This transcription is provided by artificial intelligence. We believe in technology but understand that even the smartest robots can sometimes get speech recognition wrong.

The 229 Podcast: 2026 AI Power Shifts with Reid Stephan and Sangeet Paul Choudary

Bill Russell: [00:00:00] Today on the 2 29 podcast.

Sangeet: Every part of our knowledge economy and knowledge work that we perform, we're about to see a dramatic shift in the design of every system of our workflows, our organizations, our industries.

Bill Russell: My name is Bill Russell. I'm a former health system, CIO, and creator of this Week Health, where our mission is to transform healthcare one connection at a time. Welcome to the 2 29 Podcast where we continue the conversations happening at our events with the leaders who are shaping healthcare.

Let's jump into today's conversation.

Reid: Welcome to the 2 2 9 Podcast. I'm Reid Stefan, VP and CIO of St. Luke's Health System in Boise, Idaho. And today I'm talking with someone who has a knack for dismantling assumptions about technology. He's the author of Reshuffle A Platform Revolution. He advises companies on how they should think about AI platform ecosystems.

I [00:01:00] found reading the book reshuffle his ideas about AI system transformation coordination may be exactly what healthcare needs right now. Sange, Paul c Chaudri, welcome to the podcast.

Sangeet: Thank you so much, Reid. Really excited about this conversation.

Reid: Yeah, so your book reshuffle really resonated with me largely.

Because it helped me understand something we're all feeling a little bit better, which is the ground is shifting underneath us, around us, on top of us. For those listening who may not be familiar with your book Reshuffle, if you could distill it down to a single idea that healthcare IT leaders or any, I guess, really any leader should hold onto as the world reorganizes around AI and platforms.

Sangeet: The central idea that I talk about in the book, which sort of runs counter to the prevailing narrative around AI is that AI doesn't simply impact our tasks and workflows by making things faster, [00:02:00] better, cheaper. It changes not just our tasks and workflows, but also our organizations and our industries by restructuring the assumptions around which all of those things are structured.

A simple way to think about it is that when we often think about the impact of ai, we think of the fact that something that we could, do manually in the past can now happen very quickly. For instance, transcribing this meeting and getting meeting notes out of it, you don't need, a chief of staff sitting in the meeting and doing that for you.

And you can have an agent do that instantly. That's an aptitude of faster but a cheaper, but it's sort of ignores certain assumptions of when meeting transcription was slow and expensive, how were our workflows structured around that? What were we not doing in the past just because that was slow and expensive?

And now that that becomes cheap and instant, what will be fundamentally new things we start doing? what fundamentally new capabilities that you know, we'll start using on that basis, and how will our organizations have to change on the basis of that? And that's just one example, but if you apply that to AI's impact on.

Every part of our [00:03:00] knowledge economy and knowledge work that we perform, we're about to see a dramatic shift in the design of every system of our workflows, our organizations, our industries. And we don't, we're not talking about that enough yet. And that's really what the book talks about.

Reid: So I want to talk about the human element of it, because oftentimes, mm-hmm.

The conversations, there's the existential fear about what this means for the workers. Sometime last year I heard a phrase or read a phrase that said. You won't lose your job to ai, you'll lose your job to someone who uses ai. Right? And I don't think I thought about that critically enough. I just accepted it as a universal truth.

I repeated it dutifully. One of my favorite ideas in your book, reshuffle, was that you challenge this concept, and you argue in my words that this is a comforting kind of adage, but it's fundamentally flawed because AI doesn't just improve the worker, it reconfigures the system. And you touched on this a little bit, but maybe can you walk through a little bit more about why this distinction matters so much?

Sangeet: Yeah, absolutely. This whole idea of [00:04:00] AI won't take your job, but someone using AI will. It assumes that the impact of AI plays out in only one of two ways. It either substitutes you, replaces you, or that compliments you and augments you. So it's automation versus augmentation. What both of those frames assume is that how your job is structured, why your job is structured, the larger workflows within which you participate, and the organization within which you sit, all of that is going to stay the same.

It's just a going to be a question of whether AI does the job or whether you do get to do the job using ai. And that's the fundamentally flawed assumption over here. Because all of that, as I mentioned in response to your opening question, all of that is poised to change. And it's not something that's unique to ai.

It's something that we've seen all through whenever technology comes in and changes how work is performed. And just to give a simple example, one of the examples I like that I use in the book, uh, as well, is this, idea of the impact of the. Arrival of the word processor on the job of the typist.

Mm-hmm. [00:05:00] So that example is interesting because the, we often assume the job or the role of a typist as being about the task of typing. And that's also, you know, it reflects in the in the job title a typist types. So when, uh, when the word processor came in, you would've expected that it'll either be the machine taking over the task of typing or the machine making humans better.

And so Typists would've said, well. Clearly humans still have to type. So, a, word processor is not gonna take a job, but somebody using a word processor will what they really missed in that conversation was, while that was true, it did not play out the way they had expected. Type started trying to figure out how to, you know, chain their job and on the word processor, but what.

Really happened was that with the arrival of the word processor, everybody became a typist. And the structured role of a typist no longer made sense. And that was because the job was not structured around the task. It was structured around the constraint it was solving in the system. Mm-hmm. Before the word processor, [00:06:00] the cost of editing documents was very high and mm-hmm.

And because the cost of editing documents was high. You needed or the value of typing skills? Good typing skills was really high because it reduced the cost of editing. But when the cost of editing went down to zero and you would, you know, associate this back to how we talked about the cost of transcribing meetings going down to zero when the cost of editing went down to zero.

The skill of typing was no longer valued That highly or the skill of having good typing skills was no longer valued that highly. And so anybody with reasonably good typing skills where the word processor could now start typing. So that's what really took away the job. And that's, those are the effects of AI on our job landscape that we risk ignoring when we start using things like AI won't take your job, but someone using AI will.

Reid: Yeah, and I love that example you gave. I wanna talk about a healthcare example and then just get your take on this scenario because this is a, a real thing that's happening pretty widely across healthcare. This rise of the, the AI scribe so that physicians can finish [00:07:00] the documentation faster and it's been wildly successful and it's great, but then.

Thinking about what you just described, the system around it, the handoffs, the billing, the coding, staffing, scheduling, patient throughput, that maybe really hasn't changed. So is this a case of this fallacy where we focus on upgrading the task performer without then really redesigning or thinking about how we might redesign the entire system?

Sangeet: Yeah, absolutely. I think that's a great example because, when you. Only stop at deploying the AI scribe, you are assuming, or you're really just changing the throughput of a particular task. What you're not really looking at is what were the assumptions based on which the previous workflow was structured.

So a previous workflow was structured based on the assumption of slow transcription, inefficient transcription information lost during transcription. When an AI scribe comes in, a lot of those. Issues start, getting resolved. So with that issue getting resolved, [00:08:00] what are new workflows that should be set up?

For instance insufficient transcription versus particular, you know, fully sufficient, fully efficient transcription. What does that level of information now unlock? Where can that information be used elsewhere in the organization and how does that change you know, the workflows where it can be used?

So when. You know, when a constraint changes, you have to start redesigning or rethinking your workflows based on what was the assumption based on which the previous workflow was structured. If it was structured based on slow information coming in, insufficient information coming in. And now that you have all of that information coming in smoothly, do you need to rethink your workflow?

So the first thing is think about what constraints were your previous workflows managing, and if those constraints go away how should you redesign your workflow? It's not just about improving that particular task. The second thing is now that you have the ability to use ai, not just as a scribe, but.

Eventually as some form of an organizational [00:09:00] brain where the inputs from all the doctors is being funneled into a central place where the organization can constantly learn. How can you think about this organizational brain working alongside not just the doctors, but everybody else along the workflow

job because of the ability to manage all of this knowledge much better and serve that knowledge across the workflow to the various people. So you have to think about this systemically, not just at the workflow level, but also at the organizational level. And how does everybody else's job change or, roles change because of this new intelligence available?

Reid: Yeah. As you were describing that, I was thinking about, you know, the AI scribe is really about that clinical documentation, but then if I step back and look at even the bigger system. So we have clinicians who document coders in WHO code based on that billing. Then does their rework. Payers do audits, then there's denials, then there's anger.

Then we kind of rinse and repeat, and we treat these as separate tasks rather than a single kind of architecture. And so what you're describing [00:10:00] then AI can just reimagine that entire architecture and from just these siloed lanes of work of documentation, coding, billing. Really from the ground up, designing this unified clinical to financial model that minimizes the human translation work that's so laborious and fraught within inefficiency today.

Sangeet: Yeah, absolutely. And I think that is something that we've already begun to see in, in many different industries where if I just take an example from a completely different industry, if you think of how fashion used to work, and this is an example I use in the book as well, traditional fashion used to work on the assumption of long market sensing cycles.

You had to. Have somebody fly over to Tokyo, Milan, palace figure out what was in vogue, and then come back, make designs on that basis, and run large seasonal collections, large product production runs of a seasonal collection. And what you see with a company like Shane today, which is, a Chinese fast fashion company, is that it?

[00:11:00] Totally changes that system, because the constraint of sensing the market and testing new production runs goes away. What Shane does is it's constantly taking social media signals to determine what the next trend is. It then uses a factory in China to produce a small batch. Runs that test in the market, immediately gets feedback and then determines whether to scale that particular batch or to move to the next batch based on the next hypothesis.

And so it's not just changing the workflow, it's changing the entire architecture of fashion. It's changing the speed at which new styles can be done in the market. And it's actually changed the nature of the designer's job, not because, aI has taken over design. It's not generative AI replacing designers, but it's more because with every production run, the entire system is learning what works in the market versus not, which design variations work versus not.

So the system is learning faster than the individual designer and hence is able to chip away at the individual designer's job. [00:12:00] It augments them, but eventually. Shortens their role as well. So these are all aspects that we need to think through when we think about constraints moving away with AI coming in.

Reid: I think there's value in learning from other industries and applying, especially historical. One of the examples in your book that really captivated me was, the contrast of Kmart to Walmart, right. In terms of how they looked at barcode scanning. And I'm simplifying here, but Kmart then kind of just drilling in on the task based level augmentation benefit of that.

And they kind of just then had a slow long path to I relevance, Walmart, then like reorganized all around that, created their satellite system and just really realized the value of the data they were now collecting and now their market cap is closing in on a trillion dollars. So it's really remarkable.

Just based on that, what advice or warnings would you offer to healthcare leaders who then might not be thinking about this the right way and what that might mean [00:13:00] for future viability?

Sangeet: well, I think, the Kmart versus Walmart example is, really interesting because it's not just saying that the system will be redesigned, it's also saying that the power structures within the system will change.

Who holds power within the system will change. And that's the real idea of the reshuffle that I talk about in the book. Okay. Which is that you're going to reshuffle the power structures. So Kmart changed you know, when barcodes came in, they made. Check out faster. The automated, the checkout job and then people at the checkout counter moved onto other roles.

So all of that is great. It was very local effects. Walmart changed the relationship between the retailer and the manufacturer because before that, the manufacturers had bargaining power because individual retail stores used to negoti. With the manufacturer, but with all of the barcode data coming in, you could now negotiate at the network level.

Entire Walmart network could negotiate with a Proctor and Gamble. And today, this is obvious because we live in a world of data. But back in the 1980s thinking about [00:14:00] data as a negotiation leverage and capturing that across the network was really new. So I think that's you know, we have that kind of a moment right now with AI because what the barcode did to sales data back then.

AI today does to all kinds of unstructured information that we use. There's a ton of unstructured information in our organizations. You talk about the AI scribe, every call that comes into an organization, every documentation that you generate and with ai you don't even need to generate documentation can generate documentation using inputs and artifacts.

So making sense of all of that. Can fundamentally reveal and restructure power relationships. I'll give a few examples of that. That probably helped The Kmart, uh, you know, helped to bring the Kmart, Walmart example closer to life. Think about the rise of AI agents in, customer service

in call centers. Right. On the one hand, initially when, we started seeing agent capabilities we just jumped to the other extreme and said that. We won't need human call center agents. And then, clearly that's not how it works. [00:15:00] So we kind of ricochet between the extremes of AI will do everything versus humans will always be in the loop.

But all of that argument there's a lot of nuance lost in that argument because, what's important is not whether AI is performing the job or whether they're the presence of the human or not. What's important is what's the new system going to look like and who's going to hold power in the system?

Going back to the Walmart example, who's gonna hold the new power? So if you think of projecting, what happens once AI starts getting used in customer support, there are several systemic shifts that can happen. The first is that, the human role changes. Even if the human is in the loop.

The human is not being augmented by ai. Instead, the human is filling in the gaps which AI cannot fill. So if an AI agent can handle a simple query and 80% of the queries coming in are fairly straightforward, then it's the 20% edge cases. That need to be handled with you know, human intervention and there are escalations logic that are already doing that.

But that, that, that effect is very [00:16:00] obvious. If you take that a step further, the very logic of customer support is about to change because customer support is very reactive today because the cost of serving support into the product is very high. For, with most products, you cannot also buy. You know, support on the side.

You cannot have a person working with you helping you use the product. But with AI you could potentially have an assistant embedded in the product usage, which means that your need for reactive support would go down fundamentally. So any power structures that were built on solving the reactive support need that power shifts to whoever is solving the proactive support need embedded in the product.

Right? So there's a, there's gonna be a shift of power from whoever solved that problem to whoever, whoever can embed it. Not just into the product, but into the value chain as well. You know, whoever sells it to you is better equipped. A third way to look at it is think of the, software providers who are serving that market today.

There are players who are serving call centers, contact center software as a service players who [00:17:00] provide the software to manage all of these calls coming in. In the past, they were competing with other contact centers, software as a service players, but today they're increasingly competing with anybody who is using AI to serve the customer journey.

So if you were, say. You know, a company like Genesis, which provides software to call centers. Mm-hmm. Today, they're competing, not just with other companies like themselves. They're also competing with Salesforce, which is not just in customer support, but in CRM. They're competing with ServiceNow, which is handling tickets across all kinds of workflows in the organization.

And they're potentially competing with Amazon, which is also with Amazon Connect, launching a competing capability. So. The logic of competition, who becomes your competitor, who gains power, all of that changes, and that's about to happen in healthcare as well, for very much the same reasons.

Reid: Yeah. That's fascinating.

You know, every healthcare system I know has some form of a vision statement that talks about patient-centered care, the idea that the patient's at the center, and yet in a lot of organizations, the patient [00:18:00] holds the least power. And so I'm just really curious about how. This moment we're in maybe kind of reshapes that and puts power where it should be and will cause health systems and then have to operate and think differently because of that power shift.

So a lot to think about there. One of the ideas in your book that hit me the hardest was what you described as this notion of coordination without consensus.

Sangeet: Mm-hmm.

Reid: And in healthcare, that sounds like magic because in healthcare. Not unique to us, but in healthcare specifically, we coordinate by having meetings about meetings that lead to subcommittees that maybe eventually will approve a pilot someday.

So how do platforms and then increasingly ai, how do they enable groups to coordinate their actions without ever aligning through consensus? In healthcare, you hear this phrase sometimes that because we're so consensus driven that 99 0 1 is a tie, which then just slows down decision making. So how do your concepts maybe address that and how should we think about [00:19:00] that idea of coordination without consensus?

Sangeet: Yeah. I think, when we think about, how coordination has played out there are two things that are important for multiple actors to coordinate and, one is that they have a shared view of the world, and the second is that they have a shared language through which they're coordinating.

And the later third thing is that there's agreement that they want to share their resources with each other. So you need a shared view, you need a shared language, and you need to be open to sharing your resources. That's, so the third part is the consensus part. But you always needed these three things.

There are two ways in which this has happened over the. Past you know, several decades. You either had coordination with consensus, where you got everybody in a room, made them agree to a standard like a DVD standard, and then you had players working together on that standard or you generated such high.

Access to demand that you were able to force that consensus because of your access to the market. So, you know, apple and Android cornering the smartphone [00:20:00] ecosystem or Google doing it with search and Amazon, with, e-commerce. But again, those things happen because they leverage a, disruptive technological shift to capture demand at scale.

Arguably chat GT is doing that with AI today. But, you know, we've increasingly seen those two types of coordination. What I believe is beginning to happen a lot with AI is that you don't necessarily need to get to huge market power. Nor do you need to get to consensus before you can start organizing actors together.

And, uh, I take the example of an industry where I've worked quite closely and I've seen this happen. It's in the construction industry across the construction value chain from designing a building to, planning to constructing and delivering the project and eventually to operating it.

There are different players from designers, to engineers, to contractors, and all of them use different software tools. They come out with different artifacts and so far, none of them have had a way to talk to each other and coordinate, [00:21:00] except through exactly what you talked about, meetings, emails you know, structured communication.

What is now beginning to happen with AI is a whole range of startups that are coming in. Which essentially takes the outputs of all of these different tools, the PDFs, the model diagrams, the construction plans, and they make sense of all of that to see what's the current state of the project. Is it aligned with what the owner wanted

how do we benchmark it or how do we proactively simulate where the project could go wrong? How do we proactively reduce cost overruns, which might happen later because of making sense of all of this information? So. By creating value out of unstructured information and giving that back as analytics and insights and recommendations and alerts to the various stakeholders, they're implicitly driving coordination without consensus.

If you think of an individual stakeholder's perspective, you'll think that they're just taking all these different documents and emails that they have, and they're trying to make sense of the information. [00:22:00] But when all the stakeholders start doing that, and they're working off a common model of. A common view of how the project is progressing, and they're starting to increasingly use a common language on that basis.

That coordination has started, even though they've not explicitly agreed. And once, a single player becomes the defacto decision layer where you throw in all of your unstructured information and make your decisions. That's when you can then start enforcing that coordination over time because you've kind of captured demand at that point.

And so I think even in healthcare, you know, you, you had this point about patient-centric healthcare. There, there are obviously issues. Around patient centricity because of who owns the data. Mm-hmm. That is a separate question that needs to be sold independent of the AI side of it, which is data ownership.

Who is the agent working on behalf of? Because you could create a patient centric model in the sense that the agent is on the patient side, and yet it could be working on behalf of a platform rather than on behalf of the patient. So there's one question to [00:23:00] be answered in terms of who is the agent working on behalf of, but.

The second piece of coordination without consensus. If as a patient, I, I can, even if I don't have access to structured information and structured data, if I have access to my medical history in some form unstructured information, unstructured exchanges. And if I can feed all of that and make sense with an AI assistant who is catered towards that particular use case, that starts helping me gain power back and whichever layer or agent is helping me do that.

They start gaining that power. So there's a first layer of creating that coordination without consensus because the patient has another agent working on their behalf. Eventually there's a second layer of answering the question, is the agent working for itself, which is for a tech company? Or is the agent working for the patient because of how it's been structured?

So those are, questions that we need to think through when we redesign the system around the patient.

Reid: Yeah. There, there's a lot there to [00:24:00] unpack. So thank you for that. Last question as we kind of get to the end of our time here, more of a mindset question. So for CIOs, healthcare IT leaders who are our primary listening audience

Sangeet: mm-hmm.

Reid: What are the capabilities or the mindset shifts that are gonna be most essential for them, for us to make in a world where AI is reshaping our systems? Task level optimization won't be enough. As you just described, coordination emerges from data rather than committees.

Sangeet: Right. Yeah. I think there, there are a few different things I'd like to call out over there.

The first is don't think about your internal organizational workflows in terms of the tasks that need to be speeded up. Question those workflows, because they were structured around certain constraints and when those constraints go away, a lot of those constraints were around the. The cost of executing knowledge work, whether it was cost in terms of actual money or in terms of time, and the friction associated with it.

When those frictions and constraints go away, it's an opportunity for you to [00:25:00] reimagine those workflows. When you do that, that helps you reimagine roles and organizational structures. So there's a significant, very significant, uh, reimagination opportunity that's going to be there. Complimenting that is the fact that if you really want to reimagine, you can't do that within the structure of your workflow only.

You have to think about the identity of your business, of your team. What do you do today and what constraints or assumptions is it centered on? And if those constraints are going to go away a year from now, let's say one constraint is that we have a certain identity today because the cost of accessing data for the patient is very high, and hence they cannot manage their whole patient history. But if there were technological shifts that were to, eliminate that constraint two years from now, what would your identity look like at that point in the future? So you need to keep doing this forecasting based on.

The bets you're placing on which constraints go away when. So while you are thinking about reimagining your workflow, also think about reimagining your identity [00:26:00] and ensure that what you are, the transformation you're running internally, is helping you move in the direction of that identity. The third thing I'd say is that AI is different from other technologies largely because.

It handles things which were seen as being tacit and implicit to human advantage. It starts, you know, taking over some or a lot of that work increasingly so. It's less about is AI really intelligent? It's more about the fact that a lot of knowledge work, which involved human intelligence can be replicated using ai.

And the second thing that's, unique about AI is that unlike most of the technologies. Agents in particular can be goal seeking. So you can give them a goal, they can figure out how they plan their resources and paths towards it, and they can figure out a way to achieve that goal. They may not be perfect.

Depends a lot on the deployment and on the sophistication of the technology. But we are on an arc where technology's gonna be goal seeking alongside us. When these two things happen, [00:27:00] you often start thinking about AI as digital labor. You start thinking about how can I think of my teams as humans and ai?

And I think that's the wrong way to think about it. You have to think about AI as organizational capabilities, just like you also thought of, you know, a actual work performed by humans in teams, in workflows as organizational capabilities in the past, and really think about which of these new AI capabilities are going to be unique to your new identity versus which ones are going to be commodity in the industry.

The ones that are going to be unique, you don't want to buy that from. Technology provider, you want to start building that right now. So they're important build versus by decisions that CIOs will be confronted with as well. But you won't be focused on those decisions if you keep framing AI as digital labor.

So these are, you know, the top three things I would think of if, thinking about what a CIO should be thinking about today.

Reid: Yeah. That's brilliant. Sange, thank you. This has been a, just a wonderful conversation full of ideas that healthcare desperately needs right now. Moving beyond, we've [00:28:00] talked about task automation, rethinking our system architecture.

This important concept of coordination without endless consensus. I think this will be one of those episodes that requires a couple of listenings because the things you talked about, there's incredible depth there and wisdom. So thank you. I can offer a personal recommendation, uh, for my friends.

Pick up a copy of reshuffle. Uh, it may just change how you look at your org chart, which is probably overdue for a refresh Anyway, so thank you all for joining us, and again, ASEN, thank you so much for your time.

Sangeet: Thank you so much, Reeb. It's been such a pleasure. Thank you.

Bill Russell: Thanks for listening to the 2 29 podcast. The best conversations don't end when the event does. They continue here with our community of healthcare leaders. Join us by subscribing at this week health.com/subscribe.

If you have a conversation, that's too good not to share. Reach out. Also, check out our events on the 2 29 project.com website. [00:29:00] Share this episode with a peer. It's how we grow our network, increase our collective knowledge and transform healthcare together. Thanks for listening. That's all for now.