The 4 Modes of Working with AI, The Transformation Paradox, & Building a Learning Organization | [Sidecar Sync Episode 137]

Summary:

In this jam-packed “mini” episode, Amith Nagarajan and Mallory Mejias break down a whirlwind of recent AI model releases—from Anthropic, Alibaba, Microsoft, and beyond—and what they signal about the rapidly evolving AI landscape. Then, they dive into Microsoft’s 2026 Work Trend Index Report, unpacking the “agency equation” and what it really means for organizations navigating AI adoption. From the rise of agents and the four modes of working with AI to the growing gap between employee readiness and organizational culture, this episode explores why AI transformation is less about tools and more about leadership, systems, and mindset. Plus, they introduce the concept of “owned intelligence” and what it takes to become a true learning organization in the age of AI.

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:

Microsoft Work Trend Report ➔ https://shorturl.at/yMqCH

Claude ➔ https://www.anthropic.com

Qwen ➔ https://qwenlm.github.io

MiniMax ➔ https://www.minimaxi.com

NVIDIA Cosmos ➔ https://www.nvidia.com

Microsoft Copilot ➔ https://www.microsoft.com/copilot

GitHub Copilot ➔ https://github.com/features/copilot

👍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 - 00:00:25:02]
Amith
My name is Amith Nagarajan.

[00:00:25:02 - 00:00:26:23]
Mallory
 And my name is Mallory Mejias.

[00:00:26:23 - 00:00:31:20]
Amith
 And we are your hosts. And guess what? We're going to talk about AI today.

[00:00:31:20 - 00:00:41:00]
Mallory
 Is that a surprise to all of you? I don't know if you were expecting it, but we've got a lot to cover to today. Very exciting, but it's going to be a meaty episode. I'm warning you.

[00:00:41:00 - 00:00:43:20]
Amith
 Just a bit going on in the world of AI. How are you doing, Mallory?

[00:00:43:20 - 00:00:57:11]
Mallory
 I'm doing well, Amith. This is a very busy week for me on all fronts, just lots of stuff going on. But I'm very grateful to be busy. I know you just attended MMCT, right? So how did that go?

[00:00:58:11 - 00:00:59:14]
Amith
 MMCT was wonderful.

[00:01:00:23 - 00:01:38:00]
Amith
 There was, I think, ASAE did a wonderful job at the event. I had not been to an MMCT event before, which everyone I told that was shocked because I've been around a long, long time in this space. And I've never been to that particular conference. And so it was great to get to go. Different audience, different group of people than the annual conference and some other stuff ASAE does. So it was great. I think they had about 1,000 people there. From a commercial perspective, our various companies all said they did well in terms of meeting with clients and prospects. Sessions I heard were good. A number of our folks were speaking, talking about, of course, guess what, AI. So those were all very well attended sessions.

[00:01:39:02 - 00:01:47:03]
Amith
 So it was good. I had a good experience there. I came back with lots of ideas and a cold. So I got a little bit of a bonus on the way back.

[00:01:47:03 - 00:01:55:17]
Mallory
 A bonus cold. Any ideas that you want to share with us on the podcast or any topics that you're focusing on after your conversations with association leaders?

[00:01:55:17 - 00:02:13:10]
Amith
 So MMCT, for those that aren't familiar, stands for, I believe it's Meetings, Marketing. Oh, sorry, not membership. Membership. Yeah, membership, marketing, communications, and technology. That's the problem with acronyms is that you get different ideas in your brain. And mine is like an LLM. It doesn't necessarily have perfect recall.

[00:02:13:10 - 00:02:14:07]
Mallory
 Hallucinates at times.

[00:02:14:07 - 00:04:25:11]
Amith
 Yeah. Well, I think the meetings people are important too. So I think they were there in force as well. But it was basically business unit leaders that are kind of member facing, if you will, or customer facing. And so it's a different audience than what comes to annual. Annual is mostly C-suite folks. There were some C-suite folks at this event. These are people who are in director and manager level positions, some above that, who are interested in how do you improve the member experience, how to improve communications, how do you do a better job of marketing. And of course, there's a technology element to it. But these are not technology folks for the most part. And they're not executive folks for the most part. So these are people who are actually out there doing the work to try to drive transformation, which is so cool. Because normally I'm speaking most of the time with CEOs and C-suite folks who are reaching out to me. And I love talking to people at all levels of an organization. So in conversations that I was part of or I observed, I was hearing people experimenting with a genteck AI at a level that I've not heard before. So people that are trying cloud co-work, people that are trying codecs from open AI, people that are trying the Gemini agent platform. They're trying to go beyond prompting and waiting for a response. They're trying to basically actually have AI systems rather than just AI prompts. And that's super exciting. You and I have talked about that a bunch Mallory, and we're starting to see the basic bits and pieces of that come together. So that was cool. A different perspective than the CEO would have when you're talking to the membership manager and membership director about what they're facing, what they're challenged with. So coming back, I think there's ideas for the way we can improve our content at Sidecar to be even more directed at folks at that level and in those roles. So lots of interesting things there. We already have a ton of content in courseware and the Sidecar Learning Hub focused on these topics around deep dives in marketing, deep dives in membership. But I think we're going to do even more there. And then for a product perspective, I always have ideas talking to people. When I hear people express their pain points, that is the ingredients. It's the fuel for innovation. And then we take that back into our crazy lab and think about ways we can solve problems.

[00:04:25:11 - 00:04:44:08]
Mallory
 Yeah, that's really exciting. And I think that is, I'm excited to talk about today's topic, which is the Microsoft Work Index Report because there is of course a lot about agents in there that we'll get into. But before we get there, Amith, we kind of have a mini topic today before our main topic, which is something we don't typically do.

[00:04:44:08 - 00:04:45:22]
Amith
 Not so many, many topics too, right?

[00:04:45:22 - 00:05:32:12]
Mallory
 Well, there have just been a flurry of model releases within the past couple of weeks. And we wanted to talk about the Microsoft report, but we also wanted to make sure we were providing you all with the most up-to-date information on those models. So I'm going to quickly cover a few. Amith's got some thoughts and then we'll launch into today's main topic. First, we saw Claude Opus 4.8 released, which is Anthropics new flagship with gains in coding, agentic work and reasoning. The standout improvement on that one is Anthropics is reporting Opus 4.8 is roughly four times less likely than 4.7 to leave flaws in its own code unremarked. The release is also introducing dynamic workflows, which orchestrate parallel sub-agents for tasks like code-based scale migrations.

[00:05:33:12 - 00:05:58:07]
Mallory
 We also saw Quinn 3.7 max from Alibaba, which is a closed-weight flagship built for agentic long horizon work with a 1 million token context window. The headline claim from Alibaba is a 35-hour autonomous coding run with more than 1,100 tool calls in a single session. It's also the highest ranked Chinese model ever on the artificial analysis intelligence index.

[00:05:59:09 - 00:06:26:03]
Mallory
 We saw the release of MiniMax M3, a Chinese open-weight model with a new sparse attention architecture that delivers a 1 million token context window at roughly 1 20th, the per-token compute of its predecessor. The pitch is that the first open-weight model to combine, it is the first open-weight model to combine frontier-level coding, a 1 million token context window and native multimodal input, so text, image, video, and a single model.

[00:06:27:05 - 00:06:49:04]
Mallory
 We also saw NVIDIA Cosmos 3, which is not a chat model. It is an open-source foundation model for physical AI, combining vision, world generation, and action prediction in one system aimed at robots, autonomous vehicles, and smart spaces, a signal that physical AI is becoming its own category, perhaps separate from the LLM race.

[00:06:50:04 - 00:06:57:19]
Mallory
 I mean, that was the overview, but I think you've got maybe a couple more to throw in there. What is your thought on the past few weeks of model explosions?

[00:06:57:19 - 00:10:21:10]
Amith
 Well, that's a lot, as always. And not to be outdone, Microsoft released a series of seven models just two days ago, or maybe it was yesterday, I can't remember honestly. So in the last couple of days, they released seven new models. And this is the first time Microsoft has released a close to frontier-level model of their own. And that's quite an interesting thing to unpack for just a moment. Up until now, Microsoft has been legally precluded from doing frontier-level AI work. Our open AI agreement originally basically called for open AI to give them a whole bunch of things. But one of the things Microsoft gave back in return, in addition to a bunch of money, was the commitment to not develop certain sizes of models and above. So in the past, we've covered on this pod the Microsoft Phi models, which is spelled P-H-I. And these are, at the time, they're really, really good small models. I hope they continue with that work. That most recently, I believe, is this version four of that. Really good small models. But they never really ventured into building bigger models. And that was one of the main reasons. Also, they probably didn't have the in-house chops until they hired Mustafa Saliman, who's one of the DeepMind founders from way back and a very talented guy. And he's brought a lot of people with him. And so their charge has been to build world-class AI at Microsoft. So they released a series of seven models. Their biggest model is roughly on par with Sonnet 4.6. So not as good as Opus level, but quite useful. And then they have a flash coding model, which apparently is as good as Haiku 4.5 on all the benchmarks. Or actually, I think their press release talks about it being considerably better than a Haiku 4.5, but notably a step below Sonnet. But that's actually quite useful for a lot of day-to-day coding tasks. So you mentioned Opus 4.8 is better at realizing that it made errors in its code, which is a really important thing for frontier AI. But actually writing the code doesn't necessarily take an Opus 4.8 level model. A lot of the code writing is pretty basic. That's true for a lot of domains where the bulk of the work is pretty ordinary. And then you need a little bit of extraordinary on top of that in order to make sure the ordinary work is done right. And that's actually the strategy they're taking at Microsoft. They've had dependence upon external parties for one of their most important products, which is the GitHub Copilot product. It's the coding assistant that started all this. When people first started getting really excited about AI and coding, it was autocomplete. It was GitHub Copilot. It rewinds in the days of chat GPT, actually preceding chat GPT by just a touch. GitHub Copilot was an autocomplete tool. As you were writing code in the development tool, it would just suggest the next line, the next paragraph, even sometimes an entire class or function for you. And that was remarkable at the time. And they've lost their edge. They lost their advantage because agentic coding tools like Cloud Code, the modern versions of Cursor, Codex, etc., do a lot more. You just talk to them and they produce the entire thing, which is extraordinary. And up until recently, GitHub Copilot didn't have an equivalent to that, but they do now. They have something called the GitHub Copilot CLI for command line interface, and it is their equivalent to Cloud Code. And if they have their own high quality coding models, I think they'll be a key competitor there. And it has something nobody else has in this space, which is enterprise distribution.

[00:10:22:11 - 00:12:07:19]
Amith
 Many organizations use Copilot in their office environment, not because they think it's the best AI, but it's the one that comes with Microsoft. It doesn't really come for free. You pay, I think, $30 or $40 a user. It is not necessarily the best, but it's the one that's integrated with the Microsoft suite with 365. It's under the same IT security and governance regime as the rest of their Azure stuff, so they're just comfortable with it. And so that distribution advantage is enormous. Google has this with Google Suite as well, right? And so now that they have models that they're advancing on their own timeline with their own agenda, they're going to build models that are specifically tailored to the workloads that Microsoft cares about, which of course they can't get open AI or anyone else to do just their bidding. So I think this is two things in one. Number one, it means that there's another heavy duty, 800 pound gorilla type technology player producing models. That's good. In addition to that, it means that Microsoft is probably going to be able to advance their own work with Copilot across the office suite and Copilot for coding much more rapidly than before. I think one of their constraints was the right model from the right vendor at the right cost in order to power not only the capability of the product, but the right gross margin for their business. Right. Microsoft is not interested in losing money. They want Copilot to be very, very profitable. And I suspect one of the reasons Copilot has lagged is not because Microsoft doesn't realize models are a lot better than Copilot capabilities, but because they have to constrain the consumption in order to make the price per user flat fee thing work. So I'm hopeful that Copilot and those that are kind of stuck with it, for better or worse, will see significant advancements in the next six months.

[00:12:09:06 - 00:12:33:02]
Mallory
 I was talking to an association leader just recently who, through their association, uses Copilot because that's what's approved and sanctioned and then had a personal computer next to them where they were using Claude for other tasks. And so it'll be interesting to see if Copilot can kind of close that gap a bit. Amit, of the models that I mentioned, aside from Microsoft, which do you think is most important to pay attention to if you had to pick one?

[00:12:33:02 - 00:12:50:23]
Amith
 I think collectively the non-Claud ones are the ones to just note here. I think Opus 4.8 is great. It's a nice upgrade. I mean, in normal times, in normal products, it would definitely be a major version release, but I think they're waiting for version five until they get a Mythos class model out there, which apparently is imminent, by the way.

[00:12:52:08 - 00:13:03:11]
Amith
 So we'll see what happens with that. But Opus 5, if that's what they call it, or maybe they actually get rid of Opus and they call it Mythos going forward. But the Opus 4.8 is a notable release. It's great.

[00:13:04:21 - 00:13:08:07]
Amith
 In terms of the Quinn 3.7 series, the Minimax M3,

[00:13:09:08 - 00:15:02:22]
Amith
 Chinese labs are moving forward. And the other thing that wasn't mentioned earlier, but it's notable, is that they claim that most of their training, if not all of their training, has been done on domestically sourced chip infrastructure, not on NVIDIA hardware. And if that's true, it is quite remarkable because they're producing products that are basically just as good as GPT-55, as Opus 4.7, maybe even as good as Opus 4.8 in some ways with hands tied behind their back is an expression far too weak to describe the difference in hardware capability. So that's notable. And I think ultimately that's a really good thing for the AI community. Obviously, for AI alarmists, particularly those that are thinking of this as a zero-sum game between China and the United States, it's a cause for concern that the frontier here is really not much ahead, if at all, of Chinese labs. And in some places, they're far ahead in terms of value per token, meaning the business value created arguably is higher for some of the Chinese models. As a practical matter, for those of you in the association community in the United States or in another English-speaking country, and you are saying to yourselves, "That's cool, but I don't want to run models on servers in China," just remember that, but for Quinn 3.7, you generally see these labs producing open source models that you can run in a wide array of different data centers, usually sourced local to your countries. In the United States, there's a bunch of providers. I know that's also true in many other countries that host these models. And then just to say a quick thing on Cosmos 3, this is worth checking out, like best explained through video. It's really quite stunning what they're doing here. And this feeds back to the thematic thing that Mallory and I have talked about for a while about world models, physical understanding of the world, and how it all ties together. It's quite interesting. So it's exciting times. It's also hard to cover this as a mini topic.

[00:15:02:22 - 00:15:08:00]
Mallory
 We're going to have to do a full blown episode on all these, maybe in the next few weeks. Also, everybody...

[00:15:08:00 - 00:15:13:05]
Amith
 But by then, there'll be a bunch of new models. So you know, Quinn 3.7 is going to be old news, right?

[00:15:13:05 - 00:15:32:04]
Mallory
 So moving goalposts. Everybody, if you're interested in learning more about world models, we have a good episode out there. I don't know which number it is, but it's with Thomas Altman and we get really deep into world models, if that's of interest to you. So moving to topic one, the new agency equation. Agent scale. Humans get more room to lead.

[00:15:33:04 - 00:16:18:02]
Mallory
 This year's edition, as I mentioned, is based on a survey of 20,000 AI users across 10 countries plus trillions of anonymized Microsoft 365 productivity signals. The headline framing for this 2026 report, as agents take on more execution, humans gain more agency, so more room to direct the work, make the calls and own the outcomes. Let's look at the agent data specifically first. So according to Microsoft, the number of active agents in the Microsoft 365 ecosystem grew 15x year over year, rising to 18x in large enterprises. Agents are now used in every industry. The pattern of adoption varies. So software and tech adopts broadly, nearly one in five firms using agents are in that category.

[00:16:19:09 - 00:16:25:08]
Mallory
 Manufacturing has a smaller share of firms using agents, but deploys them at much greater scale within each organization.

[00:16:26:13 - 00:17:04:14]
Mallory
 Notably, nonprofit and travel hospitality sit at the low end of both breadth and depth, a useful reference point for all of us in the association world. What does this look like, though, at the individual level? So when the work trends index survey, two thirds of AI users say AI has freed them up to spend more time on high value work and well over half say they're producing work they couldn't have produced over a year ago. Among frontier professionals, which is what we're going to call the most advanced AI users, about one in six of those surveyed, that last number doing more work than they couldn't have a year ago is actually at 80 percent.

[00:17:05:15 - 00:17:46:02]
Mallory
 Asked which human skills are more important as AI takes on more work, AI users put two at the top by a clear margin, one of those being quality control of AI output and the next one being critical thinking. And 86 percent say they treat AI output as a starting point, not a final answer, that they stay responsible for the thinking portion. And then frontier professionals, which I think is interesting, take this further. They're significantly more likely to say that they intentionally do some work without AI to keep their skills sharp. And they pause before starting to decide what should be done by AI versus a human. The report frames this as a refusal to outsource thinking.

[00:17:47:03 - 00:19:17:20]
Mallory
 Finally, on this topic, I want to talk about four modes of working with AI. So the report introduces a framework with two axes. So one is how the human engages with AI directing versus supervising. The other is how the agent is used, assistant versus teammate. And then we've got four modes that come out of that. So the first mode is asking. This is quick exchanges. You're looking up a fact, you're rewriting a sentence, you're formatting a table. Then we've got delegation. The human sets the direction the agent executes. This could be turning raw notes into a structured recap or pulling a recurring report from standard inputs. The next mode is collaboration. The work needs both refining a proposal through multiple rounds, drafting a communication where tone requires judgment. The human and the AI are working together. And then finally, exploration. This is testing what AI can do. So you're trying a new workflow and you're probing the edges of an agent's autonomy. What sets frontier professionals apart is not which mode they use, it's knowing which mode a task calls for. So I mean, this year's framing on the report is the agency equation, which is a big shift from episode 31. We covered 2024, which was more about AI adoption and anxiety of adopting AI. Do you think associations and you kind of touched on this at the beginning of the episode. Do you think associations are still struggling with that adoption and anxiety from the report back in 2024? Or do you think they're kind of shifting into that agency equation?

[00:19:18:22 - 00:20:06:03]
Amith
 I think the majority of associations are still very much struggling with adoption and having significant anxiety over data privacy and security, over who should do what, over which tools to use. Most associations have dabbled but haven't really jumped in that far where they're getting into the agency equation the way you're describing it, Mallory. I think that quick note that is important to consider for your association is study this report will include it in the show notes. I'll link to it. If your industry is moving faster than you can honestly say you were moving, that's a problem. We always say on the pod that if the rate of external change is greater than the rate of internal change, over time you have a problem. That's how you go down the obsolescence path regardless of what you do.

[00:20:07:07 - 00:21:08:15]
Amith
 And more specifically, the micro lens in your space is if the rate of pace of AI adoption in your industry is considerably higher than the rate of AI adoption in your association that represents that industry, that's probably an even more acute and immediate challenge. So something to consider and maybe just reflect on. Some of those industries are kind of natural ones that you're going to expect to see further along the technology adoption curve. But some I think are moving in interesting ways that even if the overall industry isn't moving fast, like manufacturing maybe, there are some that are going really deep. And so those members in those sectors are going to expect more from their associations in that regard. But I think coming back to your question, most associations I think are still very much in that early adoption or they have anxiety about adoption in different ways. Or they have taken the approach they've taken with other technologies where they've picked a tool, checked the box and said, "Hey, we've done basic staff training. We picked co-work. We picked GoPilot. We picked ChatGPT.

[00:21:09:18 - 00:21:15:22]
Amith
 And we're in AI shop now. Done. We implemented it." That's where a lot of people are and they're like, "Of course we adopted AI."

[00:21:17:02 - 00:21:41:19]
Mallory
 I hear that. I hear that. But I also, I can sympathize with the association leader, maybe an association within the industry of manufacturing who is moving slower but deeper with AI agents and the association saying, "Well, we perhaps our members, the organizations that are members of this association understand AI better than we do. So how can we be expected to lead this change? Maybe we should just take the backseat." What do you say to that?

[00:21:43:24 - 00:22:33:16]
Amith
 I mean, I think that that's fine in a sense. But the question is, is the industry the association or is the association the industry? Like who is on your board? Who are your volunteers? It should be those people, right? So the people that are driving change in the space should at least have meaningful representation. And sometimes what you'll find is that the people that are on the board and the people that are in key volunteer leadership positions aren't those people. They're the people that represent, you know, they're at the culmination of their career trajectory with respect to that sector or profession. And so they represent, you know, the prior way of thinking about it. So I think there's a governance and volunteer management challenge where you want to bring those people in. Because if it's happening in the industry, that's great news because you just bring those people in and say, "Hey, we want your help spreading the good word about how AI has helped your organization."

[00:22:35:02 - 00:22:42:02]
Amith
 I think a lot of people will collaborate in ways that you might not expect actually, even in competitive industries. That's what associations are really good at bringing out in people.

[00:22:44:01 - 00:23:03:11]
Mallory
 The four modes framework, asking delegation, collaboration, and exploration, even though Microsoft clarifies it's not necessarily being good at just one of these. If you're a frontier professional, it's knowing when to use which. What do you think, Amith, is the biggest unlock? Or I'm also curious, how do you think you spend most of your time with AI out of those four modes?

[00:23:03:11 - 00:23:06:07]
Amith
 Yeah, I'm definitely across all of them. And I think,

[00:23:07:12 - 00:24:28:21]
Amith
 by the way, I thought it was interesting that even the most advanced frontier people are 86% with respect to the answer to the question of, "Are you able to do work you weren't able to do a year ago?" That's, to me, a little bit mind boggling because in my brain, I'm like, "Well, I'm doing stuff I couldn't do three months ago, 100%." However, that's also part of the gap problem that we have in perception versus reality. There's a long tail on the diffusion of a technology like this, and that's what we're working with associations coming back to the unlock and the opportunity. I think it's just taking one step. So most people are just asking. That's still where most folks are. And that's good because they're starting to get accustomed to talking to a computer, which is really an unnatural act for most of us that have been around a while. For people that are just now coming out of college, it's probably all they've ever experienced because chat GPT was there when they started college or pretty close to it. So I think for most people, just that first step was the hardest step. It's like taking the first step when you want to do couch to 5K or whatever, right? You got to take the first step. So asking is important, and it still forms the foundation of what I do. I think I spend more time with quick questions and get great answers. And the AI like Claude, in my case, is my primary daily driver for this, knows a lot of the context. I can ask it like, I can spin and even like things that would be horrible, prompt engineering a year ago or two years ago and get great results because it knows a lot about me.

[00:24:30:01 - 00:26:19:08]
Amith
 So asking is still very important. I would say that if you're not in the mode yet of delegating, that's something to think about because there's power there. If you can just delegate a simple task to an agentic tool, a very simple agentic tool like Claude Cowork, which by the way, is basically Claude code with a different interface on top of it, but it's just as powerful as Claude code. But I call it a simple tool that doesn't mean it's simplistic. It means that it's simple to use. You go to Claude Cowork, the difference in that Cowork tab is it's able to take actions on your behalf. And so this is where delegation comes in, right? So if I connect it to my Gmail or I connect it to Zapier, I connect it to a number of other tools, you can have it start doing things. So you upload a spreadsheet of a few names of people that you're interested in researching, you research those names and you say, "Hey, this is a really good plan. And can you put that information back into HubSpot for me?" Well, of course it can. Are you doing it yourself manually right now if you use the AI to do some of the work? And then the actual action is something you do manually. That's where most people are. There's a simple step you can take there. Now, I will say, be thoughtful about this, because when you connect your AI systems to tools that connect to your databases and your SaaS systems, it's very powerful, but it's something you just have to at least contemplate because you're opening up those systems to the AI tools. So you have to make sure that you're working with a partner that you trust and all those obvious things, I guess. But to me, that's the big step. Go from asking to just delegating a simple task, because it's addictive. Once you get the AI to actually do a job for you, I mean, I'm just extremely lazy with things I don't find interesting. Things that I find interesting I'll spend all day and all night on. But if I find it boring, I don't want to deal with it. I don't want to make phone calls or send emails about getting the plumber to come

[00:26:20:11 - 00:27:04:00]
Amith
 do a backflow test for my water meter in New Orleans, which is a big thing here in New Orleans. Don't want to do that. So can I automate that? I have figured out how to automate it because the people in New Orleans don't really respond to email or text messages, and occasionally they'll respond to phone calls when you call plumbers. But I digress, and I need a voice tool. I'm waiting for these agentic AIs to have outbound voice where they'll make phone calls for me to do stuff that people who don't want to email and people who don't want to use websites will support. Anyway, that's my answer to your question is go from asking to testing out this delegation, which might be experimentation, but I still will call it under the delegation headline because it's a big opportunity to reduce your workload. And who doesn't want to do that?

[00:27:04:00 - 00:27:12:01]
Mallory
 Yep. I don't think I would describe you as lazy and meath, but maybe when it comes to water backflow tests, maybe then you can say you're lazy.

[00:27:12:01 - 00:27:28:07]
Amith
 Yeah, when it comes to things that are administrative or repetitive or just something I've done it before and therefore I find boring, I am extremely lazy. I don't like doing anything again that I've done before, unless it's an interesting task, which is, you know, I'm kind of annoying that way. I don't like doing certain things more than once.

[00:27:28:07 - 00:28:02:03]
Mallory
 I think I've really mastered the asking and I think I work a lot in collaboration, back and forth. I can go back and forth with Claude about an idea or even something I write. I could do that all day. I enjoy it. I think I could work on delegation a little bit more. It is addicting, as you said, but there's also there's a fear element of just not wanting to cede too much control or to mess up something that I would obviously be responsible for. And then exploration, I don't think I'm dabbling as much with. I don't know if you have any advice for that, Amith, but...

[00:28:03:11 - 00:29:19:22]
Amith
 Yeah, I think the exploration part is opening your mind to what's possible and retesting assumptions. So if you think an AI can't do certain things, go try it again a couple months later. Try a different model. Even try the models that maybe are banned from work use for trivial, like no confidential information type things. Like go check out DeepSeek on their website and quit on their website and ask a simple question, right? Don't upload your membership database, but do some basic things. So just keep opening up your mind to these new ideas. To me, that's a big part of it. You have to do that firsthand. Listening to us talk about it or watching us on YouTube with the Sidecar Sync, it might inform you a little bit, but we're not going to actually teach you a skill. You have to go do the thing. And so exploration is an important kind of forward deployed aspect of where you need to put your brain, at least part of the time. And then it can come back to like, "Oh wait, I just realized that Cowork can actually draft an email in Gmail. That's kind of cool. I don't want it to send it for me maybe, but I sure love it for it to draft it and actually put it in Gmail. And then all I have to do is go click send. And then maybe eventually I'm comfortable enough to have it send the email for me." But there's things you can do that are kind of baby steps towards that. But I think that the way you get there is that exploration experimentation piece.

[00:29:19:22 - 00:29:57:15]
Mallory
 Yep. I love the idea of the running list of things that maybe you've tried AI for in the past and it hasn't done a great job and then retrying those every couple of months to see if there's been a gain. And then also to your point, Ameth, you have to get out there and do it. I could read all the acting books in the world, but until I'm actually performing, acting in class or on a set, it's just a totally different ballgame. But I want to move to topic two. We've still got a lot to get through. Topic two, we're talking about the transformation paradox. So workers are ready for AI. Organizations aren't. And this is a topic we've talked about on the podcast several times before.

[00:29:58:23 - 00:30:44:21]
Mallory
 Microsoft maps survey respondents on two dimensions, individual capability with AI and the organization's readiness to absorb it. The result is five zones and most workers are not where their leaders think that they are. So here are the five zones. First is frontier. This is about one in five workers. This is where individual capability and organizational readiness are both high. This would be the sweet spot. The second zone is blocked agency. This is about one in 10 workers. So you've got highly skilled workers ready for artificial intelligence stuck in organizations that can't catch up or haven't caught up just yet. Then we've got unclaimed capacity. This is the smallest slice. Organizations are ready, but employees haven't caught up yet.

[00:30:45:21 - 00:31:36:02]
Mallory
 The next zone is stalled. This is about one in six. This is where individual readiness and organizational readiness are both low. And then the last zone is emergent. This is fully half of all workers, the messy middle. This is where both the individual practice and organizational conditions are still taking shape. The kicker, according to Microsoft, organizational factors. Like culture, manager support, talent practices, account for roughly twice the AI impact of individual mindset and behavior. It's pretty mind blowing. The single strongest factor is the organization's AI culture, about two and a half times stronger a signal than the top individual factor. The translation here is hiring AI savvy people doesn't solve the problem in itself. The conditions around them do.

[00:31:37:02 - 00:32:19:00]
Mallory
 There's also tension at the top. Only about one in four AI users say their leadership is clearly and consistently aligned on AI and leaders themselves see a very different reality than their employees do. They're considerably more likely to say AI driven reinvention feels safe at their organization and twice as likely to say it's rewarded regardless of the outcome. The pressure point here is that most AI users, about 65% fear falling behind if they don't adapt quickly with AI. But almost half say it's actually feel safer to focus on their current goals than to redesign their work. And only 13% say they're actually rewarded for reinvention when results don't immediately follow.

[00:32:20:00 - 00:32:34:17]
Mallory
 I think this is really an interesting part of the report, Amith. Microsoft is saying that the bottleneck isn't talent, it's the systems around the talent. It is your AI culture. How do you see that playing out in associations? Do you agree with that point?

[00:32:35:17 - 00:32:37:21]
Amith
 Totally agreed. And it's exactly what I see every day.

[00:32:38:21 - 00:34:16:08]
Amith
 You know, I'll tell you a little story, Mallory. So when I was in DC last week for MMCT, one of the things I enjoyed most is I was able to attend a dinner that Michelle from ASAE and I co-hosted with eight other CEOs of associations. And we had this format of a salon, which essentially is kind of introspection and interrogation of a topic amongst a group of collaborators and peers just to kind of discover and discuss a topic. And of course, our topic was AI related. But it was AI transformation from a leadership lens was really what we were focused on. And so we had the different people around the room share input on a variety of topics. And it really reinforced what you just said that these are leaders who obviously are interested in the topic, they're trying to lead the charge, they are pushing ahead and showcasing through their own behavior that AI is important. AI is really critical to understand and they're pushing their organization, in some cases very aggressively, in some cases with a little bit softer touch, just depending on the personality of the leader. And I was extremely encouraged by that event, that was super cool. And I said a couple things, I was primarily facilitating that event, which was very difficult for me not to talk a lot. But it was good fun. And I learned way more than I normally would, because I do have two ears and one mouth. So I have to apply that ratio more often. So it's a good opportunity for me to exercise that. But I learned a lot from the folks that we were with, tons of ideas for things that people are doing in practice. Hopefully we'll be featuring a couple of the individuals that joined us that evening on the pod in the next month or so. But in any event,

[00:34:17:11 - 00:36:40:12]
Amith
 the leadership piece is the piece. If you get leadership right, I won't say everything else is automatic, because there's a lot of hard work to be done. But this has always been true. This has been true in an era, in eras well before our time, and certainly was true in the last couple of eras of compute transformation. Pre-internet, it was true. I was around for that. And it was difficult to get people to adopt technology if leadership wasn't behind it, if leadership didn't model the behavior of actually using the tools and illustrating how to use the tools, and then also demand that people become proficient in these technologies. That was true, obviously, as the internet took shape and became an amazing wave. And it's true now. So two things I would say really to punctuate that. Number one is that there is no other leader than the CEO who can choose to make an organization AI ready. You cannot delegate this. You cannot bypass the responsibility. And the only way you yourself as the CEO of the association can be ready is you have to become proficient in AI yourself. This is not something you can delegate and say, "I've got an AI person. I've got a CTO." Or, "I've got a really savvy membership VP who has taken it upon themselves to be AI savvy." No. You have to know what these capabilities are. Not knowing AI deeply yourself as a user. I'm not talking about a coder. I'm talking about a user. Not knowing it deeply yourself is like not knowing what a telephone can do, or not knowing what electricity can do, or not knowing that airplanes exist. There's a fundamentally transformative shift in strategy and ability. And you have to know that for yourself. And then what you have to do is propagate that throughout your organization. But the key step is you yourself adopting this attitude and demonstrating it. One of the favorite things about our partnership at Sidecar with ASAE is that Michelle, the CEO there, took it upon herself to be the first employee at ASAE to be AAIP certified. Actually, I think she had a couple that came in individually just on their own in the preceding six months. But she was the first as part of our partnership to go get certified. She spent time on her holiday break in December, and Michelle is a proud AAIP now. And it was wonderful hosting this event with her because we were able to bring together a group of other people who were very AI forward.

[00:36:41:16 - 00:37:23:10]
Amith
 And the certification, of course, we think it's awesome, but it's just one way you can showcase expertise. But then she followed up on that and said, "Hey, I'd really love for everyone to get certified in the next 30 days. Kudos if you do that. And oh, by the way, you need to get certified. You must get certified by the end of the quarter." So she gave people a little bit of time, not a ton of time, not like typical association timescales were like, "Hey, take 2026 to do it." She said, "No, you have to get it done in 90 days. And by the way, if you want to show me that you're serious about this and basically be one of my AI forward thinkers in the organization, get it done this month." And that was the beginning of the year. So that was very powerful. And she achieved that 100% of ASAE staff are AAIP certified now.

[00:37:23:10 - 00:37:25:04]
Mallory
 I didn't know that to me. That's amazing.

[00:37:25:04 - 00:37:48:02]
Amith
 Oh, really? Okay. It's super cool. We should talk about it more. But she's done a great job leading that organization forward. And Michelle herself will be the first person to say she's not at all the deepest expert in AI. But by showcasing her own commitment to learning, her own commitment to continually driving forward the organization, she's really sent the right message to her team. There's a lot of change happening at ASAE, just like at most organizations, and she's really trying to push the envelope this year.

[00:37:49:04 - 00:37:58:22]
Amith
 Leaders can do that themselves. You have to focus on your own commitment and then you have to follow through on that in mandating that your team learn. And if the learning happens, I think magic happens after that.

[00:38:00:04 - 00:38:23:06]
Mallory
 So what if someone's listening to this episode and they are the CEO and they have done their due diligence, they're listening to this podcast, they're taking AI courses, and they feel like their staff is doing the same. I feel like there's still a gap between individual readiness, even if all the individual parts are ready versus your association is ready to take the next step. How do you speak to leaders that feel like they're in that boat?

[00:38:24:09 - 00:38:53:00]
Amith
 Yeah, that's a great question because we definitely have a lot of people like that in our community. I know one CEO who immediately comes to mind runs about a 200 staff association. He himself is very AI forward, has a bunch of other people in the organization scattered about who are as well, but there are some folks in leadership that are not as AI forward. And so that's a challenge because these people do add value in other ways, but how do you bring them along? How do you coach people that may be your senior most staff or maybe they're a level below that?

[00:38:54:04 - 00:39:10:18]
Amith
 I think it's carrot and stick. I think you have to first get with people and sit with them and say, "Look, if you have some detractors out there that are slowing AI progress, you probably know who they are." You have to have both the leadership courage and capacity to sit with them individually and say, "Listen,

[00:39:11:22 - 00:41:04:04]
Amith
 tell me why you're concerned. Hear them out, of course. I always think that's important." And then say, "Listen, I hear where you're coming from, but even though those concerns may be valid, this is what we're doing. And you're on this bus. And to be on this bus, I need you to get on the bus fully. You can't be looking backwards. You have to be looking forward and be part of this team." And to be willing to make changes in your staff. If you have people who are unwilling after, obviously you have to approach this with empathy, with kindness, but also with firmness. And so the absence of willingness to be firm and to make changes when necessary is a critical ingredient for a leader to be successful. So what I see in organizations that have the problem you're talking about is usually there's key people. They might be senior staff. They might be elsewhere, but it's influential folks who are not on board and they're stopping progress. That's one issue that requires the CEO or the executive director to be bold and to move more assertively than you might be comfortable. But this is a time where if you want a direct change, you need to be empathetic and kind, but you need to be firm. The other part of it is making sure that you have the tools available for people who want to be in the delegate and experiment modes of work. A lot of people have co-pilot and they're like, "Hey, this is our approved tool or maybe it's chat GPT." And people say, "Hey, I hear about this cool stuff you can do with Claude or Gemini or whatever." And they're like, "Well, we can't use that." Well, think about it this way. Maybe consider a smaller group of people that are in this experiment group that are willing or that are interested in doing this. Give them access to that tool. Make it possible for them to do that. In many organizations I run into, I do talk to people who are like, "Yeah, we all want to go do this, but we approved this thing." Maybe we made it a bylaw. How that it's co-pilot. I hope it's not a bylaw, but could be.

[00:41:05:16 - 00:41:18:15]
Amith
 And bylaws take like an act of Congress or an act of God to change basically. So I'm kind of kidding about that. But once you get a policy in place in an association, man, that's part of it in a cockroach to kill. It's tough.

[00:41:19:20 - 00:41:23:20]
Amith
 So I'd say you got to be willing to change that policy if you need to.

[00:41:23:20 - 00:41:28:21]
Mallory
 Yeah. Oh, I hate cockroaches. No more mentions of cockroaches on the sidecar sink pod.

[00:41:28:21 - 00:41:31:21]
Amith
 Well, we have a lot of them run around in New Orleans this time of year.

[00:41:31:21 - 00:41:46:17]
Mallory
 I think I must have a phobia or something. I'm very scared of them. I also want to touch on this because we haven't really talked about this as much, but the gap between how leaders perceive their AI culture and then how employees experience it.

[00:41:47:21 - 00:41:59:18]
Mallory
 Do you have any recommendations for how you can stay truly in touch with the way your organization's functioning and not just grasping onto this? Well, this is how I hope we're functioning. So this is what I'm going to stick with.

[00:41:59:18 - 00:43:16:03]
Amith
 I think one of the earliest career lessons I learned when I was an intern at Hewlett-Packard Laboratories in Palo Alto, California, across the street, by the way, from where a lot of things were invented at Xerox Park and within my building that I was in, the inkjet was invented along with optical character recognition. So it was a really cool place to be a kid and get an opportunity to play around. One of the things I learned when I was there is part of the HP way, which is a book they have, they encourage everyone to read there. At least this goes back to the early 1990s, right? So I don't know if they do that anymore. It was that Bill and Dave, the founders of Hewlett-Packard, did this thing called management by walking around or it was MBWA, or MBWA. I can't do acronyms today. I got to have a CT wrong and I can't even spell. But management by walking around. What this basically is, is that it's essentially designed to flatten the organization. A lot of associations have kind of steep hierarchies where there's a CEO that has six direct reports, each of those people might have five direct reports, each of those people have, and on and on. And so all of a sudden, the CEO, you know, it's the emperor with no clothes, right? They're being told what they want to hear. And all of a sudden, they don't realize that many of the people that are even a couple levels removed from them are totally, totally disengaged and do not want to do what they're being asked to do or they can't.

[00:43:17:03 - 00:44:52:17]
Amith
 So what you do is you go talk to everyone, right? Your association is not Amazon.com or Microsoft. You don't have hundreds of thousands or tens of thousands of employees. You can talk to a lot of people, even if you have 200 staff, you can connect with a lot of people. This is not adding to your meeting rhythm. This is not saying, "Hey, schedule a meeting with every employee. That would be insane. It wouldn't work and it would immediately die and it would be torture for everyone." This is randomly connecting with people who you haven't talked to in a while. So back in the days of cubicles, Hewlett-Packard, you know, Bill and Dave would literally walk around and talk to people just, "Hey, what are you up to? What are you working on?" Just curiosity. And those kinds of conversations flatten the organization. They make leaders more approachable. They actually increase the level of buying and investment people have way down lower in the organization and deep in the culture's intent because you've spread your values through contact. You spread your values by having conversations with people. So if you're in one of these very hierarchical organizations, put it on your task list for this week that each day you're going to spend five minutes. It doesn't need to be a long conversation. Just randomly talking to someone in the organization who you haven't connected with in a while. And if you haven't done this in forever or ever, you will probably freak people out a little bit. They're going to go, "Why is someone high up the food chain bugging me? Are they mad at me?" Just do it. And just say, "Hey, I'm just trying to learn more about what's going on. We're in a period of change. I want to understand what you do better. I appreciate your work. And just tell me more about what you're working on." People love talking about themselves. Your employees will tell you all you want to know about their work and what they like and don't like about it if you open up to them.

[00:44:52:17 - 00:45:11:03]
Mallory
 Yeah. I love that advice, Amitha. I think that would go a long way in figuring out what exactly your culture is and also just strengthening your culture, flattening your organization in that way. Also, picturing an intern, Amitha, is definitely funny. I'm sure you had lots of ideas buzzing up there. Maybe not quite about associations just yet, but soon.

[00:45:12:08 - 00:45:15:02]
Amith
 Yeah. It basically looked the same. I just had a little bit of hair.

[00:45:16:21 - 00:46:26:00]
Mallory
 Okay. Moving to topic three of today's episode, becoming a learning system. This is what the leading firms are doing differently according to Microsoft. The firms pulling ahead are focused on AI absorption rather than just adoption. They redesign how work gets done. They turn outputs into insight and build that insight back into how the organization operates. This is what Microsoft is calling a learning system. Per the Work Trends Index Survey, frontier professionals work in a noticeably different environment. Their managers are dramatically more likely to openly use AI, set quality standards for AI work, create space for experimentation, and encourage ambitious work redesign. Gaps of 20 to 30 percentage points across every category. Their teams are also roughly twice as likely to brainstorm processes AI could change and share learnings out in the open. A separate Microsoft people study of 1800 workers backs this up. When managers actively modeled AI use, employees reported meaningful lifts across the board, including a striking 30-point lift in trust in agentic AI.

[00:46:27:16 - 00:47:02:20]
Mallory
 Microsoft is also introducing a concept worth pulling out called owned intelligence. As agents execute more work, they generate signals. What worked, what failed, where outcomes drifted. In most organizations, these signals stay local or they spread very slowly. But frontier firms are able to capture these signals, encode them into shared routines, and turn them into institutional know-how that compounds over time, is unique to the firm, and very hard to replicate. This is the difference between an organization that has AI tools and one that is genuinely learning faster than its competitors.

[00:47:03:22 - 00:48:11:09]
Mallory
 According to Microsoft, this takes four coordinated roles. The first is employees re-architecting their work around intent and review. The next is leaders redesigning processes around outcomes. And then IT, building the infrastructure for agent operations at scale and security IT weaving the trust into the system itself. The IT and security pieces are the most novel. Microsoft is actually arguing IT should treat agents as managed entities with identities, permissions, policy enforcement, and lifecycle management and become the control plane for agent operations. Don't worry, we're going to get into this. Security has to account for a genuinely new risk surface. So data exfiltration, unintended system actions, unauthorized access, monitoring and auditability built into the platform not bolted on after. Microsoft states there's three questions every leader needs to be able to answer. Who reviews agent performance? Who has the authority to update the workflows that agents run? And how does a local win get captured and scaled across the organization?

[00:48:12:16 - 00:48:36:10]
Mallory
 So I mentioned the IT and security moment because I know that on a recent episode, we talked about Ethan Mollick referring to IT as a graveyard for AI innovation. And it seemed like you were maybe in agreement with Ethan Mollick or at least could see his points. I'm wondering, do you think Microsoft's taking a different perspective on it? Can they both coexist?

[00:48:36:10 - 00:48:55:22]
Amith
 I think it's actually quite similar in some ways in that Microsoft is saying that you need to treat agents as first class entities and provision them with certain capabilities and deny them certain capabilities so that you have governance and control and auditability. And I think that's absolutely IT's function. At the same time, going back to the Mollick reference,

[00:48:56:23 - 00:50:00:05]
Amith
 it does impose a factor of a reduction in speed. Because when you're doing those things with proper governance, with proper audit trails and all this, it's slower than just doing whatever you want. That's important. IT's job is to protect the organization digitally and to make sure the infrastructure is sound as well as being secure. And so IT's role is really, really important in the AI era. The key is that they shouldn't own the business process. Business owners should drive the process forward. IT needs to provide this robust control plane. And so, of course, Microsoft's comment, I agree with it. It's the way we do things in our Member Junction agent framework. Everything has roles. It's all controlled in that way. There's an audit trail. You have to do that regardless of the tool that you're using. I think it's really important. But I do think that IT's job is to certainly encourage innovation so long as there is a sanity check to it so that you do have controls over your data and do have controls over which systems are allowed. It's really important. You can't just let stuff happen however it needs to happen.

[00:50:00:05 - 00:50:02:23]
Mallory
 Right. So they can coexist, it sounds like.

[00:50:02:23 - 00:50:21:10]
Amith
 Totally. They need to coexist. They both need to be strong. IT needs to be a critical partner in the AI equation. They're just not the owner of it. And in the world of AMS's and other last generation or many generation ago technologies, IT typically was the owner of the system. And actually, in my prior life as an AMS company,

[00:50:22:18 - 00:53:28:08]
Amith
 I always saw that that was actually the weakest AMS implementation was when IT fully owned the business case around an AMS. It was always best when there was a business owner. And IT was certainly a key partner in that process but didn't own the system itself. And that's a partnership that worked well 30 years ago. And it can work well in the age of AI too. You know, Mallory, the thing that I wanted to point out about owned intelligence that you commented on earlier, first of all, I love the term. And I'm so happy that Microsoft agrees with us here at the Sidecar Sync about the fact that you need to model the behavior here, which I was talking about earlier for leadership. It's just another punctuation at that same point. It's so key. But what I would say about owned intelligence is it's another term for process power. So we've talked a bit in the past about Hamilton Helmer and his book, The Seven Powers of Strategy. And that book is something that's foundational to the way we think about strategy in the age of AI. In fact, we teach that course on the Sidecar Learning Hub all about strategy in the age of AI. And we center the course around how to think about AI through the lens of the seven powers. And I won't go through all the seven powers here, but one of the powers or what you'd call a route to power is process power. It's an exceedingly rare power, but it's essentially this collective wisdom that is harnessed to be able to create differentiated and durable returns. Meaning that this power, like the best example that he talks about in the book that a lot of people know is the Toyota total quality system, which is the Toyota production process. And that process is, it's a power, by the way, it's very difficult to replicate. So much so that even when Toyota tries to help another auto manufacturer implement the Toyota systems, it's almost always a failure. There's something in that culture, there's something in that DNA, literally like floating in the air of the building that defines what that process power is. So owned intelligence is a frame that I think is extremely useful. And it directly correlates to how strong your culture is. You can't have owned intelligence if you have a weak AI culture. So you have to give people the authority, the autonomy, to be able to go do these different modes of working that we've discussed in this pod. And then over time, you have to nurture and grow that owned intelligence, you have to be able to work openly with AI, share what worked, even more critically share what didn't work. You know, Mallory, if you figured out something was an absolute failure with Claude Kowork, sharing that's actually probably more important than sharing what did work. It's great to share what worked too. But people are really hesitant to do that. You know, there's just a fear of looking bad. That's, of course, again, leadership and culture to make failure that comes from experimentation to be totally fine. People just not doing their work, of course, is not a form of failure we tolerate. But people trying new things and not having them work the way they thought is absolutely what we want. So it's a lot of different things mixed into one. I think it's, you know, the old is new again kind of thing that's happening here, that this term is wonderful. But it essentially means a culture that is like, you know, harvesting and nurturing and continuing to grow that knowledge base, that wisdom.

[00:53:30:00 - 00:54:13:05]
Mallory
 I wish we hadn't, I hadn't put this at the end of the episode, because I really do love this idea as well of owned intelligence. But I want to talk about it in practice, Amith, because the idea of capturing signals from what has worked with AI, what hasn't worked, and then re-encoding those back in your organization. I love it. It makes sense. I feel like in practice, that can be tough when maybe one person ran an experiment and in their mind, they know what worked and what didn't work. You obviously have a ton of experience with this. Across the Blue Cypress family of companies with AI products, with the Member Junction platform, can you talk about what that process looks like? Coming out of an experiment and then being able to use those, actually use the insights that you've learned to make products better or services?

[00:54:14:05 - 00:55:08:05]
Amith
 So the fundamental baseline that you need is some kind of observability of your agents. So you need to be using an agent framework of some sort that logs its actions. So you need to basically track whatever an agent is doing something, what its inputs and outputs were. You need to track that information, preferably in a database or someplace where you control, so you can interrogate it however you need to. And then you have to do something with that. So if you have your agents logging all of their executions and you have all the inputs and outputs and you know which ones went well and which ones didn't, then you can actually use AI to help you analyze that data to say, "Okay, well, let's look for common patterns in situations where the user was pleased with the result. Let's look for common patterns where the user was displeased with the result." Both are very valuable signals. So in the Member Junction AI Data Platform, which is our free open source AI Data Platform specifically built for associations,

[00:55:09:06 - 00:56:44:08]
Amith
 we do exactly that. So we have observability built into the agents and there in fact is something called the continuous learning loop. It's kind of, we've talked about it before, like sleep cycles where the agents go to sleep. They don't literally do that, but they basically kind of emulate what happens in a biological neural network when you're in arrest mode through REM sleep where we're kind of processing, "Hey, what happened during the day? How many conversations did I have? Oh, I'm an agent and I had this conversation with Mallory and certain things that she seemed to like, certain things she corrected. How can I learn from that? How can I be better?" That's basically what these continuous learning loops do in a good agent framework. And what that does is it bubbles up this collective wisdom into the agent's memory so that it can serve you better. So the agent isn't the same agent you talked to yesterday or even necessarily an hour ago. It's continually understanding you and your organization better, both at the individual level, but also at the organizational level. And contemporary agent frameworks are doing this already. And then what you can do to bubble that up into the human brain is you can again interrogate that data with various different kinds of analytical tools to understand what's going well, what's not going well, and then to kind of have a meta-analysis on top of all the agents across all of these different time periods. So I think there's a lot of powerful tools out there, but you have to start off with a good foundation. If you're using some consumer-grade agent tool that just, everybody's got co-work on their desktop, there's no centralized observability or tracing of that. So I have no idea what's happening there, which is a problem from a governance perspective, but also robs you of the ability to do this organization-wide understanding.

[00:56:44:08 - 00:56:56:04]
Mallory
 So making sure it's baked into the agent structure itself sounds like a key first step, and then bubbling that up to the humans. But at least you know if it doesn't bubble up to the humans, the agents can continuously get better.

[00:56:56:04 - 00:57:32:04]
Amith
 Yeah, for sure. And there is downside to that because sometimes you mentioned something in passing and the agent takes that as a literal instruction to change the way it behaves, and you can have some variations in behavior. So there's a lot of tuning that you have to do with agents. There's turning on an agent and getting it to work for you in an area like member service or knowledge work or analytics. It's not, I turn it on and just run with it and it's perfect immediately. Just like a human worker, you have to help the agent get better, and you have to pay attention to the way you're training it. And over time, you do have to review how it's performed, just like a human would hopefully get better if you gave them a thoughtful performance evaluation from time to time.

[00:57:33:24 - 00:57:37:17]
Mallory
 Amith, do you have a key takeaway from this episode?

[00:57:37:17 - 00:59:00:06]
Amith
 You know, I talked a lot earlier about the CEO role and how you have to model the behavior. I think that the reinforcement here that I'd like to share as a thought for all of our listeners, regardless of what role you're in, is you are the CEO of you, and you are the CEO of your department potentially, even if you have no direct reports. The agency that comes from being the chief executive or the executive director is something all of us should try to model and in taking initiative and showcasing our own behaviors, and then to share them with others. Even if you're in an organization where you say, "Well, Amith, I heard you talk about how the CEO should do this and this, and my CEO is just not going to do that." Okay, that doesn't mean that you throw your hands in there and give up. You can try to bubble up change from the bottom up. In the Ascend book, we talk a lot about change from the bottom up as well as top down. They're both important. But from the bottom up side, an example of something you can do is, let's say everyone in your organization, other than you, is a stalwart and not interested in AI and all this stuff. You go do something with AI and you go showcase it to people. Don't be afraid to explain that you're using it. Show something you did that will blow people's minds or just let them see the results as you do your work, which is markedly different than what you did previously or what others are doing that are peers of yours. I think there's opportunities for everyone to model behavior.

[00:59:01:17 - 00:59:37:08]
Amith
 It goes back to the ideas that we've talked about over and over on the pod and in our writing. It's that you have to make change happen through incremental and continuous action. You can't say, "This is what I'm going to go do," and then do nothing after that. If I say, "Hey, I want to change my lifestyle. I'm going to eat differently or I'm going to work out," or whatever. If I don't follow through on that with a lot of small actions, it's not going to result in any change. Organizations are like a larger scale version of an organism just like ourselves. They're very complex. They're nuanced. They have feelings collectively and you have to work through all that, but it starts off by modeling behavior at every level.

[00:59:38:19 - 00:59:59:13]
Mallory
 It makes me think of the book Atomic Habits. Amith, I don't know if you've read that one by James Clear, but just little tiny steps in the right direction, being consistent every day, really adds up in the long run. It sounds like from this episode some practical advice we have is to, one, model AI use for yourself, for your team, your department, for your association, potentially managing by walking around. I

[00:59:59:13 - 01:00:05:02]
 (Music Playing)

[01:00:15:20 - 01:00:32:19]
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.

[01:00:32:19 - 01:00:36:00]
 (Music Playing)