Sidecar Blog

The Intelligence Sandwich, Claude & Gemini Updates, & Data Centers in Space | [Sidecar Sync Episode 160]

Written by Mallory Mejias | Nov 16, 2026, 7:37:28 PM

Summary:

Frontier models are now strong enough that the harder question is what to point them at. Back from digitalNow 2026, Amith Nagarajan and Mallory Mejias work through a crowded week of releases for association leaders trying to keep pace. They break down Gemini 4 Carbon and its million-token output, unpack Ethan Mollick's argument that Claude Fable 5.2 nearly clears his own bar for AGI, and explain the intelligence sandwich Amith uses to pair frontier models with cheaper ones. They also trace Anthropic's new Claude Sheets, Whiteboards, and Claude Data, challenge the switching costs and data exposure that come with them, and look at Google's Project Suncatcher running Gemma 4 on TPUs in orbit. Along the way, Amith makes the case for unifying your data before connecting it to any AI platform, and for keeping your focus on how this pace of change is reshaping your members' work.

Timestamps:

00:00 - digitalNow 2026 Recap
06:15 - Gemini 4 Carbon's Million-Token Output
11:27 - Fable 5.2 Nearly Clears the AGI Bar
15:55 - A Film Written Entirely in Code
19:34 - Fable 5.2 Reviews MemberJunction
23:51 - The Intelligence Sandwich
27:11 - Claude Sheets Takes On Excel
33:42 - Unify Data Before Claude Data
38:25 - Gemma 4 Runs in Orbit
41:27 - Modular Data Centers in Space
45:13 - Watch What Your Members Face

 

 

👥Provide comprehensive AI education for your team

https://learn.sidecar.ai/teams

 🤖 Join the AI Mastermind:

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

🎀 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

 📅 Join the pre-registration list for digitalNow 2027:

https://sidecar.ai/digitalnow 

🛠 AI Tools and Resources Mentioned in This Episode:

Fable 5.2 and Jagged AGI (Ethan Mollick's Article) ➔ https://tinyurl.com/52jpw6we

Project Suncatcher ➔ https://tinyurl.com/yc7jn892

Gemini 4 Carbon ➔ https://tinyurl.com/mte2w3t9

Google Antigravity ➔ https://antigravity.google

Claude Fable 5.2 ➔ https://tinyurl.com/yt46zb4d

Claude Sheets, Whiteboards, and Claude Data ➔ https://tinyurl.com/5h64cms7

MemberJunction ➔ 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: 

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 - 00:00:27:03]
Amith
 My name is Amith Nagarajan.

[00:00:27:03 - 00:00:29:11]
Mallory
 And my name is Mallory Mejias.

[00:00:29:11 - 00:00:40:01]
Amith
 And we are your hosts. And as always, we have an action packed episode literally today. Mallory, I don't think we're going to catch a breath for more than a second, because there's so much to go over, isn't there?

[00:00:40:01 - 00:00:56:17]
Mallory
 Oh, there's so much. I know we always say that, but typically every week, there's a pretty obvious two to three topics that we must cover this week. In particular, I was having trouble with because there's probably about 10 things we could have covered. So I tried to narrow it down to the ones that I think are most relevant for our audience today.

[00:00:56:17 - 00:01:15:15]
Amith
 Yeah. And that's actually a tough job is, you know, cutting things down and reducing the scope of something, whether that's in a business or in software or in a podcast, you know, listening to this great podcast featuring Toby Luca. I don't know if you've ever heard of him. He's the Shopify founder. He's a fascinating guy, really brilliant individual.

[00:01:16:24 - 00:01:20:09]
Amith
 I'm going to pull up the name of this podcast, still listening to it.

[00:01:21:17 - 00:01:51:23]
Amith
 I'll pull it up later, but this podcast is all about interviewing thought leaders and it's called the Knowledge Project. A friend of mine turned me on to it and he was talking about this idea of people think that they're making progress by addition, but often it's actually subtraction that results in the greatest gains in your business. I think that's more true than ever in the world of AI when generating stuff is so incredibly easy that you can produce, you know, blogs and software and all sorts of things so quickly.

[00:01:51:23 - 00:02:10:00]
Mallory
 Yep. Amith, this is our first live podcast, you and I since Digital Now 2026. So I just wanted to give you a second to talk about it. I had a great time. I know, I think you had a great time too. It seems like it was a fantastic event. Great keynote speakers, great content across the board. What did you think?

[00:02:11:03 - 00:02:27:16]
Amith
 I really did have a great time, Mallory. It was, it's fun. I mean, Digital Now is always a great time. This is the sixth year I think we've run the event since 2021 was the first time we kind of brought the event back to life after a pause during the thing that happened in 2020 that we don't talk about too much anymore, but

[00:02:28:22 - 00:02:53:24]
Amith
 something happened in the world back then that had an impact on events. And so we brought it back to life in 2021 and we've run it ever since. And the first couple of years were on disruptive technologies, including AI, but we included other things as well. And then in 23, we made it a pure AI show. We launched the AI learning hub in 2024. I think we launched the pod in 2024 as well, didn't we? Right after Digital Now.

[00:02:53:24 - 00:02:57:15]
Mallory
 Late, I always forget, was it 2024 or 2023?

[00:02:57:15 - 00:04:30:23]
Amith
 It might've been late 23, but it was right after Digital Now in Denver. Cause we talked about Digital Now, we talked about it at the Digital Now in Denver. So Sidecar and Blue Cypress more broadly has a long history with Digital Now. And actually well before we owned the company and ran the event, cause we bought Digital Now from the founders in 2020 and they ran the event for 19 years. And so in a former life, I used to tend to show us an attendee, you know, at various locations all the time. So, but anyway, this year in Washington, DC, we were there in late October. We had 330 registrants, which is a high watermark for us. And it was fantastic. We probably expect to get over 400 next year based on the growth trend that we've seen. So that was great, but more importantly, the conversations within the community is always what excites me. So I like to just bounce around to different lunch tables every day and, you know, eat a little bit of my lunch at one table and then eat a little bit of my lunch at another table. So I guess it's kind of weird to do that, but I like to just participate in a lot of conversations. Mostly I just sit there and listen cause I do all my talking on the stage. So I just like to sit down and hear what folks are talking about and practitioners are practicing. We've got people out there doing stuff with AI, you know, my specific talk, which we're not going to really be getting into today, but it's a big theme for us content-wise for some time now will be is sovereignty, how to ensure that your AI stays sovereign, which I think is a really important topic. And Mallory, your keynote was fantastic as well. And definitely to our listeners, a tiny bit about what you spoke about on stage.

[00:04:30:23 - 00:05:43:01]
Mallory
 Yes, I definitely will. And I did, I mentioned it briefly. I think it was the last episode we put out the one that was live at Digital Now. I touched on it briefly, but my keynote was kind of fusing my two worlds. One side as an actor, one side is this podcast co-host, somebody who speaks about, you know, AI and innovation for associations all the time and seeing if I could put those two worlds together and realizing they have a lot more in common than I thought, which is that as actors, we navigate a very saturated competitive market and we're auditioning against many other people for limited roles and realizing that you all as associations historically may have been the only one auditioning for relevance in your members' lives. But now the room that you're sitting in, the audition room is getting fuller as well. So the whole keynote, you know, it was a full 50 minutes of yapping, but it was basically about identifying the stories, the member stories that exist in your data, figuring out which assets you have that you could bring to that story and then taking narrow action based on those two, whether that's a product you're launching or a new service for your members, basically creating new forms of value. So that was very exciting for me. I had said it on the podcast, also very nerve wracking. I was pretty nervous, but I had a great time on the stage and I am appreciative that I had the opportunity.

[00:05:43:01 - 00:06:03:17]
Amith
 Well, you did a great job. I think you broke, but what do they say in acting? Break legs. I think you broke both legs. Oh, okay. I'm saying that you did a great job. A lot of people said great things about your talk. I think the thing is you leaned into what makes you you. And that was your advice to associations is, you know, embrace your uniqueness. Don't try to be a carbon copy of someone else. So

[00:06:04:18 - 00:06:05:03]
Amith
 I think it's.

[00:06:05:03 - 00:06:05:22]
Mallory
 Good segue.

[00:06:07:00 - 00:06:15:11]
Amith
 So it's it's important to really be distinctive. And Google is doing a good job with that right now, actually. So,

[00:06:15:11 - 00:06:19:05]
Mallory
 man, I don't know how you came up with that. Everybody, we basically have a mini topic.

[00:06:19:05 - 00:06:19:19]
Amith
 A lifetime of dad jokes.

[00:06:20:21 - 00:07:02:21]
Mallory
 A mini topic today because we just had too many topics, like I mentioned. So I want to briefly talk about Gemini for carbon carbon copy. Google announced a second Gemini for model called Carbon in late October and opened it on November 10th. It is the same introductory price as Argonne, which if you remember, we gave a full episode to recently and it's included with Google, A.I. Ultra, not the free or pro tiers. Google is calling this its strongest model aimed at cybersecurity defense and the hardest engineering jobs like migrations that span many packages at once. We don't have a ton of time to dedicate to this. Unfortunately, we might have to do a future episode on it again. But, Amith, what is your experience with Gemini for carbon?

[00:07:02:21 - 00:07:20:11]
Amith
 It's pretty pretty amazing. Actually, this probably predates you quite a bit, Mallory, and most of our listeners, but the carbon copy. Have you ever actually seen a carbon copy of something like a physical carbon copy of a form that you filled out or a credit card slip?

[00:07:20:11 - 00:07:24:23]
Mallory
 Is that when it's like multiple pages and if it reflects the signature?

[00:07:24:23 - 00:07:31:11]
Amith
 Well, that's the fancier, more modern versions is when the paper had it built in where you could like kind of, you know, write through it and it like the subsequent lines.

[00:07:31:11 - 00:07:32:16]
Mallory
 Yes, that's what I've seen. What's an original?

[00:07:32:16 - 00:08:00:05]
Amith
 There actually used to be an individual sheet in between the pieces of paper. So with credit card slips back in the day, they would actually have the top slip and there'd be like a black sheet between and there'd be a sheet below it. And you could really easily smudge it and it was pretty low tech, but it didn't work. And so you'd have like, you know, raised numbers on the plastic of the credit card and you'd put it on a machine and you'd roll this roller over it and then people would sign it. So I'm guessing you probably never experienced that.

[00:08:00:05 - 00:08:02:07]
Mallory
 We've come so far with these.

[00:08:02:07 - 00:10:24:12]
Amith
 Yeah, this is like going back to almost like the days of punch cards, which I never personally experienced that. But when I was growing up, I remember those credit card slips quite well. And I started working my first jobs in retail. I was a clerk at a Sears and I had to take credit cards that way. And then we got electronic machines shortly after I started. So a little bit out of the history books, but coming back to Gemini for carbon, and most certainly is not a carbon copy of anything else out there, because like with Argonne, it has a million token output option, you know, I shouldn't say option, but output capability. And that has really changed the game in terms of its practical use on really good tasks, because think about it this way. All of the other Frontier models are limited at 128K tokens, which is a lot of tokens. That's 128,000 words essentially, which is what is that roughly equivalent to maybe about 100 pages, I think, if I remember correctly, it's a decent chunk. Maybe it's more than that actually, but it's a lot of content. But if you're dealing with a really complex project, whatever that may be, if you stop at that number of tokens and then you have to continue with the next chunk, it's breaking up your thinking. And so with Gemini for Argonne and now Carbon, they can output up to a million tokens continuously, which is a lot. And that allows the model, even if it doesn't go to a million, even if it goes to 200,000 or 300,000, to have a more continuous stream of output. And I found that to be very helpful in the kind of work that I do, which tends to be very involved, complex, like software architecture type stuff. And it's been pretty amazing. Now, we'll be talking about Fable 5.2. I'd say they're comparable to each other, but I do think Gemini is really onto something with that. Other people, of course, are going to carbon copy that feature at some point, but model architecture-wise, they're ahead in this area. And it's just an outstanding model overall. Google continues to very much play second fiddle to anthropic open AI and maybe some might even say Grok in terms of the press they get on their models. But I think it's worth taking a look. If you haven't played with Antigravity or with Google's Gemini, I would encourage you to check it out. It's really, really strong and it's also inexpensive, particularly with our Ultra plan. You can get a whole lot of use for 100 bucks a month. So it's worth checking out if you haven't looked at it.

[00:10:24:12 - 00:10:28:20]
Mallory
 Would you say carbon is noticeably better than argon from your experience?

[00:10:28:20 - 00:10:51:21]
Amith
 I think so. I haven't tested it heavily enough. I just got access to it, I think, yesterday maybe or maybe the day before. Fair point. And with argon, we've been good friends for three weeks now, so I know argon very well. And that model is extraordinary and definitely not a throwaway model. It's an incredible model. I think carbon might be a little bit slower because it's generating more tokens, it seems, in its thinking.

[00:10:53:07 - 00:11:25:22]
Amith
 But I would say it definitely seems to be a step above in terms of its ability to reason through really complex things. Like if I say, hey, I'm going to load up a 20 or 30 million line code base and talk about some really complex and challenging software architecture ideas for what we want to do for the next few years. Carbon is a really good thought partner across something like that, whereas argon, I think, even argon could get limited in that area. And I haven't used table 5.2 for that type of thing yet, so can't compare it directly. But carbon is definitely a step up over argon. And the benchmarks do tell that story as well.

[00:11:27:03 - 00:12:39:15]
Mallory
 Well, moving to Fable 5.2, as Amit just mentioned, and the concept of jagged AGI and anthropic released Claude Fable 5.2 on Monday, November 9th, Fable is the tier above Opus, built for long open-ended technical work. And anthropic's own guidance is that most people should start with Opus 5.5 and only reach for Fable on the hardest problems. Ethan Mollick, someone we've covered on the podcast many times before, he is a leading AI voice and professor at the Wharton School, published a post called Fable 5.2 and Jagged AGI. He sets his own bar for AGI up front and he makes it a narrow one. His test is whether an AI is as good as a typical human expert at most of the thinking work a person could do on a computer. That rules out anything physical, anything that involves dealing with other people, and most real jobs where sitting at a computer is only part of what you do. His verdict is that Fable 5.2 almost clears even that narrow bar. Jagged is his word for why it almost clears it. The model performs at expert level in some areas and falls well short in others. Amith, AGI, Jagged AGI, does the terminology really matter so much as the model capability?

[00:12:39:15 - 00:13:12:19]
Amith
 I think it's the latter that matters. What matters is what it does for you. So whether or not it's extraordinary at some particular discipline that you never do doesn't necessarily matter, right? You have this extraordinary capability that's in the case of Fable and Carbon. These are still somewhat expensive, but you have this extraordinary capability that's incredibly cheap compared to the human equivalent. So you can now untap it for things that you really have to actually stretch your creativity to figure out what to do with these models.

[00:13:13:20 - 00:13:27:16]
Amith
 I was talking to some folks here at Blue Cypress yesterday and just saying like, well, what do we do with this stuff? You know, we're just kind of kicking around saying like, what do you want to do with this extra intelligence? You know, it's like basically you've doubled your intelligence from Fable 5 or whatever.

[00:13:28:19 - 00:15:54:20]
Amith
 It is interesting. There's all sorts of outlets for it. There's definitely a smoothing of the jagged edges that makes it the best expert at everything. That's obviously going to happen in the next couple of years probably at some point, definitely. The question is what matters to you as an association leader. So I'd break it down into two general buckets. First is what you do internally, your internal processes. What are you doing day to day, week to week, month to month? I would argue that some of the more complex thinking that you're doing, like thinking through your business strategy, a model like Fable 5.2 would be where you should go because its ability to understand subtleties, nuance, understand you and think at a higher level is very useful when you're dealing with something that's more abstract like that. If however, what you're doing is saying, hey, I've got this spreadsheet and I need to take these numbers and move them from over here to over here and I want to do these calculations or I've got all these documents and I want to summarize them and I want to, you know, do whatever. Like the typical work streams of most associations, even at like a director or VP level day to day, you're fine with Opus 5.5. You're probably fine with Opus 4.6 from earlier this year. The model intelligence has been so strong. So I think it's more of the latter part of what you said. It's the model capability relative to your needs. The second bucket is what your members need. And so this is where I think in some cases frontier level intelligence will in fact be very helpful. If you're solving a really complex problem in your field, right, you're in a scientific or engineering discipline or a field of law where there's, you know, tremendous complexity, tons of content, lots of nuance. The power of these models may perform at a level that all of a sudden unlocks new use cases, new ways for you to combine your unique content, your unique brand with frontier level intelligence to do things you never thought you could do. You know, being the world's greatest expert in your discipline available on tap every hour of the day as part of a member benefit or a paid product. There's ways to do that at scale now that are just different qualitatively than before. We could have done that three years ago and we've been doing stuff exactly like I just mentioned for several years. But the quality now is a different conversation. So it's hard to actually really describe the answer to that question in a more succinct way. At least I'm not great at succinct anyway. But like I think it's hard to describe it succinctly because there's so much to unpack there and there's so many different ways to use these things.

[00:15:54:20 - 00:16:59:04]
Mallory
 Yep. So Moloch always puts these models to the test and his post opens with kind of a little short film with a twist ending. The model made all of it and it also asked Moloch not to give the ending away before people had watched the whole thing. So we will not be spoiling the ending of that video either, but we'll link it in the show notes. Moloch did not ask for a video. What he gave was a request for the single most impressive thing he could share with people at launch, something he could embed in a post that would hit people hard and grab them fast. And he told the model to be clever about it. That was the entire brief. He gave it no feedback once it started. Choosing to make a film at all was the model's decision. And so was the little twist at the end. But here's the thing. If you're an avid Claude user, you probably know this. Claude doesn't generate video. It outputs text and code is text. So what it actually did was basically write a program that draws the film. That is a fundamentally different thing from Sora or Vio, which generate video by predicting pixels. So, Amit, you saw the video, right? What did you think about the video? And can you help us understand why that's so crazy?

[00:16:59:04 - 00:17:57:17]
Amith
 It is really nuts because to think about that and say, hey, essentially what you're doing is like what would be the code to do to animate a frame and the next frame and the next frame and then to look at that over the course of a storyboard of, in this case, I think it was 45 seconds or something. And then to create music that goes with it and to tie it all together in this beautiful way. So there's obviously a lot of planning that went into it. But the code that you're describing is literally just writing code that says render this frame like down to, you know, how am I going to draw these objects? This is the way this tree is decomposed out of these shapes to draw these shapes and put them together. So it's pretty low level. I don't think a human programmer could ever do that or would ever do that. You know, the higher level tools that are used for these things are at a different kind of level of abstraction. But I was totally blown away by it. And by the way, the point that you made that I thought was really interesting is the model asked Malik not to share.

[00:17:58:17 - 00:18:07:18]
Amith
 And he didn't tell it like what to do or not to. It just said, hey, you should do this, which I think was a smart approach to like getting it's a hook, right? It gets people to want to watch the video.

[00:18:07:18 - 00:19:11:13]
Mallory
 Yeah, very clever, very creative. Everyone really go check out the video. It's I think it's actually over a minute, Amith, but it's very impressive and catchy. And at the end, you'll have a hard time kind of thinking, oh, did an AI really create that? Because it's pretty engaging. He ran a few more experiments with the model. One of them turned a tiny USB keyboard into a game console. It built a business from scratch. Thought that was pretty interesting. I rebuilt a 1941 literature course as an interactive website and it wrote an original economic study that ran to one hundred and eighty two pages on the model itself. The headline from artificial analysis, which benchmarks models independently, ranks Fable five point two first ahead of Gemini four carbon and Opus five point five tokens are about twenty percent cheaper than the last version. And output is up to 30 percent faster. So, Amith, what has been your experience thus far with Fable five point two? And especially since you're such good friends with Argonne, I'm curious, do you have any comparisons with your benchmarks?

[00:19:11:13 - 00:21:45:20]
Amith
 Yeah, so Fable five point two is definitely next level. And it I haven't tried a video generation project like Moloch did. By the way, just as a quick side note, if you're not a subscriber to Ethan's newsletter, one useful thing dot com, I think I would definitely recommend checking it out or following him on LinkedIn. Super smart guy. What I actually really like about him is he's not technical. So he's a business professor and he's been involved in tons of technology over a long period of time before I. But it's a super sharp guy with like a very blended background, you know, in the classics and all these other areas that he's led him to where he's at in life. And what is interesting about him in my mind is how he is creative in the way he asks the AI to solve problems right in the way he's framing the problem and thinking through it. So he's definitely one of the thought-blowing. I follow I like the way he kind of unpacks these things. And because he's not from the world of AI or computer science or software engineering, I think he's got a really different take than a lot of the people that you might listen to. Coming back to my experience. So I fired it up right away. I gave it a prompt. I said, hey, here are 20 different code repositories in GitHub, member junction, as well as a whole bunch of other apps that sit on top of it. I want you to unpack this, study it deeply and come back to me with a critique. Tell me what is bad, what is good and give me a plan to improve it to be beyond anything that exists in the world or something along those lines. And it thought for about, I want to say 30 to 45 minutes and came back with some really novel insights, some stuff that Carbon didn't tell me because I run the same thing through every frontier model when it comes out. And actually, it's been amazing because a lot of times I think we do a pretty good job with software. We've been doing a long time, but we're a bunch of humans. And so we're sitting around looking at the same patterns over and over. And we miss things. And so it's unbelievable what it finds because there's a couple of different things that I specifically ask too, is we want the experience for our customers to be more delightful, which is obviously totally qualitative. We want this software to be faster. We want it to cost less to run, which means less AI inference and also less general compute memory that also makes it more green, you know, in terms of running the software. And we want it to be more hardened in terms of cybersecurity. And so those are kind of the general ideas. But then we just said anything else that you can think of that would make it a better piece of software and comes back with all these ideas for making it, you know, both better, faster and cheaper to run. So it's pretty tremendous. And I've been doing this for a while with different models. And generally speaking in the past, their

[00:21:47:02 - 00:22:25:04]
Amith
 thoughts have been fairly limited and fairly underwhelming because when you take into account this massive code base and the complexity of it and understand the context of thousands of associations using this type of product, it's not an easy problem to solve. It takes a lot of this nuance and context. So Fable has access to my context in Clod. So Clod has a lot of knowledge about me and about Blue Cypress and about Member Junction. And so it's able to use that along with Clod's just intellectual horsepower. And that's my version of Molex tests. It's obviously a lot nerdier kind of software version, but it's a very complex task. I was extremely impressed with it.

[00:22:25:04 - 00:22:28:06]
Mallory
 Do you think you'll implement any of the changes that were recommended?

[00:22:28:06 - 00:22:29:13]
Amith
 They're already underway.

[00:22:29:13 - 00:22:31:09]
Mallory
 Oh, they're already underway. Yeah.

[00:22:31:09 - 00:25:42:04]
Amith
 And actually we had one of the things we like to do is we'll take Fable and Argon, or now Fable and Carbon and have them kind of duel each other in a way where, like for example, we do a lot of what's called plan based development, which is a fancy sounding term. It simply means write a plan before you start writing software. It's kind of like this, Mallory. If you want to build a new house, you might want to have some blueprints, right? Just don't start digging a big hole and hope you had a nice house. That's shockingly, the software development community doesn't always do this. Sometimes people just start writing code. And sometimes that's actually helpful because you're kind of exploring the problem space and it's like creative exercise. That's great. But what you need is a really good plan when you're going to go build something industrial strength. And so the planning phase is iterative. You know, plan gets created, it gets critiqued, gets revised, et cetera, just like writing a blog post or writing a book. And so we use high end models to help us with planning, right? So we'll talk to a Fable 5.2 and we have done this with Fable 5.2, talk to carbon, talk about a problem space, how we're trying to solve an issue, build a plan, and then we'll go back and forth between the models like ping pong and have them improve the plan. And ultimately the plan dramatically improves. And then we set it off to be implemented. Now when you implement the plan, once you have an extraordinary blueprint, you don't need the world's greatest architect on site at the construction site every day to make sure that the blueprint is implemented to their exact specifications. There's ways to test that and validate that even if the architect isn't there on site. They're going to come, they should come and check it fairly regularly, but they don't need to be there every day. There's construction supervisors. There are obviously people that are tradespeople in various areas, but then there's people who check the work. There's also building inspectors. So if you think about it in the physical world, which I think is a little bit easier to relate to for most of our friends listening compared to the kind of strange world of software where it's all abstract, it's the same concept supply. And so we use these agents, these AIs in agent systems and we like to use Fable 5.2 now and Carbon for the planning and for the review at the backend. So it's kind of like sandwiching the work with really, really high intelligence models and putting much simpler stuff in the middle. It's still hard. It's still complex, but we can use way less than models. For example, using Gemini 3.8 Flash or using Claude Haiku 5.5, which are good models, but they're way, way less capable than the Frontier. So Mallory, my bottom line is I think a model like Fable 5.2 is something that people should test out, but ultimately I would recommend doing it. It's something you think is so complex that models of the past were unsuccessful or you never even attempted to do it. By all means throw it at Fable 5.2, throw it at Cardin. But another way to approach it is to say, I'm going to start a project with Fable 5.2, then I'm going to have it ask Opus or Sonnet or Haiku to do the bulk of the work and then have it check the work. So it's kind of like this intelligence sandwich, right? Where you have the highest intelligence on the outside and you have a somewhat lesser intelligence in the middle doing kind of the actual grunt work, so to speak.

[00:25:42:04 - 00:27:16:16]
Mallory
 Intelligence sandwich. We're going to keep using that, Amith. Moving on to or continuing the Anthropic theme on Tuesday, November 10th, Anthropic announced Claude can now build spreadsheets, whiteboards, and dashboards. Completing a set it's been assembling for months. Here's a full lineup since we haven't walked through most of these in a minute. Claude has Docs, which is a real document editor. Claude drafts you and your team edit the same file alongside it. They have Slides, which is a deck builder. You can describe your presentation. It builds and formats the slides. Also exports to PowerPoint or Google Slides. They have Claude Design, a visual canvas with real design tools for mockups, posters, landing pages, Claude Motion, short animated explainers. They've just launched Sheets, which is basically spreadsheets with tabs, formulas, and formatting exportable to Excel or CSV. Whiteboards is also new. It's a shared whiteboard where Claude groups, the sticky notes, and drafts action items out of those. And then Claude Data, which was formerly dashboards, and this is now generally available. Finally, Anthropic says Meetings is coming soon, which would allow Claude to join live video calls as a visible participant, take notes, answer chat questions, and draft plans in real time. Docs, Slides, and Design are free. Sheets and whiteboards need a paid plan. There's no new price tag on any of these. It just draws on your own existing usage limits. Amith, do you see this as Anthropic coming for Microsoft or just a natural progression of the work they're already doing?

[00:27:16:16 - 00:28:12:20]
Amith
 Well, Mallory, I think it is in a way, Anthropic coming from Microsoft, but indirectly. So the essence of how I describe it is this. If you were to compare Claude Sheets, what it is today, what it'll probably end up being over time, it's probably not going to take the full feature set of Excel and replicate it. But if they implement, let's say 10 percent or 20 percent of Excel, would that be useful is the question. And in enterprise software and also for desktop productivity software like Word and Excel, for a long, long time, there's been a pretty well studied notion of the 80-20 rule, which in this case is that 20 percent of your features are the ones that are used by roughly 80 percent of your users. So or put another way, 80 percent of your users only use 20 percent of your software, which is kind of crazy if you think about it. And actually I found in my experience, it's closer to like 90 to 10.

[00:28:14:15 - 00:29:19:18]
Amith
 And so I mean, just think about Excel and all the craziness that's in there and how much have you ever actually used even people who are reasonably good at Excel? Now, I would say if you're the CFO of a company and you have super sophisticated financial modeling and an Excel workbook and link sheets and all these things, will you be bringing that to Claude sheets anytime soon? Probably not. It probably doesn't make a lot of sense. But if Claude as an AI working for you can spin up a sheet or spin up a doc or spin up a slide and it's editable by you, shareable by you, and it becomes a work surface for collaboration, there's incredible value there because they've lowered friction. They've made it so that the thing is possible right in the work stream of talking to the AI. So it's very helpful even if it is only 10 or 20% of the functionality. That is the logic, I think. And in addition to that, I think there's an incredibly beneficial business model for Anthropic because if you think about it, just with Anthropic's artifacts, have you ever shared an artifact with someone? Not exported it, but actually shared a link back to Claude, Mallory?

[00:29:19:18 - 00:29:21:20]
Mallory
 Like with a team member, you mean?

[00:29:21:20 - 00:29:35:17]
Amith
 Yeah, with a team member or even with someone outside of the team. Just sent them a link, say, "Hey, here's something I built. I'm going to send it out." We do it all the time, right? Because it's convenient. And what that does is essentially it drives more traffic back to Claude.

[00:29:36:19 - 00:32:15:23]
Amith
 It also means that people can now start to comment on documents, on sheets, on slides, on other things inside the AI system, which also adds more mass to the value, right, if you think about it that way in that system. And then as you add more and more of these tools that were once external and are now brought in, it does give the AI system more understanding of what you're doing, which is helpful for the AI to be smarter on your behalf, but it also dramatically increases your switching costs because it's not easy to take those documents, those comments, those links, all these things that you're building up, all the context, all the memory with you. So it's pretty clear to all the labs, they're all a bunch of smart folks. That the model itself is a treadmill, meaning it's not a place where they're really going to make money long-term. What they have to do is create an ecosystem that's so valuable that people are choosing to be trapped. You're choosing to be a captive of their environment because it's so good. And that is the choice people are making right now in choosing TrackGPT, Claude, Gemini, these other systems, which by the way, ties back to the conversation I was having with our community at Digital Now about sovereignty, because there are alternatives to this where you can build your own context layer, your own infrastructure, and I think that's worth at least considering before you make the choice to be on someone else's land, which is where you always be if you choose to build all of this infrastructure in a cloud. So on the one hand, I'm super excited to see these products evolve. I guarantee you OpenAI is going to be no slouch and they're going to have the same features in TrackGPT. They'll probably leapfrog them in some areas because OpenAI has voice models and Anthropic doesn't have their own voice models. OpenAI does have image models and they're bringing back video generation, by the way. They killed Sora that they're going to be doing video generation again. And Google has voice and image and video models. Anthropic doesn't have those. So when these other companies say, hey, we're truly multimodal, yes, Claude can generate videos through code, but we can actually do these other photo realistic videos and all these other things, let's bring it together. The idea is that the suite of services collectively is so powerful to draw you in. And I think that's a reasonable business thesis for them. I don't know that it's winner take all, but I think there's going to be multiple camps, just like Google Workspace and Microsoft Office is something of a duopoly. So my general take is, yes, they're coming from Microsoft, but not because they're coming from Microsoft. They're coming for both WalletShare and Mindshare. They're coming to make it easier for you to do your work, easier for the agent to work for you, and then ultimately that accrues to benefit them as well.

[00:32:17:02 - 00:32:57:04]
Mallory
 I think that's the right read, Amith. Microsoft stock dropped around 3% on the news, but Anthropic was pretty modest about it. It's product lead called Sheets Fairly Basic, and the documentation tells you to export to Excel for real functionality. Claude Data is something I wanted to discuss with you. It connects to a data warehouse or CRM and builds live dashboards off the actual system, certified queries let someone lock in the organization's official metric, so every dashboard calculates active member in the same way. For example, Live Sharing lets you hand someone a dashboard that runs on your access where they can't see the underlying data and can only run approved queries.

[00:32:58:06 - 00:33:09:20]
Mallory
 Do you recommend people listening to this podcast use Claude Data if they want to do a quick kind of analysis or attach this to their CRM or their AMS? Is that something that you would recommend doing?

[00:33:09:20 - 00:36:35:04]
Amith
 I will speak to it in the general sense as opposed to speaking to Anthropic. First of all, what they built is great. It's very powerful, very functional. It can connect to many different kinds of systems. There's been versions of this that Anthropic has made over the course of the last roughly year once the AI got smart enough to be able to do this. I know this category really well because this is exactly what we do with Skip, which is our AI analytics agent, and it does a lot of the same kinds of things. The thing that you have to think through is where are you building residual value, which is when you go create a chart or a graph or a dashboard, is it again in a system that you have some degree of control over and you have portability, you have an eject button where you can take your stuff with you, or are you essentially stuck in that environment? That is very much the case here with data. I made those exact same points a moment ago with regards to sheets and docs and so forth. Here I'm making the point in the context of your data analytics, and I think it's an even more pointed observation here because it has to do both with that issue of portability and sovereignty, but also the issue of data safety and security. When you open up a pipe to your AMS or open up a pipe to your CRM or whatever it is, you are authorizing those companies to suck in all of that data to do these kinds of things. And so there is governance around it to an extent, but you are essentially creating a deeper level of trust to do that with this organization. You might be fine with that. That's OK if that's the case. Just be aware of what you're doing and specifically be aware of the fact that your switching costs are radically increasing. So that's the downside to it. The other downside to this architecture is Claude doesn't have any form of storage. So when Claude data connects to your AMS and your LMS and your other systems, Claude has to actually dynamically connect them together, which is a lot harder than if you bring your data together physically in one place. So what I would argue for is if you're going to do this thing with Claude data or the chat GPT equivalent, which is also really exciting, I would recommend that you unify your data first so that you're connecting Claude not to the AMS and LMS and other systems individually, but connect Claude to a system that has all of your data unified. Something like Member Junction obviously does this. This is the AI data platform that we have, which is totally free and it's open. And so what you could do is unify all your data to MJ and then connect MJ to Claude and then Claude data will actually be way more performance because Claude data will be able to understand what's inside that database on a unified basis rather than having to figure out all these things on a separate basis. It also reduces your exposure a little bit because you have kind of an intermediary in there. And that's a totally free thing that you can do with with MJ. There's other ways to do the same thing as well. But it is outstanding that this is out there. It's really exciting. I don't want to like, you know, take away from the enthusiasm people might have for this because the idea is to go explore your data as quickly as your mind thinks of the question. You can get the answer, which is what people have wanted forever. So I find this extraordinarily exciting. It's just something that has some pretty significant downstream implications. So it's worth stopping for a second and thinking it through. By all means, check it out. But then stop, pause and think about what this means for your overall data governance and your business strategy and your risk profile.

[00:36:35:04 - 00:36:42:18]
Mallory
 And so when you're talking about the member junction unifying your data first and then using something like Claude data, that's just via API.

[00:36:43:18 - 00:36:58:09]
Amith
 Yeah, I suspect it's probably going through MCP, which is a form of an API. But yes, and you're connecting whatever it is, you're choosing to connect to Claude. Claude's smart enough to look at that and say, OK, now I have a general idea of this data. And then can do all sorts of really cool things.

[00:36:58:09 - 00:38:33:01]
Mallory
 I want to move to our last topic for today, Project Suncatcher, which we actually covered on this podcast. I don't know. I think it was over six months at this point. Google is calling Project Suncatcher a moonshot. A moonshot means taking an enormous problem and seriously testing a solution that sounds absurd, knowing most of them will probably fail. The idea is to put TPUs, Google's in-house AI chips, on solar powered satellites in an orbit that rides the line between day and night. So the panels stay lit and there's no atmosphere in the way. Clusters of them would be linked by laser. So they work together as one data center. What's actually up there right now is a single prototype about the size of a refrigerator carrying four TPUs on roughly a kilowatt of power. It launched October 1st. The AI data center is being built on the ground, draw many orders of magnitude more than just as a note. On October 30th, the team sent up a prompt to Gemma 4, which is Google's open weight model running on the chips on board. Open weight, meaning Google publishes the model for anyone to download and run themselves, which is why it can live on a satellite instead of calling home. The answer came back as the satellite passed over Norway as a haiku. Blue marble below, data streams in silent grace, starlight feeds my core. Beautiful. Heat and radiation tests come next, then twin linked satellites in 2027. Amit, do you see this as a small step or a big step in the journey for AI compute demands?

[00:38:33:01 - 00:40:00:01]
Amith
 I think it's a giant leap. I'm pumped about it. And you know, SpaceX is doing the same thing and I know Amazon has a plan for something similar. So I'm pumped about it because the idea is that you can harness a free resource in space. You don't have the water problems, the environmental impact, the local community impact, all the issues that are causing blowback on the data center construction at scale right now. You eliminate all of that and you also put things in a way where networking becomes a lot easier because these laser links you're talking about between the satellites is interesting. But also we've already seen that proven at scale with Starlink, which is a different type of thing because it's for connectivity. But putting the compute in space makes total sense. Musk is talking about that being in the next like three to five years. He tends to be a bit on the optimistic side in terms of his time scales. He may be right this time because the compounding of what's happening with AI in space as well, which he obviously knows a lot about both topics, is actually happening so fast. It may make these things more possible than most people would estimate. But whether it's five years or 10 years, it's going to happen. And it's going to be a big thing because it's going to dramatically reduce the power consumption needs here on Earth. And be a much cleaner way of doing it. So I find it incredibly exciting. And I think there are obviously a whole ton of challenges to make this work at scale. But you know, we've got a lot of smart people thinking about this problem. So it's exciting.

[00:40:00:01 - 00:40:53:01]
Mallory
 As you said, I wanted to add in a note. SpaceX has asked the FCC for authority to run up to a million orbital data center satellites linked to Starlink. Blue Origin has filed for more than 50,000. A startup called StarCloud has already trained a small model in orbit and China has been launching its own computing constellation since last year. But as you said, there are some issues. Nothing can be repaired once it's up there. Heat is much harder to shed in a vacuum than people assume. And the entire economic argument depends on launch costs falling by something like a factor of 18. So definitely still progress that needs to be made. What's interesting to me, Amith, is I thought a moonshot. I want to say Google came up with this moonshot just last year and maybe the year before that. And I always think of moonshots as like really long term goals. But it is pretty insane to think they've already made this much progress on it.

[00:40:53:01 - 00:41:15:10]
Amith
 I suspect people have been working on this for quite a while. I mean, I think it's been in the popular conversation for the last six to 12 months. But you know, if you looked at kind of the arc of AI and power consumption and growth, a lot of people have been thinking about this for a bit. So I don't know when Google's Suncatcher project officially started. But I know that like SpaceX, I believe, has been talking about it for several years. So

[00:41:16:11 - 00:43:34:06]
Amith
 I think for my my take on it is when you think about sustainable, renewable energy and you think about solar, it's it's very seductive to think we can power everything through solar. The problem is here down on Earth, we don't get light all the time everywhere. And we also have challenges with things called the atmosphere and clouds, which block some of that radiation coming down to warm us up and feed our solar panels. And so when you put the satellites up in orbit, obviously you can eliminate those issues and have nearly constant power. I think there would be many other things that might work that way. And the point you made actually is really good about this idea that it's essentially it's a one way trip. And the trip also is one that essentially has to be like a sealed container if anything goes wrong. It's a throwaway. You send it back out of the atmosphere to burn up. That's true right now. But if you also think about the trajectory of what's happening with physical AI, aka robotics, and if we were to say, hey, what if you ship a robot up along not like a humanoid robot, probably, but like a small robot in each of these small little self contained data centers that is capable of doing the most common repairs? That might be interesting. Or what if you make it modular and you allow the data center, these little mini data centers around to have construction such that they can eject modules and the modules burn up or the modules somehow return safely to Earth and are recycled. But they're taken care of and the new modules arrive, right? Because they're able to like these little mini things like dock and land and then they get inserted into their modular spots. I guarantee people I'm making this up as I talk about it. But it's like it's not a really hard concept to think through. So this idea of reducing cost reuse, making things better, faster, cheaper, because the infrastructure of the satellite, like the physical housing and the solar panels and like the communications and the lasers and all this stuff, most likely that stuff will far outlive the lifetime of a GPU or a TPU or an LPU. And you want to replace those every couple of years probably, but maybe maybe the satellite can have a lifetime service of 10, 15, 20 years. I don't know. I'm speculating. But therefore there's a big economic incentive to find ways to make these things more modular. So I think we're going to see a lot of innovation there. Good news is we have lots of near free intelligence helping us think through these problems and all these companies.

[00:43:34:06 - 00:43:39:16]
Mallory
 I know. I'm actually thinking maybe for digital now next year we should try to get someone from Project Suncatcher.

[00:43:39:16 - 00:43:47:00]
Amith
 That'd be pretty fun. Yeah, we'll definitely work on that. That or a similar project from another organization thinking about this stuff would be great.

[00:43:47:00 - 00:43:57:13]
Mallory
 What do you think is the takeaway for association leaders who are listening to all of this maybe technical sounding stuff about satellites and orbital compute? What do they need to keep an eye on?

[00:43:57:13 - 00:46:10:11]
Amith
 I probably provide a very similar line on this each time we talk about what's happening in the world of AI because AI is itself. It's interesting to the extent you find it interesting. What's more interesting to everyone is the speed of change that's happening in the world. And the idea that change is happening as quickly as it is means that your members are impacted. Forget about yourself for a second as the association leader thinking about your processes and your issues internally. But think about what your members are facing in their worlds. In some cases they might be in sectors where you think you know what some of this stuff isn't going to change too much for us anytime soon. And maybe that's true or maybe it's not because maybe there's alternatives or there's new ways of doing things or maybe it's affecting your members in a personal way rather than a professional way. Right. But those are still things that are relevant to you as an association. So that's the first hat that I'd put on is the member centric hat saying hey how is this changing our members lives and why that's super important aside from being empathetic to your customer is understanding what they may need. So thinking about where are my members going. What are they facing. What pressures what challenges and what opportunities exist in my field because of these changes and what should the association be doing to not only show leadership but to create services and products that help the member fully benefit from the speed of change. That's the first thing that we should all be doing when we think about this is it's kind of a little bit selfless in a sense saying take care of your customer figure out what's going on in their worlds from these changes. It's obviously what we're trying to do with our association clients. We think a lot about our internal business too obviously in the world of A.I. And you know for example like will the services we provide even be relevant in five years. Obviously we don't know for sure. We think that associations that are going through transformative times will want a helping hand from experts in that field at any point in time even 20 30 years from now. And that's what we think of our business as not about A.I. even though we're experts in A.I. Obviously so I think associations have to really think deeply about how this affects their members. That's my biggest comment and the internal stuff will follow.

[00:46:10:11 - 00:46:13:15]
Mallory
 Well everybody Google and anthropic both shipped

[00:46:13:15 - 00:46:18:24]
 (Music Playing)

[00:46:29:17 - 00:46:46:16]
Amith
 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.

[00:46:46:16 - 00:46:49:22]
 (Music Playing)