Deploying Claude across your organization

0:11 Welcome everyone. I'm excited to talk to you about how co-work is actually playing out in the real world within a ton of different organizations and enterprises. I'm Jonathan Dolberg. I head up our applied eye team that focuses on industries. These are companies in healthcare, in finance, in manufacturing, in retail, in a ton of different organizations and teams that are sometimes struggling to figure out how to adopt AI. And I have the pleasure of really digging into how they're using co-work, what this means in terms of plugins and some of the adoption patterns that are again playing out live as we speak. Now, a quick framing for where we are and how we got here. Back in 2024, just 2 years ago, Claude was really an assistant. It would answer questions. It would help people accelerate what they were doing. It was an autocomplete. It really helped people work more quickly, but it didn't completely delegate actions for them.

1:03 And for me, what really changed in 2025 was cla code. This was true for a ton of engineers. For the first time, instead of trying to find three, four, five hours in the day where I could have dedicated time to code, find solutions, work in a terminal or an IDE, I could suddenly delegate things to Claude and have it magically just do things for me. This was the biggest step change that I think most people on Twitter, if you read that in the news, really experienced at the end of last year, back when people were on their winter break and starting to experience this firsthand. The big shift, you suddenly had a developer, someone that you could actually pair program with, talk to, plan, but you could also just give it tasks and you could give the next claude another task. And you started to see people realized that there was a fundamental change in how engineers were operating. They became more managers and orchestrators. The vision and taste that they had mattered more sometimes than the individual technical skills in a given language or given implementation.

2:04 But you still had a barrier to entry. You still had to be comfortable with an IDE. You had to know what that meant. You had to know what a terminal was. And we saw some people in legal, in product, in design, in a ton of different areas starting to use this in pockets, but we didn't see widespread adoption. And now in 2026 with co-work, we're extending that out in a way that really shifts what's possible. I have always been probably unduly proud of secretly my Excel skills. My first job out of school, I lived in SAS, in Excel, and PowerPoint. And I still to this day know alt codes and Excel shortcuts that no longer exist. And I do not use a Windows machine every day. So, it's pretty much useless for me. But what I always thought in the back of my head was whether it's in Python, whether it's in Excel, whether it's in various dashboarding software, I could always do something to look further into the data.

2:57 I could take a dashboard and then do that next step. But I rarely would. I would just take things at face value, be a little bit annoyed that I couldn't dig deeper and just say, "Well, it would be nice if our BI team or analytics team would do that, but it's not worth the effort of trying to wrangle exactly what I mean and what I'm looking for because it's kind of one-off." And even for me, somebody who has excitedly been using cloud code and whose team has been embracing this and rolling this out in a ton of different companies, co-work changes how I operate. And more importantly, it changes how my colleagues operate. I'm now working with sales leaders that have built out command centers that let me click in and see everything from their hiring pipelines to what their sales prospects looks like to actually getting information on risks, blockers, feature requests that I can bring back to product. And I think a lot of this is the interface, having this available in something that's not the terminal that gives me an interactive way to really ask questions, challenge what I'm thinking, and get results back. Not just again a faster autocomplete changes how I operate day-to-day and it has unlocked a ton of productivity for me personally and more importantly it's enabled people that don't use the terminal that are not familiar with what cloud code is to suddenly start to have the same capability day in day and day out. So a little bit deeper, what is co-work for me? It is a way that anyone technical or otherwise can suddenly start to delegate tasks and again have that paired programmer experience of having an assistant or a very good dedicated focused colleague that never sleeps is happy to do any task no matter how menial go and try something for you. You can give it a problem and get back a result. You can suddenly see what if I did this a little bit differently. If I did some AB testing, if I did some sensitivity analyses, if I redid this slide in a different form or in Excel or something else, by meeting people where they are, where they operate, it suddenly unlocks the capabilities of a lot of this that we're both locked behind technical limitations. You have to know what a terminal is, but also from a change management perspective.

4:59 The biggest piece what I'm suddenly seeing both internally within Anthropic and with a lot of our customers and partners is that co-work sets a floor for quality. your Excel wizards, your PowerPoint designers, you can suddenly take what they do, build that into a plug-in, have a workflow that makes sense for not only your company, but for your team, that specific pattern and use case, and apply that everywhere so that people can build on this. They can still delegate tasks, they can accelerate what they're doing, but it's doing so not at the cost of quality, but at additional speed with better quality and better interactivity.

5:36 And where that really comes into play is that you suddenly accelerate that output. It's actual work being done. It's hours saved. It's faster, higher quality across the board. And this is what this looks like. If you suddenly start using co-work out of the blue you get access, you start to see that claude understands what you're doing. And this has always been the case. But having this in a new form factor within co-work itself lets you start to realize that when claude challenges you on some assumptions when it says why not try this it's understanding what's happening and as part of that it's then building out a plan you can see what the next steps will be you can say maybe don't start there maybe try a few different things give me back some intermediate steps let me try these in five different forms five different ways to present this chart that for a lot of things is the first step to adoption because you don't really change anything you're you just become again better at what you're doing without fundamentally changing what that looks like. But the next step and not everyone takes this step and we often partner with people to figure out the right ways to really change behavior is moving beyond that planning and understanding to actual execution.

6:42 getting the right step of delivering better results with Claude checking the quality of these and verifying that they're accurate, that you have built-in capabilities to go and do reference checks to see how Claude is thinking too and trace back to what it was doing, why it was doing it, how it was following that plan. That fundamentally changes again in the same way for me back last year, how people are adopting and using AI in their day-to-day. It changes from an accelerator of things that they're using to actually adjusting what they're getting, how they're doing it, and how they're actually interacting with that output. It lets you prototype. It lets you build. It lets you accelerate what otherwise would be very painful things that you could do but don't do. And if you look at this, a big piece of this is actually connecting to your data. It's living where you are. And it's less about a specific feature or functionality. It's more that you can suddenly collaborate with someone that again knows you, knows your business, and knows the workflows that you're operating on. And if you find the right way to operate here, this is the biggest thing. You suddenly have a sales team that can connect to their CRM, to their pipeline, to their processes, and you can build that in. If you have one person who is amazing at forecasting, how are they doing it? What sort of data are they looking at? You can have this as the baseline for everybody, and you can extend that to the legal team. We have other people who are speaking to some of the work that I think it's Mark Pike built for our legal team. He built a plugin in an afternoon that's accelerating how our team actually can review contracts and build those out.

8:07 The same thing is happening for product for analytics for even our people team for nonsensitive data that makes sense to accelerate again some of the drudgery of living in Excel changing things and powering a lot of these dashboards. At the end of the day, if you think about all the work that I might be doing or someone on our finance team or marketing or legal, you have a workflow and you have domain experts who know what they're doing and have kind of been living and breathing in the space for years or decades. A plugin bundles all of that together. It connects in the right skills. Think of those as like the design format for PowerPoint that your company uses or your team uses or how you actually structure an Excel file. It takes all of the different data connectors and most importantly it takes how someone would use all of those things and builds them again as a floor that you can just get up and running. It reduces the friction for someone who's new to this or new to the company or new to this workload getting up and running again at the same quality floor that would take them weeks or months or never to really learn. And it's easy. It's something that doesn't take a developer or an engineer to build. It's something that you can write down that you can actually build and integrate or take from the outside. We have a lot of external parties that are building these plugins too that we're trying to expose to make it easier and easier to get up and running and not only get up and running but start to build on top of these foundations and further customize them so that whether it's you, your team, your company, some combination, you end up with a combination of these plugins that accelerates your work and again changes claude from an autocomplete or a collaborator to something that is actually working on your behalf that you can delegate things to that you can see the results back day in and day out. Now as you're actually starting the first biggest piece that you see is just time savings. It makes everything faster. The second is that quality of output. You suddenly get better floor raising capabilities. And the third you're extending this. You're helping with discoverability for how people are doing this at scale or across different teams or different individuals. So that again you're not sacrificing quality here for speed.

10:10 You're combining both faster acceleration with better results and better extensibility. The big piece you need to connect into existing systems. You need to start with workflows that are used day-to-day and you need to then scale that. If this is just a one-off, we see that this gets adopted in small pockets or people. You have amazing sort of change agents within an organization and it can die there. So having them take that final step of building a skill, helping to create a plugin that then is where we see that step change in adoption and the capabilities extending out through an organization. Now a couple of examples.

10:46 Deote incredible organization. They operate at a scale that is way beyond anything that Anthropic does. And the way that they're using this is to take what their best consultants are doing, not engineers, not like super technical people, but the teams that live and breathe in PowerPoint and Excel Word docs and building in plugins is what's helping them actually build prototypes so that instead of going back and forth internally for days or weeks, you can have a consultant build a prototype, look at it, and then bring that to a client or bring that to internal stakeholders. They're also starting to create custom plugins. These are domain specific tasks across accounting and sales legal all of these different dimensions. A little bit different view if you think about what Zapier is doing.

11:29 They're more so connecting out to their tech stack into Jira into Slack into documentation all these things that again someone who's been at the company for a while could access if they know about these and building claude to actually figure out where the bottlenecks are and not only identify those bottlenecks but start to synthesize that data run these analyses and bring this together so that as they're digging and drilling deeper this is like that example I gave of how I start to use co-work to accelerate analytics and go beyond a dashboard you start to be able to ask follow-on questions and they're using this at scale, especially in product and design and teams that otherwise might have been working with engineering but didn't have enough capacity to meet every single day to ask questions that they want to ask but couldn't get the data for or the information. So, Zapier is a great example of how crossunctionally teams are starting to adopt co-work to really bridge the divide between teams and make it easier for them to collaborate together and again raise the acceleration and quality of this. And finally, if you think about skills, plugins, the words don't matter as much as thinking these as workflows that can be packaged up.

12:44 to track everything you've done over a year to make sure that you're filling this in against competencies and matrix and complex sort of like ways to measure the impact that someone has had. But it's hard and if you think about people having a day job, you also often end up in circum or certain circumstances where Claude can now prompt an engineer to say this is an okay example, but what about this? Can you go a little bit deeper?

13:07 can you expand on what you were doing or re-ransate what you said from a technical perspective that only you know to something that your manager or wider team would appreciate and what ended up happening they built a plugin for this in I think it was under 45 minutes it again is not a huge lift but what that is doing is both saving time for every single engineer and wider teams going through these performance reviews it's making these better it's adding more comprehensive information and they're already starting to look at ways to proactively identify and plug in with Jira and other systems to enhance and complement what that information looks like. So across these three examples, there are phenomenal examples of real world companies using this to either help with prototyping, tying together different systems and teams and collaboration or really just reducing the drudgery and inc and increasing the quality of day-to-day tasks or things like performance reviews. So the way to get this going, start with believers.

14:04 Start with those change agents and don't leave it there. You need them to actually codify what they're doing into a skill, into a plugin, into something that can be shared and extended and show what they're doing. Not just a demo, but a real world application. This is where you see adoption. You have people say, "I want to do the same thing." That's the biggest step change that we see. AI and Claude and co-work make it easy to go beyond demos into real world prototypes and actual working applications and workflows and then share what the results are. Do lunch alerts. Highlight the wins. Highlight what time is being saved.

14:40 We have a ton of resources to help here too. This is some asynchronous blog post and resources and getting started. You can contact us. Let us know how we can help. We are more than happy to talk about what we're seeing and help with everything from rollout plans to evaluation to giving you again a sense of how this actually works in the real world. And we are also partnering with a huge range of companies. Please just try it. I think that you'll find that both the models themselves and the form factor of co-work change different in how you can actually adopt this. But let me actually talk in a more concrete example. I have the pleasure of welcoming Seth to the stage from Epic and we're going to chat about some of what he's seeing again in the actual companies and teams that are adopting this.

15:28 Thank you for braving the blizzard. It has been an interesting time here in New York, but it's also an interesting time over across the world and over in Madison. Can you walk me through over the last few years, decades, whatever the time frame is? How did you start to get into AI? Was it a sort of step change? Was it gradual? >> Gradual I think you know I think I think about it in sort of two layers.

15:55 one was a establishing kind of a a fertile foundation. so my background is in mathematics. while at Epic I I've been there for a while now, I spent a number of years focusing on the lowlevel architectures. How you scale systems for the next generation of technology and across applications to support the large cell systems and then simultaneously we require developers at the company to do immersion. What that means is they get on site at the elbow of the users and the key there is understanding the workflows what's happening within the organization what comes next step by step and where the software aids that where human interactions take place etc. So that foundation was set and it ended up serving me as I started to use AI both personally and professionally to identify and help work across the company in ways that it could be implemented. And I think a key there is when you're using it personally, you end up being open to these moments of surprise. Mhm.

17:06 >> And I think over the last 5 6 years, you started to see those moments come faster and faster and they start to build on each other to inform what you might do next. >> Are you seeing it accelerating? >> Yes. The short answer is yes. I think it's but you're also predicting that it's going to accelerate. I think that when you're spending time in it, you start to acclimate to the change in a way that you can then work with your teams and with the organizations that we have the privilege of providing software to to account for and anticipate that changes in new ways. Well, of course, the core models are improving, the cloud models are improving. but also these agentic toolings, these things that are kind of a layer above the models are also improving. And as you're watching both of those compound, you're able to then take it into those workflows, into that experiences that developers had when they were elbowto elbow to anticipate how that can change and help acclimate to it. Taking a step back, I mentioned sort of the idea of change agents or champions or things along those lines.

18:24 >> Did you start with a person, a team, sort of fits and starts? Like what did that look like? >> it was a it was a very natural progression that came out of having open exploration be a possibility across the entire company, >> all roles. And what you started to see was pockets develop. It would often be an individual, but then it would be kind of their neighbors. We're we are kind of famously an in-person company. and a key part of that is this natural interactions you might have at the coffee cart or in the hallway and these quick notes that just get traded back and forth. the excitement of something you build over the weekend on your own using cloud code as an example and showing it to a colleague generates excitement going forward. And so we ended up creating at Epic a number of internal tools to help folks see who was using it and what they were doing with it and then have others be able to adopt that quickly and try a similar thing.

19:35 >> What's the coolest thing you've seen so far that you could at least share >> internally or externally? I mean I am deeply excited and so one of the for folks that don't know Epic we provide healthcare technology software and general public that generally means my chart which is in their pocket and we now have capabilities for folks to not only directly understand their lab results and be able to discuss those in my chart but importantly connect back to their care team and have the same their agents have the same tools that they have within that and I think that type of connectivity to be deeply involved in their care provides an opportunity for people to be healthier. so I I can't not talk about that because that's what we do.

20:32 but I think internally some of the most exciting things I have seen have been non-developers taking tools like clawed code and applying it into their own workflows. this might be helping get to the root cause of a customer issue more efficiently by understanding both the knowledge bases of old tickets that we have had and the code base and being able to suggest and pull those different disparate pieces of information together to help suggest to our technical staff what they might want to investigate next or build that they might need to configure in the system. and having that come from the ground up by allowing this type of open exploration.

21:19 >> We I think had a hackathon that just ended this last weekend and we found well my dad's a cardiologist so I avoided medicine as long as possible but no epic well and I think it was the third place winner was a cardiologist who was building things to investigate like patient activities and things like that. I think the first place winner was again a non-engineer and second place was a software engineer that was using cloud in various ways but some of the things that I've seen even with an anthropic have been the teams that are not engineers that are coming up with new solutions are some of the most excited to show what they're doing because it suddenly is something that they could never do.

21:57 >> You know it it presents an opportunity to communicate ideas and concepts in new manners. Yes. >> So, I had a pharmacist at Epic that I know well, and on a trip up north, he came back with a full prototype of the new pieces of functionality he wanted in my chart. And he was able to show that to others in a way that previously would have been maybe slides or some written documentation >> or 10 meetings, >> right? Or 10 meetings. And now it's a quick link and and folks are exploring it and then iterating on it together.

22:37 How do you actually get people to not only want to do that, but start to find the time to do it? And you mentioned sort of giving people the space and the energy, but there's the reality that not everyone people have day jobs, frankly, and they know what they're doing, and they know what they're doing well. >> I mentioned a lot of things coming from the ground up. I think this is one thing that should come from the top down >> and that starts with leaders at organizations spending time with these tools every day. They need to lead and they need to have that familiarity and sense of where they're developing and then use that to both extend time to their employees to explore. We took we have a Friday afternoon meeting with many of our development leads across the company where we deeply investigate new designs that are crossf functional in regards to our applications or other concepts. We took an entire one of those meetings an entire afternoon and said all of you need to sit down and just experience this yourself. You need to have that aha moment. And it wasn't about those four hours. it was this establishes a foundation for you to continue to iterate and now you've opened that door.

23:58 U many of them had done it already and so there was cross coaching between them. but I think it demonstrated the importance of it to the organization. >> We've been experimenting with similar things internally where new managers come in and need to ship a PR. They need to actually have built something and understood what these tools are. >> It's hard. I guess the biggest piece that's hard is often getting people to not only try this but keep on using it.

24:22 I mentioned some of the different aspects of plugins helping with discoverability and sort of extensibility. How do you find that you can help we're reframed? Do you need consistency in what people are doing and to sort of standardize or is it better to let a thousand flowers bloom people experiment and use what they're doing? >> I I think the latter. once you establish something that is deeply meaningful I think standardization can happen but I think if you start with standardization you're going to miss out both miss out in regards to what's possible but I think importantly what we've been finding is that internally you don't anticipate some of the bottlenecks in your own processes and until you kind of share these tools across the company you don't see things like your maybe legal review processes that you have internally for uses of new tools becoming a bottleneck in ways that they hadn't been at the speed they were before as an example.

25:22 or even simply getting access to and deploying a website which sounds simple but you suddenly have non-developers doing this in a way that they had never done before. And you start to recognize the opportunity you have and you can quickly solve these problems. They're not hard, but they're hard to anticipate. >> Yes. The experimentation part we see across the board. You need people just trying things and using things and >> seeing what develops. How though do you go from seeing that having a thousand flowers. You work in one of the most important systems in the US. Healthcare matters. If you mess up, patients lives are impacted. How do you balance the speed and velocity and experimentation with flowing that through to healthcare systems and everything else that Epic is known for? Well, I I think there's sort of multiple different layers here.

26:11 first off, you need to maintain I mean, first do no harm, right, is is key here. And that starts all the way in regards to your development pipelines within the organization. Understanding how to maintain and improve the level of quality of the code going out the door. and understanding what that means as the tools evolve. There's a variety of new capabilities with skills and MCP servers, etc. that allow you to even do additional checks than you'd done in the past to help with that. even while the velocity of what you're releasing continues to increase, it means as you're releasing functionality that's dependent on it, you sort of have a couple of different layers. the first layer at the health system organization is key visibility into understanding both and evaluating functionality before it's put into production context, understanding its profile and then being able to monitor it in that same framework ongoing after it's released. It's interesting. I I tend to we have a capability as an example for physicians to and nurses to be able to quickly quickly respond to messages from patients.

27:29 And I tend to think and we I've tal we've talked about this with leaders across the health systems. You know, if you have an individual that is 100% of the time using those drafts without any edit, that's incorrect. And if you have an individual that 0% of the time is using any of them, that's also incorrect. And how do you how do you have the tooling to be able to monitor that? So that's the the ne the layer at the health system. But then importantly, the user experience needs to continue to adapt and help acclimate folks across healthcare to what is continuing to come. And that's where a number of us spend a lot of time at the company is helping think about how product design evolves through these different stages as both the models and the agentic capabilities continue to improve and help users be able to acclimate to that.

28:19 >> For the health systems themselves and the nurses and physicians and administrators, do you think they're going to change how they operate or will it just be a qualitative change in what they're doing today? >> They are changing how they operate. I think I think both of these will happen. I think the what we're seeing right now, the way the the framing I often tend to think about from a user perspective is we've gone through a few years here where AI just like it has been candidly inside of Epic is an assisted technology.

28:53 >> Yeah, >> there are pieces of the workflow. for example, responding to a message from a patient or preparing a letter in regards to a charge that was say denied. That helps. You can prepare that draft. You can suggest a likely code for a visit as an example or you can help patients have a discrete list of things that they might need to follow up on after their appointment and remind them about it.

29:26 But those are assistive tasks. What's happening is those are starting to compound together into an automated set of workflows. And I think as we go from these sort of assistive tasks into more automation, we can talk about where automation makes sense in healthcare and where it doesn't because it's not clear. It should not take place in certain contexts. we will see operational changes happen I think more and more extensively on this latter half.

29:57 >> Amazing. What about from a people perspective in terms of what are the skills that are going to matter the most as a lot of this technology evolves knowing that everything is still changing? Is it deep technical specialization? Is it broad generalist ability? Is it almost like design or management? Is it a combination of these? Do you have a sense of what that looks like even internally so far? The thing that immediately comes to mind is openness to adapting.

30:24 >> where we have tried to guess this, we get it wrong. Folks that are exploring and quickly adapting to what the technology provides and makes possible are those that are sort of moving ahead in leaps and bounds. And we're able to then take that and share it with the the broader group. I do think that and this gets back to kind of where we started the conversation in a way increasingly a deeper understanding of workflow and kind of end to end what is trying to be accomplished is important.

31:04 I was listening to a podcast actually as real Klein interviewing Jack from Anthropic this morning. and they were highlighting the importance of being able to give Claude code kind of a a full understanding of what it needs to be developed. Not just ask it a paragraph, but interview me about this design in many of these very complicated organizations. I mean, when you're thinking about a medical record moving from the ED up to the floor to the back office and then to the patient taking care of themselves as they go home with follow-ups, right? There's a over a hundred handoffs between individuals just in that type of care.

31:49 You the individual that understands that and has spent time immersed in it is going to be able to guide these tools to create the next generation of software and capabilities. and importantly help folks understand how to operationalize and adopt it within their organizations. What about more junior people? Do you think that this is something that's going to accelerate how they can get up to speed in an organization and help them grow and adapt faster? Is it going to be more beneficial to more senior people that have that expertise and knowledge? This is an impossible question. I want to acknowledge it up front, too.

32:28 >> I don't think that tenure directly correlates to success. I think often times there's a correlation between tenure and where they need to apply energy and time and focus. At that point, a relatively new employee to the company and relatively new employee to the the workforce at large, was chatting with one of our our leads recently about how as we go forward, we're going to make sure that developers get that workflow experience, that immersion experience, because he's seeing that be one of the core skills he needs going forward. It's not just the technical pieces. And I think identifying where folks are at in this trade-off between workflow understanding to guide agents and technical expertise to be able to critique their outputs and and what's ultimately going out the door. You need to figure out where you're at on that spectrum. And different people are at different paces.

33:43 >> I think we're seeing the exact same thing. There was we do onboarding for new hires and especially on my team. We need people to understand a lot of the tech and the tech changes so fast that we need to adapt in so many ways and last week we had a lot of people going through onboarding and a technical base camp or boot camp >> and we realized oh no like we haven't really done this specific session before and what if some people don't get started at the same time? If you have people following along building things, what if someone gets stuck at the very beginning and can't actually get started? What if someone sprints to the end or sprints to the middle? Usually, you kind of just accept that and hope that people are figuring things out and if they get too far behind, they'll follow up and do this at night or on the next day or something else. And that was my response of like, let's have three tracks. Let's figure this. One of our most junior people, a woman who just graduated and is on the team, she looked at that and said, I've done some of these sessions. Let me just build a way to say like where people are step by step. Have them be able to like not even self-report, but just see where they are. be able to say this person is stuck. You don't need to raise your hand. You don't need to feel embarrassed, but it makes it easier for them for people who are further ahead to know and be able to just go and help that person. And this was not just like a spreadsheet. It wasn't a like raise your hand, see what's going on. She built a working application and shipped it by the time that afternoon before this thing was starting. And I my response was, "Yeah, this is a hard problem. good luck. Here's some recommendations." And she solved this.

35:09 It is unbelievable what you can do with a sense of autonomy. Maybe maybe this is a bit of a digression, but I I I do think folks like you and I that are a little more established have a bit of a handicap on some of this stuff. >> Yeah. >> and you know, we we have a mental model of how these tools are supposed to work, how we've used them in the past. You know, one of the most fascinating things to me, there was a guy on our floor, whole whole team and a couple of them, none none of these folks are technical. They're working on building out and working with customers on operationalizing particularly around access to care.

35:52 Making sure that as you're coming in for an appointment, you get scheduled with the right specialist right out of the gate. Those opportunities are available directly in my chart to schedule, etc. in being able to get more people access to care. And you know, historically when a team is going and doing these types of engagements, we would have spreadsheets that would document who's there, what are the outcomes, etc. They built a cloud project that they shared among each other >> to be able to have dialogues with it about around what was happening at each of their customers with those engagements. sort of a a CRM like tool, but it wasn't for CRM purposes. And now whenever I ask him what's up, he sends me a link to a conversation he had with Claude that answers the question. In fact, I kid you not, be yesterday afternoon as I was preparing for this conversation, I asked him, how would you describe where this has helped? And yet again, he sent me a link to Claude answering the question for how I should answer this question. and that is a team that had no technical experience.

37:03 and quickly built this out and is now using it to meaningfully help keep everybody informed where Claude essentially is functioning as part of the team in a way, not just as an assistant to one of the folks. >> This goes back to I think you agreed and if you disagreed tell me, but I'd asked if this is all accelerating and it feels like there has been a step change in some way. If you do agree and that there is sort of a qualitative shift in adoption and traction, do you think that's because of the models, the products around them, what you've built to make it easy to access things, something outside of that?

37:38 >> Well, I I there isn't a single thing. I first off, I agree with the acceleration. I actually I think one thing that's being experienced right now and maybe we'll say it's the last 8 to 12 weeks and this gets a little bit back to that assistive to auto automation that I was talking about. I one of the things I noticed both internally as well as with health systems if you go back 12 24 months was that AI was productively helping in segments of a person's day >> y >> or segments of a person's workflow and so it would save them time it would save a physician time in regards to completing their notes as an example based off the conversation in the exam room or it would save folks internally time in regards to finding information from some knowledge base. but it was narrow and the result of that was that folks still sort of spent the total amount of time they couldn't insert truly new work into that day and I think in the last 8 12 months it's probably 8 12 weeks excuse me >> all the time >> you started to see this combine those moments where the human had to jump into the middle have become fewer and fewer and so you have this contiguous block where the agents are able to do work and then a result that somebody can engage with and spend time on. And so that I think feels qualitatively different now.

39:15 it results in automation in the back office in the health system. It results in improved security review of a of a PR that's being checked in back in Verona at Epic. And so I think that that difference feels large, but it's actually an accumulation of these individual moments that have been building up over the last 24 to 36 months. >> Taking a step back, I did have a few questions.

39:45 One is what about for companies that are not as advanced as Epic, either in terms of skill sets, where they are today. What would you recommend for a company or a team that is just getting started today knowing maybe they should have been experimenting before but now is the next best time? >> I think the first thing to do and I'll I'll come back to a little bit of what we talked about but maybe dive into a little bit more. One leaders need to be spending time in it. if they're not that right now and not just at work but also personally I think you need to figure out how to integrate this in to different contexts but then I think the second one is really look at your governance processes and where you're setting up challenging roadblocks internally that you might not anticipate you know I was talking with a friend different company who was this is a few months ago trying to experiment with cloud code >> and I got a text over the weekend that said how is your company handling access to various tools and libraries that are typically locked down and you don't have access to and I thought oh I wonder what kind of intense strange thing he's up to and I said well what what is an example and he responded with Python and and it got me thinking you know and and you know his the IT team there is and and at at organizations is spending time when I talk to them thinking about what are the security risks of giving everybody access to Python. Will they make good decisions with what they develop? And I understand that those conversations need to be had. I'll offer an analogy. You you were talking about Excel.

41:25 >> I do love Excel >> 45 minutes ago. Imagine coming up to an organization and them saying, "We don't roll Excel out to our employees. We don't trust that they have enough understanding of math in stats to be able to use it to analyze data. So we don't provide it to them. I I think everybody would agree that that's a not going to be a recipe for success long term. There there may be skills you need to help build, familiarity you need to help build in regards to that and getting deeper, but at the same time access and exploration is important. and those would be the two things. Start with leadership, make sure you've got governance, and open the door to this type of exploration.

42:10 >> Amazing. >> What about the reverse? Are there any challenges that you hit that someone can learn from or avoid or at least know that they're likely to hit the same challenges? you likely do have employees at your organization that know they need to adapt, that want to learn and want to grow and are struggling to figure out how. Mhm. >> And I do think that this is an important part of workforce development and figuring out how to have your team leads, your HR staff familiar and ready to work through these questions as they start to arise, I think is going to be very important. because there's a lot of opportunities for impact and growth and I think working through those aspects is going to be key.

43:04 What do you think is going to be the biggest surprise six months from now, 12 months from now where you knew that something was going to be possible but discovered something unexpected? And in this case, it is a known unknown because if it's truly unknown, you obviously can't predict this, but what do you think is going to be the biggest shift that you don't guarantee is going to happen, but suspect it might and wouldn't be completely shocked by?

43:30 I have my own answer for this too and I promise I'll give an answer. >> My my immediate thought is that for many years we've understood that there are patterns both in medical data and patterns in biology, in chemistry that could do more around improving the information and the research available at the point of care. And in the same ways that we have seen scaling laws applied to natural language models like the claude models as we're doing research on building medical event models internally at Epic and as you look across others in the life sciences you see additional models starting to be developed that I think can work in coordination with these and I think we're going to see a rapid adoption of these at the point of care in the clinics. The opportunity that for every patient when they are in the exam room, the patient and the physician have the opportunity to deeply understand how other patients like them progressed and what different interventions might mean for what's coming next. I think this is truly within reach. It's unknown exactly when it will happen or I guess some folks would say if it would happen, but I think it's going to come far faster than folks are anticipating.

45:00 >> I think you're right. I also think that the biggest shift and I don't know how this is going to be perceived or what the diffusion of this will be will be a shift to proactivity in some way where right now a big piece for me in the step change that I felt was faster autocomplete. something to delegate and be pleasantly surprised by the results. But if your zero to 100% sort of like are you using nothing or everything if things are just happening in the background and then I can review and adopt them and I didn't prompt those or ask for those that I think is going to change how I do my job every day instead of needing to set up those workflows and I don't know how that's going to feel personally. I think this gets back to the workflow conversation we were having in a way. Right now >> in many of these cases, >> we are passing in context. We're figuring out how to collect that different information. Be that in the exam room, be that in preparing for an appointment with a patient or be that in a design document if I'm a developer back at Epic. And I do think we are on the verge of that context being essentially ambiently available >> to the models. And as that context becomes ambiently available and this agentic run is able to happen, you suddenly just have natural insights available to folks when they need them to be able to evaluate what to do next.

46:30 And I don't know what that's going to do dayto-day for people's Monday mornings is something that just going to be done and you're reviewing it. Are you going to have that happen every hour? That's a part where I don't know and we're going to be surprised by how this plays out. But I think that's right that ambient intelligence knowledge context. That's going to be the big shift that happens next. >> I think it gets back to that the importance of operational change.

46:50 >> Also the the adoption of all of this means that that fluidity and adaptability needs to be there. be that in regards to developing software academic or within the health system and how it changes. I think that's right. And I know we are at time. So, thank you, Seth. This has been a pleasure. Really enjoyed it. >> And thank you for coming out.