 - 00:34:33:09]
Amith
 But I would say that your other comment, this part of what you said, that data platform adoption has been picking up is definitely true, it's slow but it's been picking up in a quite exciting way. It's actually perceived to be a bigger deal than it is. It sounds like, oh my gosh, I need to go and implement like a new thing and people think of it like an AMS or something terrible like that. But really what this is is just putting in place a cloud-based database and then setting up some pipes to suck the data in on a regular basis from your other systems. That's what's involved, it sounds pretty straightforward when I describe it that way. And it really actually is, it's not super complicated. You're not gonna run your financial reports from this data platform, you're not gonna do membership and yields from this data platform, you're not gonna sign people up for events. So you're not doing the right operations, if you will, the things that modify the data, the things that deal with financial transactions. You have systems for all of that. And they generally work, not always work perfectly but they obviously process the business of the association. What we're talking about is sucking the data out of those systems on a read-only basis and doing it on a high frequency basis, like could be every few minutes, it could be once a day, into neutral ground, into a data platform that you own and control. And then that data platform can then wire up because that data platform,

[00:34:34:13 - 00:34:49:24]
Amith
 my assumption is, is that it's a very contemporary, modern AI forward data platform that has MCP server built in and all this other stuff we're talking about. That's a matter of literally a handful of clicks for your end users to be able to connect to these tools and then make them work with all this amazing AI stuff.

[00:34:51:01 - 00:36:13:07]
Amith
 The last thing I'll say is, it's, think of it this way, if you took the most brilliant, best trained, customer service rep from the most amazing company in the world that's extraordinary customer service, pick whichever brand that is in your mind that has the best customer service in the world, you can say, if I could just pick out their very best customer service rep who has the best training, the best education, et cetera, and drop them into a chair in my association, they'd be able to provide amazing customer service. And I would say that probably is true, but what would happen if that person didn't have access to your database? They couldn't do a whole lot. And that's the problem with Claude or open, open AI stuff or anything else, is if you don't give them access to these tools, they're just as effective as that employee who might be brilliant, incredibly well trained, but does not have access to your systems. And because these association systems are often so difficult to interact with, it's just so much easier to have a single surface area to connect all of your AI systems to. It also provides you a lot of security when you do it that way, rather than just having individual connections to all these different things that becomes a governance nightmare, which by the way, is one of the reasons associations have moved slowly. They're worried about their data, which is an understandable concern, but it's a solvable challenge, especially if you take an approach like I'm describing. So people are starting to get this. People are really starting to get this at scale. We're seeing exciting adoption with member junctions specifically. There's tons of other ways to do this as well.

[00:36:14:13 - 00:36:39:04]
Amith
 So overall, I'm very optimistic about it, but I think it's one of the most critical things people need to do. It's way more important than replacing an old AMS with a new AMS because it's still gonna be an AMS. No matter how many times you replace your AMS, it's gonna be an AMS. And so, it might be a better AMS. It might have certain things built in, but it's still gonna be an AMS. So in comparison, AI is not waiting for you. So you need to focus on that is what I'm trying to say.

[00:36:40:09 - 00:36:48:12]
Mallory
 Well, hopefully this year in 2026, we see more associations implementing AI data platforms. Maybe we'll have one on the pod, do a little interview, that could be fun.

[00:36:49:18 - 00:38:12:19]
Mallory
 All right, moving on to topic three. On February 2nd, Elon Musk announced that SpaceX has acquired his AI company, XAI. The transaction values SpaceX at $1 trillion and XAI at 250 billion, with the combined company aiming to go public at about 1.25 trillion, which would make it one of the biggest IPOs ever and immediately place it in the top 10 most valuable, publicly traded US companies. So what is this combining? We've got SpaceX. That's Rocket's Starlink satellite internet with 9,000 satellites and about 9 million customers. And then we've got XAI, which is the Grok chatbot, Grok with a K, not Grok with a Q, competing with OpenAI and Google and AI. And then we've also got X or Twitter, which XAI already absorbed last year. Musk called it the most ambitious, vertically integrated innovation engine on and off Earth. So what is the rationale behind this space-based AI data centers? Musk's argument is that global electricity demand for AI simply cannot be met with terrestrial solutions without harming communities and the environment. So his solution is put data centers in space where they can harness near constant solar power. We're seeing a bold timeline as it is with Musk, claiming that within two to three years, the lowest cost way to generate AI compute will be in space.

[00:38:13:22 - 00:39:05:12]
Mallory
 Musk is talking about launching a million satellites, adding 100 gigawatts of AI compute capacity per year with an eventual goal of reaching one terawatt per year from Earth and even more from moon-based manufacturing. Some of you all might remember we covered Project Suncatcher from Google a few months ago on the pod. Essentially, this is their own research into space-based AI data centers, but they framed it differently as a moonshot goal. So they were planning to launch just two prototype satellites for early 2027 to test the concept and they estimate viability maybe in the mid 2030s. Sundar Pichai said a decade or so away, we'll be viewing it as a more normal way to build data centers. So we're seeing kind of two differing views. Of course, Google and Elon Musk, two different entities for sure. Amith, what is your take on this merger?

[00:39:06:23 - 00:40:04:04]
Amith
 Well, I think it's definitely worth paying close attention to. I think the comment that Musk made is worth unpacking a little bit. He's talking about vertical integration. And so think about what is being controlled under one roof. You have the leading company in space exploration. This is a company that has the lowest internal cost and even the lowest public cost for launching each kilogram of mass into space. So that's an incredible advantage if you think about that. And of course, Musk is relentless if nothing else in terms of driving his businesses forward, lowering costs, et cetera. And so SpaceX will continue to be a leader if not the leader in that whole category. And I think he's right that data centers that orbit the earth make a ton of sense. It's free power. Down here on earth, we're trying to figure out things like fusion and other types of ways of doing carbon neutral,

[00:40:05:04 - 00:40:41:05]
Amith
 environmentally friendly and most importantly, scalable power solutions to fuel everything else on earth, but obviously also AI. But out in the solar system there, in fact, not even that far from earth, there's effectively abundant, unlimited free power. So that's a big, big thing. Also, it's pretty cold in space. So if you can figure out the heat dissipation problem, which is in a vacuum, how do you actually cool without liquid, which needs to be replenished and without air? How do you figure out how to pipe that heat out into space, which is an interesting problem. And there's definitely solutions for that.

[00:40:42:06 - 00:43:00:10]
Amith
 But Musk is also known to be extremely aggressive with all of his timelines. Tesla's full self-driving was something promised, I think about a decade ago, if I remember correctly. And that's not poking fun. It's just the reality is, is there's some people that are out there that are just ultra aggressive and their ambition sometimes is something that is well directionally correct, but perhaps a little bit slower than what they predicted. So my suspicion is that somewhere between what Musk is saying and what Senator Pichai is saying will actually end up being a reality. Google is pretty conservative in the way they estimate things. As you've seen in the last two years, they have been able to really turn a big organization on a dime and do amazing things with AI. I have no doubt they'll do great work in this area. But I think Musk might be a little bit on the fast track in terms of what he is saying he can do. That being said, you know, Grok, which is XAI, those guys put up data centers faster than anyone said was possible because everyone else in kind of the normal orbit says, it takes two years to put up a data center. Musk put up his first data center for X.AI in six months with 100,000 GPUs in it and figured out the power and figured out the cooling. And so when you have a lot of resources, obviously he's a smart guy, obviously he's a lot of smart people around him and the will to just kind of bend timeframes, you can do some amazing things. So I'm rooting for this to happen because I think it solves a big environmental problem we have as well as a power availability problem. I think many other companies are going to be going after this exact same thing, you know, the satellite constellations business, which is the low Earth orbit, low latency, high bandwidth internet access thing, right? That's going to be the big game and you're going to see Amazon and they're going to see Google undoubtedly will be part of that. And it's going to completely destroy the current telecom industry. You know, the current companies that are there, Verizon, AT&T, they will either cease to exist or they might become little rounding errors on the balance sheet of some of these other companies because you won't need cell service through cell towers and all the infrastructure for that. If you have ultra reliable, super high bandwidth, low latency satellites that are available at all times everywhere on Earth, it's just a fundamentally superior technology. Now there will be certain backbone things that you want to run on fiber, but for most consumer grade stuff that you and I use, you know,

[00:43:01:10 - 00:43:36:12]
Amith
 if we have our phones connect to a satellite and we have instant free, not free, but like very close to free type of access, you're talking about a very different game. It's also capital intensive to put this stuff in space, but so is building cell towers and upgrading from 5G to 6G, which is in progress now and all these things. They take sometimes decades. So I'm excited about it. It's going to ultimately be great for consumers. It's going to solve a choke point in terms of AI capacity and power. And what I'm also excited about is getting all of that carbon emission off of Earth, right? You make that carbon neutral, but you put that in space. That solves a big problem that we have here.

[00:43:37:15 - 00:43:51:24]
Mallory
 This is incredibly interesting. And I look forward to seeing how this plays out and we'll definitely cover that on the pod. But Amith, for our association leader listeners, what do you think is the takeaway here watching this infrastructure story play out over the next few years?

[00:43:53:06 - 00:44:14:10]
Amith
 Well, I think there's two things. One is don't believe that constraints really exist unless there's a law of physics that demands so. So when you think about like, I can't do this or we can't do it because of that, when you hear these stories, regardless of what you think about Musk, you might love him, you might hate him. But just think about what is being accomplished here and the time scale it's being accomplished in.

[00:44:15:10 - 00:44:54:07]
Amith
 This is a situation where you see you see someone, really a team of people that are, you know, bending reality to suit their time scales, right? And so associations are often victim to the thinking that certain things just have to take longer. There's certain things they have to wait longer for. And so hopefully this inspires some folks to think, well, you know, maybe we can actually suspend disbelief and set an ultra aggressive goal to do something that we never would have thought possible in a pre-AI era. So that's one observation is maybe this can inspire people to think differently about how quickly they can adapt and adopt new ideas. As far as the broader implication,

[00:44:55:09 - 00:46:12:04]
Amith
 this is going to be yet another accelerant to an already fast moving, fast accelerating curve. When we talk about exponential curves, we're talking about the convergence of multiple exponential curves, right? You have the power of AI, you have the availability of bandwidth, and we're about to put power on an exponential curve. Granted, it's in space, but it's essentially a straight line up in terms of both figuratively and literally in terms of the power availability up there. And it's free. So if you're able to do that, we solve for an amazing number of problems at the same time. And what that all boils down to is abundant power, and what I mean by that term of power is power in your hands, right? Abundant power of AI that you will have, and it's going to transform your sector, your industry, your profession very, very quickly. If we have unlimited intelligence and we can apply it to any problem that humanity chooses to apply it to, we're going to see a different world. And we already are seeing that happen, but it's just going to accelerate further. So I think this should be exciting for everyone. I don't think that it's directly applicable unless your association is in the space industry, let's say, or in the power industry. Of course, it's directly relevant to you right now, but it'll have second order effects and third order effects on every sector and every association that's listening.

[00:46:13:18 - 00:47:07:23]
Mallory
 Yeah, that was great of me. That was a really great way to sum it up. And I want to double down too on this idea of challenging your own beliefs and assumptions. Hopefully you don't take offense to this of me because I'm not directly comparing you to Elon Musk, but I will say you have a bit of that as well of, okay, people typically say it should take this long to do this thing. Why couldn't we do it faster? Is there a law of physics that prevents us from doing so? And oftentimes what happens in those situations is a lot of us team members or me, for example, I'm thinking of building the Sidecar website on HubSpot and being told that's going to take months and months. There's no way. And we're saying, well, maybe we could do it faster. Why does it have to take three to six months? And we did it in, I don't know, maybe six weeks. And so I think there's something to be said of that. Challenge your own assumptions, challenge your own beliefs. If it typically takes you six months to roll out a new offering to your members, why couldn't it take four or three? I don't know. Just something to think about.

[00:47:07:23 - 00:48:31:07]
Amith
 The simple way to think about this is just ask the question, why? More frequently when someone tells you, I need three months or six months or three days, ask why. And don't set objectives that you think are reasonable. Set objectives that you think are needed. When you think about like, I need to have this project done by tomorrow. You think, well, that's a lot of work. I better give this team member a week to do it. I never think that way. I think about what I need. I need it tomorrow. And I'm going to say it's due tomorrow. Now, of course, if someone has a reasonable reason, and if I ask why a few times and there's actually a reason why it can't be done tomorrow, then I'm not an entirely unreasonable person, but you have to set ambitious goals. But the most important thing is to ask why and to be intelligent about the way you explore and investigate a problem so that you're really trying to boil it down to first principles, right? It's that fundamental idea of what is the choke point, what is the constraint, and is it a constraint manufactured by ideology or by culture? Or is it actually a constraint that's limiting you? And the law of physics may also be financial laws. Like if you literally need more money than you have to do this thing and there's no other way that you can find to do it, okay, maybe that's an actual first principles constraint, but oftentimes there's very few of those. And usually what I just mentioned is an artifact of our imaginations and often other ways, especially these days where everything's becoming so much cheaper, so much faster, there's usually ways to work around that.

[00:48:31:07 - 00:48:58:00]
Mallory
 Well, wrapping up this episode, KimmyK2.5 shows AI can run teams of itself, 100 agents working in parallel. Claude's new plugins means big AI is coming for specific industries. Associations, take note. And Elon Musk is merging rockets with AI to put data centers in space. Google calls that a moonshot and Musk says two years. So the through line here, as we all know, AI is not waiting around. It's

[00:48:58:00 - 00:49:03:10]
 (Music Playing)

[00:49:14:03 - 00:49:31:02]
Mallory
 Thanks for tuning into the Sidecar Sync podcast. If you want to dive deeper into anything mentioned in this episode, please check out the links in our show notes. And if you're looking for more in-depth AI education for you, your entire team, or your members, head to sidecar.ai.

[00:49:31:02 - 00:49:34:08]
 (Music Playing)

[01:12:32:12 - 01:12:35:14]
Speaker 5
 Are you ready to become the AI leader your association needs?

[01:12:37:02 - 01:13:03:04]
Speaker 5
 Sidecar's Association Artificial Intelligence Professional Certification, or AAIP, is your fast track to mastering AI skills built specifically for the association world. Whether you're just getting started or you're already experimenting with AI, getting your AAIP will give you the practical tools, real-world examples, and hands-on training to boost productivity, enhance member value, and confidently guide your organization into the future.

[01:13:03:04 - 01:13:08:05]
Speaker 6
 And right now, listeners of this podcast get an exclusive discount.

[01:13:09:09 - 01:13:21:23]
Speaker 6
 Visit sidecar.ai.ai.p and use code AIPOD50 today for $50 off your year-long pro-level subscription. That's AIPOD50.

[01:13:23:13 - 01:13:28:17]
Speaker 6
 Become the AI savvy leader your association deserves with Sidecar's AAIP certification.

[01:13:28:17 - 01:13:54:22]
Mallory
 Our mission at Sidecar is to educate one million people in the association community around the world on artificial intelligence. And the good news is you can help. If you're enjoying this podcast, please take on the challenge to share it with one friend or colleague each time you listen. This will help us spread the word and hopefully bring one million association leaders and volunteers into the age of AI.

[01:13:54:22 - 01:16:23:07]