1 00:00:00,120 --> 00:00:04,570 2026 has been the year of agents from open claw at the beginning of the year to now 2 00:00:04,653 --> 00:00:08,338 platforms like muse and grokbot and instincts that are getting people 3 00:00:08,421 --> 00:00:12,270 to actually take advantage of these incredibly powerful autonomous tools 4 00:00:12,353 --> 00:00:16,327 that are getting increasingly large portions of their work done for them. 5 00:00:16,410 --> 00:00:20,528 We really have gone from agents being the next big thing to just being here. 6 00:00:20,612 --> 00:00:23,071 The problem is our work isn't just done alone. 7 00:00:23,155 --> 00:00:26,610 We tend to work in teams with other people. And yet up till now, 8 00:00:26,693 --> 00:00:31,088 most agents have been solo affairs, only covering the portion of our work that we 9 00:00:31,171 --> 00:00:33,520 do on our own. I think that is shifting now. 10 00:00:33,604 --> 00:00:37,750 A trend which I've talked about as multiplayer AI or shared or team agents. 11 00:00:38,070 --> 00:00:40,309 But what does it mean to even build a team agent? 12 00:00:40,392 --> 00:00:42,821 What are the types of considerations that go into it? 13 00:00:42,904 --> 00:00:46,234 And how different is it really than just building an agent for yourself? 14 00:00:46,317 --> 00:00:50,025 Those are the questions that I get into with Nufar Gaspar on this Operator's Cut 15 00:00:50,109 --> 00:00:53,722 edition of the AI Daily Brief. The AI Daily Brief is a daily podcast and video 16 00:00:53,805 --> 00:00:56,405 about the most important news and discussions in AI. 17 00:01:08,070 --> 00:01:12,069 All right, friends, quick announcements before we dive in. First of all, 18 00:01:12,153 --> 00:01:15,925 thank you to today's sponsors, KPMG, Blitzy, Harbor, and HyperAgent. 19 00:01:16,009 --> 00:01:20,235 To get an ad-free version of the show, go to patreon.com slash aidailybrief, 20 00:01:20,318 --> 00:01:22,530 or you can subscribe on Apple Podcasts. 21 00:01:22,620 --> 00:01:24,649 And to learn more about sponsoring the show, 22 00:01:24,732 --> 00:01:27,032 send us a note at sponsors at AIDailyBrief.ai. 23 00:01:27,660 --> 00:01:30,053 Just a couple other notes before we get in. Obviously, 24 00:01:30,136 --> 00:01:33,445 this is a pre-recorded episode. There are a bunch of things cooking today. 25 00:01:33,529 --> 00:01:36,884 We will have a lot to talk about, so we will be back with our normal format 26 00:01:36,967 --> 00:01:41,917 tomorrow. I also wanted to share a couple of upcoming opportunities. First of all, 27 00:01:42,000 --> 00:01:46,203 this Thursday, October 1st, we have a free live webinar all about building 28 00:01:46,286 --> 00:01:47,797 your personal AI benchmark. 29 00:01:47,880 --> 00:01:52,392 The whole idea is that when you get a new model like Opus 5-5 or Sonnet 5-5 30 00:01:52,476 --> 00:01:57,417 or Gemini 4 or whatever model comes next, this will help you put together your own 31 00:01:57,500 --> 00:02:02,135 standard benchmark to better understand where that model is going to fit into 32 00:02:02,218 --> 00:02:05,664 your own process. That is completely free, and if you register, 33 00:02:05,747 --> 00:02:08,910 you will get all the materials after, even if you can't attend. Again, 34 00:02:08,993 --> 00:02:11,017 that is coming up this Thursday, October 1st. 35 00:02:11,100 --> 00:02:14,238 Now, speaking of training, if you want to go a little bit deeper, 36 00:02:14,321 --> 00:02:18,252 the next cohort of our super intelligent executive AI and agent training programs 37 00:02:18,336 --> 00:02:21,969 is coming up. The executive agent leadership program is where you learn how 38 00:02:22,053 --> 00:02:24,101 to build AI agents for real business needs, 39 00:02:24,184 --> 00:02:28,034 as well as building a playbook to scale them safely across your organization. 40 00:02:28,380 --> 00:02:31,645 And if you feel you need a little bit more background before you get into that, 41 00:02:31,728 --> 00:02:34,078 you can also do the executive catch-up program. 42 00:02:34,170 --> 00:02:37,048 The next agent leadership cohort starts on October 5th, 43 00:02:37,131 --> 00:02:41,330 while the next executive catch-up program starts a week later on October 12th. 44 00:02:41,580 --> 00:02:45,730 All right, with all that out of the way, let's talk about how to build team agents. 45 00:02:47,220 --> 00:02:52,027 All right, Nufar, welcome back to the show. We got an operator's cut today. 46 00:02:52,110 --> 00:02:56,864 Yes, happy to be here again. This one has its genesis in some conversations we were 47 00:02:56,948 --> 00:03:01,469 having as we were coming up into the fall around what we wanted to do with the, 48 00:03:01,552 --> 00:03:05,690 you know, this fall's edition of a free self-directed training program. 49 00:03:22,110 --> 00:03:26,706 in between people's shared workspace. And this kind of just follows the natural way 50 00:03:26,789 --> 00:03:29,975 that people work. A lot of your work is done individually, 51 00:03:30,059 --> 00:03:34,102 but then lots and lots is also done at the intersection with other people, 52 00:03:34,186 --> 00:03:38,819 and that's where team agents can live. And as we were building out that course that's 53 00:03:38,902 --> 00:03:43,202 available right now at Multiplayer AI and just thinking about this concept more 54 00:03:43,285 --> 00:03:47,585 broadly, one of the things that we kept coming back to was that this is nascent 55 00:03:47,668 --> 00:03:51,857 enough that what it means to actually build a team agent won't necessarily be 56 00:03:51,941 --> 00:03:56,407 super obvious. And so the goal of today's Operator's Cut is to help actually think 57 00:03:56,490 --> 00:03:57,600 through how to build 58 00:04:01,800 --> 00:04:04,733 are, different from or I think probably what we'll argue here, 59 00:04:04,816 --> 00:04:08,597 similar to the types of agents that people might have already built and where they 60 00:04:08,680 --> 00:04:12,680 can go from here. So super excited to have you back and excited to dive in here. 61 00:04:12,810 --> 00:04:16,773 Amazing. So I'm going to broaden your definition and I'm going to call them team 62 00:04:16,857 --> 00:04:20,264 agents. And the concept is teams that your entire team can work with, 63 00:04:20,347 --> 00:04:24,411 whether it's because work happened between them or just because they are something 64 00:04:24,495 --> 00:04:25,810 that can be shared across. 65 00:04:26,070 --> 00:04:30,184 team members. And kind of the short version is that some agents should stay 66 00:04:30,267 --> 00:04:34,101 yours and private, while others should become the team's level agents. 67 00:04:34,185 --> 00:04:38,298 And the ones that do become the teams need a few decisions made on purpose. 68 00:04:38,382 --> 00:04:42,160 Some of them, as you said, overlap with any good agent configuration, 69 00:04:42,243 --> 00:04:45,629 and some of them are more unique or at least more intentional. 70 00:04:45,713 --> 00:04:50,218 And that's what we'll walk through. And I wanted to start as a means of motivation 71 00:04:50,302 --> 00:04:54,639 to give you like two stories that you will probably recognize from your company 72 00:04:54,723 --> 00:04:59,571 or your ecosystem. So the first is about a person that everybody that I work with, 73 00:04:59,654 --> 00:05:03,837 every company that I work with has at least one like that. And this person, 74 00:05:03,921 --> 00:05:08,104 they really know how their price and exception work or what the data is all 75 00:05:08,187 --> 00:05:12,143 about or how our biggest customer setup was configured three years ago. 76 00:05:12,226 --> 00:05:16,979 And when they're swamped, then work has to wait for them because they're the only one 77 00:05:17,062 --> 00:05:20,733 who knows. And when they're on vacation, someone still calls them. 78 00:05:20,817 --> 00:05:24,230 And when they leave, a piece of the company lives with them. 79 00:05:26,070 --> 00:05:30,203 work with. They had, I think, a person like that for each and every domain. 80 00:05:30,287 --> 00:05:34,251 So no one gets to take vacation without getting a call from their peers. 81 00:05:34,335 --> 00:05:36,613 And obviously, that's not a desired state. 82 00:05:36,696 --> 00:05:39,480 The second scenario is work that nobody fully owns. 83 00:05:39,564 --> 00:05:43,247 So a customer can move from sales to marketing to customer success. 84 00:05:43,331 --> 00:05:46,845 And then sales made them a promise during the deal conversation. 85 00:05:46,929 --> 00:05:50,781 And marketing is running a campaign with slightly different messaging. 86 00:05:50,864 --> 00:05:52,270 And then customer success 87 00:05:56,070 --> 00:06:00,313 the renewal and each team has their own piece and nobody has the whole picture 88 00:06:00,396 --> 00:06:03,252 because the work sits between them. So to your point, 89 00:06:03,335 --> 00:06:06,469 and even if they are using AI in each step of the process, 90 00:06:06,552 --> 00:06:10,906 these agents don't talk to each other and only worsen the problem in many cases. 91 00:06:10,989 --> 00:06:15,454 So an agent that is built for the whole team can help with both of these problems. 92 00:06:15,537 --> 00:06:18,310 And they are, of course, quite different problems. 93 00:06:18,630 --> 00:06:23,182 So there is another reason, I think, why this matters right now and why you should 94 00:06:23,266 --> 00:06:27,592 pay attention now, even if you feel a little bit like this is above your head. 95 00:06:27,675 --> 00:06:32,171 And that's the pattern that I think what we're starting to see across the most AI 96 00:06:32,254 --> 00:06:35,337 forward companies. And it goes in basically three steps. 97 00:06:35,420 --> 00:06:39,464 The step one is that everybody builds their own agents and they are happy 98 00:06:39,547 --> 00:06:41,887 with their productivity boost only with technology. 99 00:06:41,970 --> 00:06:45,735 Enough of those running around, we kind of get into an agent's poll. 100 00:06:45,819 --> 00:06:49,641 Lots of agents doing overlapping work, each maintained by one person, 101 00:06:49,724 --> 00:06:52,584 each with slightly different picture of the company. 102 00:06:52,667 --> 00:06:57,055 And each one stops being useful the day that the owner loses interest or leaves 103 00:06:57,139 --> 00:07:00,791 the company. And then what you see in the most AI forward company, 104 00:07:00,874 --> 00:07:04,610 they started to merge some of those agents into team level agents. 105 00:07:11,970 --> 00:07:16,032 people and ideally refined over time as the team learns what they should 106 00:07:16,115 --> 00:07:19,140 and shouldn't do. And this is happening very publicly. 107 00:07:19,224 --> 00:07:22,019 There are many companies already talking about it. 108 00:07:22,102 --> 00:07:26,567 I think you mentioned Avery's experience. They started by giving every employee 109 00:07:26,650 --> 00:07:31,230 an agent early in the year and then by May they have moved to shared team agents. 110 00:07:31,313 --> 00:07:34,511 Sierra merged many of their agents, specialists into one. 111 00:07:34,595 --> 00:07:38,970 Shopify is internal agents. So we see a lot of these in very public speaking 112 00:07:41,970 --> 00:07:44,233 see are probably either in step one or two, 113 00:07:44,317 --> 00:07:48,490 but I think that it's very important for all of us to look at these AI forward 114 00:07:48,574 --> 00:07:52,747 companies and understand how to get to number three and how to do it properly. 115 00:07:52,831 --> 00:07:57,168 And that's the entire purpose of today. So to make sure that we are talking about 116 00:07:57,251 --> 00:08:01,370 the same thing, because there are multiple names to basically the same thing. 117 00:08:01,454 --> 00:08:04,427 Some people, including yourself, call it multiplayer AI. 118 00:08:04,510 --> 00:08:07,730 You probably also heard shared agents. Some people refer to 119 00:08:11,970 --> 00:08:15,881 into the mix or interchangeably used to mean agents being used with shared 120 00:08:15,965 --> 00:08:20,146 knowledge across the company. I'm going to refer to them throughout the episode 121 00:08:20,229 --> 00:08:24,572 as team agents. And what I mean by that is we have one agent that many people talk 122 00:08:24,656 --> 00:08:27,973 to with shared knowledge, shared memory, and one configuration. 123 00:08:28,057 --> 00:08:32,130 The instructions, the skills, the access, and the owner, they are all shared. 124 00:08:32,213 --> 00:08:36,394 And you might be thinking when you hear me saying that we already share skills, 125 00:08:36,478 --> 00:08:40,335 right? Most companies have an amazing skill library or working on a skill 126 00:08:40,419 --> 00:08:41,887 library. And that's awesome. 127 00:08:41,970 --> 00:08:46,113 and that's great standardization of how you do the work in the company. 128 00:08:46,196 --> 00:08:50,695 But a skill is a playbook for a specific task where a team agent is something 129 00:08:50,779 --> 00:08:54,029 that your whole team works with on diverse set of tasks, 130 00:08:54,112 --> 00:08:58,016 ad hoc as well as repeated stuff. It does carry the team knowledge, 131 00:08:58,100 --> 00:09:02,480 remembers what it learns, and using the team level skills if you have them. 132 00:09:02,564 --> 00:09:06,409 Those, of course, can also tap into skills marketplaces and so on. 133 00:09:06,492 --> 00:09:09,563 So a skill library perhaps is one of the ingredients, 134 00:09:09,646 --> 00:09:11,670 but they are not one and the same. 135 00:09:11,970 --> 00:09:16,542 And I want you to today think about how and when to start building your next team 136 00:09:16,625 --> 00:09:21,197 agent. And one more note on scope, because there is a lot of excitement right now 137 00:09:21,280 --> 00:09:25,679 about all of the personal agents. I'm talking about Muse and Instinct and some 138 00:09:25,762 --> 00:09:29,874 of the other in this category. Those are for like a home or private life. 139 00:09:29,958 --> 00:09:34,644 Today, the focus is going to be on work. So that's one thing to make sure that it's 140 00:09:34,727 --> 00:09:39,210 clear about the scope. We're talking about agents that you build for your job. 141 00:09:41,970 --> 00:09:46,926 A new study from KPMG in the University of Texas at Austin found that when people 142 00:09:47,009 --> 00:09:50,783 work with AI, similar skills don't guarantee similar outcomes. 143 00:09:50,867 --> 00:09:55,449 Researchers studied more than 500 early career professionals and found that 144 00:09:55,533 --> 00:09:59,927 the best performers consistently amplified the value of AI by guiding a 145 00:10:00,010 --> 00:10:02,250 evaluating and refining its outputs. 146 00:10:02,500 --> 00:10:05,329 These top performers, called AI amplifiers, 147 00:10:05,412 --> 00:10:10,205 weren't defined by what they knew alone, but by how they worked with AI. 148 00:10:10,289 --> 00:10:15,353 Learn more about what separates AI amplifiers from everyone else at kpmg.com 149 00:10:15,436 --> 00:10:17,400 slash us slash AI amplifiers. 150 00:10:18,900 --> 00:10:22,170 Blitzy deeply understands your code base before it writes code. 151 00:10:22,253 --> 00:10:25,660 Here's the first place that pays off. Security in the age of AI. 152 00:10:26,140 --> 00:10:29,735 Vulnerabilities don't live in isolation. They live buried inside millions of lines 153 00:10:29,818 --> 00:10:33,189 of interconnected code, where patching one thing quietly breaks three others. 154 00:10:33,272 --> 00:10:36,553 That's why surface-level scans fail. Blitzy starts from its Knowledge Graph 155 00:10:36,636 --> 00:10:40,141 of your entire application, identifies and surfaces CVEs across the full estate, 156 00:10:40,225 --> 00:10:42,609 proactively recommends patches, and can execute the PR. 157 00:10:42,692 --> 00:10:46,018 Each fix is grounded in how your systems connect and validate so nothing new 158 00:10:46,101 --> 00:10:48,350 breaks, and the Knowledge Graph dynamically updates, 159 00:10:48,433 --> 00:10:51,107 keeping you ahead of an ever-accelerating threat landscape. 160 00:10:51,190 --> 00:10:55,439 One Blitzy customer resolved 21 active CDEs across 6 core microservices in 4 161 00:10:55,522 --> 00:10:58,574 days. Zero compile errors, every validation scan clean, 162 00:10:58,657 --> 00:11:01,450 months of planned work fixed in less than a week. 163 00:11:01,600 --> 00:11:05,497 Security remediation grounded in real architectural context at the speed 164 00:11:05,581 --> 00:11:08,096 of compute. Harden your codebase at blitzy.com. 165 00:11:08,179 --> 00:11:10,418 That's B-L-I-T-Z-Y dot com. Every episode, 166 00:11:10,501 --> 00:11:14,509 I talk about the competition between OpenAI, Anthropic, SpaceX AI, Google, 167 00:11:14,593 --> 00:11:18,822 and Meta. And if you've been listening for a while, you might have a favorite. 168 00:11:18,905 --> 00:11:21,697 Maybe you think OpenAI and Anthropic can stay ahead, 169 00:11:21,780 --> 00:11:24,600 or perhaps Meta's open source strategy can win out. 170 00:11:24,790 --> 00:11:28,760 Whatever your view, every AI lab creates a different investment opportunity. 171 00:11:28,843 --> 00:11:33,187 Harbor Capital Advisors AI Lab Ecosystem ETF Suite lets you invest in the ecosystem 172 00:11:33,270 --> 00:11:35,030 behind the AI lab you believe in. 173 00:11:35,260 --> 00:11:38,448 Search Harbor AI Lab ecosystem ETFs wherever you invest or follow at Harbor 174 00:11:38,531 --> 00:11:41,414 Capital on X to learn more. Visit harborcapital.com for a prospectus 175 00:11:41,497 --> 00:11:43,856 containing investment objectives, risks, fees, expenses, 176 00:11:43,939 --> 00:11:47,389 and other important information. Read and consider it carefully before investing. 177 00:11:47,472 --> 00:11:50,485 Risks include principal loss and artificial intelligence related risks. 178 00:11:50,569 --> 00:11:53,102 Harbor ETFs are distributed by Foresight Fund Services, LLC. 179 00:11:53,186 --> 00:11:56,461 Harbor is not affiliated with AI Daily Brief and the funds are not affiliated 180 00:11:56,544 --> 00:11:58,467 with, sponsored by, or endorsed by any AI lab. 181 00:11:58,550 --> 00:12:01,433 This is a paid advertisement and not personalized investment advice. 182 00:12:01,516 --> 00:12:04,616 Investing involves risk, including possible loss of principal. 183 00:12:05,260 --> 00:12:08,129 This episode of the AI Daily Brief is brought to you by HyperAgent, 184 00:12:08,212 --> 00:12:11,262 where you run fleets of agents your team can manage together. 185 00:12:11,380 --> 00:12:15,193 Forget local agents and chat workflows waiting on your laptop to be prompted. 186 00:12:15,276 --> 00:12:17,672 Hyperagent deploys always-on agents in the cloud, 187 00:12:17,755 --> 00:12:20,505 doing real work across the tools your team already uses. 188 00:12:20,589 --> 00:12:23,440 Marketing agents turn competitor moves into landing pages. 189 00:12:23,523 --> 00:12:26,526 Sales agents enrich leads, draft emails, and updates the CRM. 190 00:12:26,610 --> 00:12:29,208 Ops agent chases the paperwork and tracks the budget. 191 00:12:29,291 --> 00:12:33,003 Every agent has access to shared context and follows your rules about scope 192 00:12:33,086 --> 00:12:36,292 and approvals. It's time you add agents that feel like teammates. 193 00:12:36,375 --> 00:12:37,640 Hire yours at Hyperagent. 194 00:12:41,380 --> 00:12:46,412 all right i want to make sure that we understand like who i build the episode 195 00:12:46,495 --> 00:12:51,727 for and i think it's built for everybody and not just the frontier professionals 196 00:12:51,810 --> 00:12:56,510 that are building the absolute cutting edge if you're about to build one 197 00:12:56,593 --> 00:13:01,758 of course pay attention because it will provide you or verify the full playbook 198 00:13:01,842 --> 00:13:06,160 it's also aimed at people who are not quite there yet because the 199 00:13:11,380 --> 00:13:14,068 Team agent just makes some of them even more critical. 200 00:13:14,151 --> 00:13:17,763 And even if you are working solo and you have a team of agents or you're 201 00:13:17,846 --> 00:13:21,303 contemplating building a team of agents, you have the same decisions. 202 00:13:21,387 --> 00:13:25,409 The other player in your ecosystems are probably not your peers because you work 203 00:13:25,492 --> 00:13:29,617 alone, but perhaps you're building it for your customers or you're building it for 204 00:13:29,700 --> 00:13:33,465 a future you to make sure that it's robust enough and representative enough 205 00:13:33,549 --> 00:13:37,500 of diverse set of work. So that's the motivation or who should pay attention. 206 00:13:41,380 --> 00:13:44,339 very clear is that not every agent should be shared. 207 00:13:44,422 --> 00:13:47,556 There is a dial here or a spectrum with three settings. 208 00:13:47,639 --> 00:13:52,411 We have a private agent that's yours for your work with your taste and your access. 209 00:13:52,495 --> 00:13:55,512 For example, my own social media agent stays private. 210 00:13:55,595 --> 00:14:00,367 I'm probably not going to be able to share it with anybody because I'm the only one 211 00:14:00,450 --> 00:14:03,701 that wants to share or write in social in a specific way. 212 00:14:03,785 --> 00:14:07,855 So nobody should ever sound like me. And then we have shared knowledge. 213 00:14:07,938 --> 00:14:08,771 Those can be 214 00:14:11,380 --> 00:14:15,762 meaning the team maintains one body of knowledge. For example, what we sell, 215 00:14:15,845 --> 00:14:19,757 how we work, what our words mean, often with a shared skill library, 216 00:14:19,840 --> 00:14:22,460 and everybody points their own agents at that. 217 00:14:22,543 --> 00:14:25,632 But this is where the skill library lives, by the way. 218 00:14:25,716 --> 00:14:29,686 And it's often the right answer, and it's the easiest place to start. 219 00:14:29,770 --> 00:14:34,210 Another concrete example, say every salesperson has a prospecting agent tuned 220 00:14:34,294 --> 00:14:36,854 to their own style and their own preferences. 221 00:14:36,937 --> 00:14:38,700 Keep those agents as they are, 222 00:14:41,380 --> 00:14:45,370 voice, but give them all the same well-maintained picture of the ideal 223 00:14:45,453 --> 00:14:47,872 customer and the messaging for the company. 224 00:14:47,955 --> 00:14:52,294 So that's a hybrid mode that some companies or some use cases should remain. 225 00:14:52,378 --> 00:14:56,950 And lastly, we do have the team agents, where we have one agent that many people 226 00:14:57,033 --> 00:14:59,801 work with that has its own job and its own owner, 227 00:14:59,884 --> 00:15:04,689 and agents can move along the dial. And if you have a scenario where your colleagues 228 00:15:04,772 --> 00:15:09,020 keep asking to borrow your private agent, that's probably a sign that you 229 00:15:11,380 --> 00:15:16,359 to consider sharing it with others. So the next question that I want to answer is, 230 00:15:16,442 --> 00:15:19,076 are all team agents from the same archetype, 231 00:15:19,159 --> 00:15:23,521 or do they all follow the same type of use cases? And the answer is not. 232 00:15:23,604 --> 00:15:28,274 Like across the teams that I work with, I can roughly categorize the existing 233 00:15:28,358 --> 00:15:32,966 or future built team agents into four kinds. And knowing which your kind is, 234 00:15:33,050 --> 00:15:37,412 it helps you not only identify use cases, but also refine the use cases. 235 00:15:37,495 --> 00:15:40,520 And also it can tell you what to pay attention to 236 00:15:41,380 --> 00:15:46,150 right. So the first type of team agent, I'm calling it the expert agent. 237 00:15:46,233 --> 00:15:50,194 It's the one person's know-how or one small team's know-how. 238 00:15:50,278 --> 00:15:52,247 It's available to everyone who 239 00:15:52,330 --> 00:15:56,506 Depends on it. You'll recognize it by the person who can take the vacation. 240 00:15:56,589 --> 00:15:59,970 You remember from the beginning, if you want another example, 241 00:16:00,053 --> 00:16:04,172 it can be the data agent that can answer any data question across multiple 242 00:16:04,255 --> 00:16:08,545 departments in the company or a pricing and deal desk agent that serves a lot 243 00:16:08,628 --> 00:16:11,838 of go to market organizations, compliance agent and so on. 244 00:16:11,922 --> 00:16:16,495 In order to get it right, the knowledge has to come from the experts that holds it 245 00:16:16,578 --> 00:16:17,430 in the company. 246 00:16:17,800 --> 00:16:22,618 They need to be interviewed and they need to you need to collect the answers they 247 00:16:22,701 --> 00:16:27,277 already gave in multiple forums, whether those are direct messaging or emails 248 00:16:27,360 --> 00:16:32,299 and other places. And they have to be they the experts have to be involved from day 249 00:16:32,382 --> 00:16:36,656 one, because for them, this is what finally makes the vacation possible. 250 00:16:36,739 --> 00:16:41,640 But also a lot of job insecurity. So tread carefully when working in this domain. 251 00:16:47,800 --> 00:16:50,808 similar recurring work with one shared way to do it. 252 00:16:50,891 --> 00:16:55,682 The way to recognize a use case that falls into this category is when three people 253 00:16:55,765 --> 00:16:58,773 have each built their own version of the same agent. 254 00:16:58,856 --> 00:17:02,280 You can think about a team research and meeting prep agent. 255 00:17:02,364 --> 00:17:05,966 That's a very classical one. Or marketing teams content agent. 256 00:17:06,049 --> 00:17:10,424 And I'm sure you can think of others. In order to get this archetype right, 257 00:17:10,507 --> 00:17:15,061 you have to agree on how work is done. And it's easier to say than to actually 258 00:17:15,144 --> 00:17:15,977 execute 259 00:17:17,800 --> 00:17:22,096 three versions. And that's really a conversation about what you agree in terms 260 00:17:22,179 --> 00:17:26,644 of the standard for the company. So an interesting conversation at the very least 261 00:17:26,727 --> 00:17:29,844 once you start contemplating unifying an agent like that. 262 00:17:29,927 --> 00:17:34,392 And then we have the bridge agent. That's the work that flows between roles where 263 00:17:34,475 --> 00:17:39,052 nobody can do it alone. You'll recognize it when the handoffs break and every stage 264 00:17:39,135 --> 00:17:43,431 has to explain the context or the agents have to somehow work together between 265 00:17:43,515 --> 00:17:44,348 different 266 00:17:47,800 --> 00:17:51,705 solution, customer success and delivery. That's a very classical one. 267 00:17:51,789 --> 00:17:54,653 And to get it right, we have to have each function, 268 00:17:54,737 --> 00:17:58,295 their own piece of knowledge that is being fed into this agent. 269 00:17:58,378 --> 00:18:01,937 And people with different access will be eventually using that. 270 00:18:02,020 --> 00:18:06,619 So we have to also pay attention very, very carefully to permissions and you have 271 00:18:06,702 --> 00:18:10,434 to work hard through that. And lastly, we have the chief of staff. 272 00:18:10,518 --> 00:18:13,035 That's the agent that own the team operating, 273 00:18:13,119 --> 00:18:15,720 like operationalizing of the day-to-day work. 274 00:18:17,800 --> 00:18:21,907 status onboarding and you will recognize this one when the team keeps repeating 275 00:18:21,991 --> 00:18:26,363 itself and new joiners take weeks to find their footing and in order to get this one 276 00:18:26,446 --> 00:18:30,607 right what you need to do you need to clearly define what it is allowed to learn 277 00:18:30,690 --> 00:18:34,320 how can it learn their processes and the ongoing and how does it do so 278 00:18:34,403 --> 00:18:38,457 automatically which is not very trivial and then there are also of course many 279 00:18:38,541 --> 00:18:42,436 questions around permissions and so on so while you're thinking about these 280 00:18:42,519 --> 00:18:43,580 archetypes and which 281 00:18:47,800 --> 00:18:52,346 about. And before I give you the playbook on how to actually build these team 282 00:18:52,430 --> 00:18:56,676 agents, a quick detour, because I do want to give a quick reality check. 283 00:18:56,759 --> 00:19:01,606 There are some signs that a team agent is the wrong move or at least not the right 284 00:19:01,689 --> 00:19:05,935 move for you at this moment. So one indication where you shouldn't build 285 00:19:06,019 --> 00:19:09,423 a team agent, at least yet, is when taste beats standards. 286 00:19:09,506 --> 00:19:13,932 If different people truly need different answers or not willing to agree on 287 00:19:14,015 --> 00:19:16,180 a standard and their own judgment or 288 00:19:17,800 --> 00:19:22,448 those need to remain private agents so people can remain authentic and not have 289 00:19:22,531 --> 00:19:27,299 to fight about the ground truth. And the second indication not to build is nobody 290 00:19:27,382 --> 00:19:28,700 can own the knowledge. 291 00:19:29,080 --> 00:19:33,371 If the team cannot agree on how the work is done or nobody is willing to own 292 00:19:33,455 --> 00:19:35,904 and maintain the shared knowledge over time, 293 00:19:35,987 --> 00:19:39,185 the agent will drift within weeks, sometimes within days. 294 00:19:39,268 --> 00:19:42,927 So sort out the ownership before you go and build the team agent, 295 00:19:43,010 --> 00:19:45,632 because that's going to be a no go. And lastly, 296 00:19:45,715 --> 00:19:49,201 whenever you're realizing that trying to build a shared agent, 297 00:19:49,284 --> 00:19:53,921 a team agent only complicates more than it simplifies because you have conflicting 298 00:19:54,004 --> 00:19:56,940 needs or tangled permissions, endless coordination. 299 00:19:59,080 --> 00:20:02,100 work than what the agent can provide for you, 300 00:20:02,270 --> 00:20:06,325 That's the answer. Don't build it, at least not until you are able to untangle 301 00:20:06,408 --> 00:20:10,410 some of the complexities. And of course, notice what's missing from the list, 302 00:20:10,493 --> 00:20:14,707 sensitive data and high stakes. Those are design questions and they shape how you 303 00:20:14,791 --> 00:20:16,936 build it, which is where we're going next. 304 00:20:17,019 --> 00:20:20,755 So I don't think that when data is overly sensitive is a reason against. 305 00:20:20,839 --> 00:20:23,673 It's just something that needs extra careful attention. 306 00:20:23,757 --> 00:20:27,652 In order to build the candidate use case that hopefully you've gone through 307 00:20:27,736 --> 00:20:30,070 the decision checklist that I shared before, 308 00:20:32,270 --> 00:20:36,336 design decisions for your agent. It goes to what it does, where it lives, 309 00:20:36,420 --> 00:20:39,576 what it knows, what it can touch, and how can you run it. 310 00:20:39,660 --> 00:20:43,214 These are the core questions. In order to make it more concrete, 311 00:20:43,298 --> 00:20:45,886 I'll use one example, the whole way throughout. 312 00:20:45,969 --> 00:20:50,149 So it's going to be a customer agent, and it's going to be the bridge kind, 313 00:20:50,233 --> 00:20:54,185 meaning one that holds everything the company knows about each customer 314 00:20:54,269 --> 00:20:58,676 and everything we've promised them. And it's probably going to be used by sales 315 00:20:58,759 --> 00:20:59,593 and 316 00:21:02,270 --> 00:21:07,300 on. they will be using that example agent to do various customer-related activities. 317 00:21:07,520 --> 00:21:11,266 So let's break down some of these decisions to make it actionable. 318 00:21:11,349 --> 00:21:14,804 So first, the decision that you have to make is what it does. 319 00:21:14,888 --> 00:21:19,155 A quick caveat here, there is a lot of scoping that is very similar for any 320 00:21:19,239 --> 00:21:23,796 serious agent at work. It needs to have a clear job and a definition of done and 321 00:21:23,880 --> 00:21:25,620 a list of what it does and do. 322 00:21:25,730 --> 00:21:30,269 I'm going to stick here to what changes when it's built for a team versus an agent 323 00:21:30,352 --> 00:21:34,835 that you build just for yourself. So the first decision is who it serves by role. 324 00:21:34,918 --> 00:21:38,386 For example, sales asks it different things than delivery does. 325 00:21:38,469 --> 00:21:42,614 So write down each role and what they'll come to the agent for to make sure 326 00:21:42,697 --> 00:21:44,670 that you're covering all the scope. 327 00:21:44,780 --> 00:21:49,119 And of course, you can aim for one broad area of work because I think the team 328 00:21:49,203 --> 00:21:51,897 agents should be quite capable agents. Otherwise, 329 00:21:51,981 --> 00:21:55,016 it's harder to justify their existence. In our example, 330 00:21:55,099 --> 00:21:59,495 I would expect that the customer agent will be able to prepare meeting answers. 331 00:21:59,579 --> 00:22:04,031 Where do we stand with this customer? We'll be able to flag promises that commit 332 00:22:04,114 --> 00:22:08,397 other teams and draft every handoff. For example, I have quite a broad scope. 333 00:22:08,480 --> 00:22:11,005 And what keeps it focused is the area of work. 334 00:22:11,089 --> 00:22:13,640 And that is primarily in our case, customers. 335 00:22:14,780 --> 00:22:17,690 Also to pay attention to the team don'ts list. 336 00:22:17,774 --> 00:22:21,335 This is the part people often tend to skip. For example, 337 00:22:21,418 --> 00:22:26,411 it never makes commitment on someone's behalf or it never settles disagreement 338 00:22:26,494 --> 00:22:31,617 between people. Those go to its owner. You can think of another such examples in 339 00:22:31,701 --> 00:22:36,498 your case. Also make sure that the don't list includes permissions and data 340 00:22:36,582 --> 00:22:41,965 handling stuff. So, for example, it never carries information from one private space 341 00:22:42,049 --> 00:22:43,220 into a shared one. 342 00:22:44,780 --> 00:22:49,534 customer in another customer space. And it doesn't speak for one person to another. 343 00:22:49,617 --> 00:22:53,613 And of course, like with any agent, ideally start with narrower scope, 344 00:22:53,696 --> 00:22:58,525 reading and drafting. Only when it earns sufficient trust and was validated enough, 345 00:22:58,609 --> 00:23:02,087 then you can increase the scope as you gain more and more confidence. 346 00:23:02,170 --> 00:23:05,235 So that's the first decision and we can move to the next one. 347 00:23:05,318 --> 00:23:07,816 The second question will be where the agent lives. 348 00:23:07,899 --> 00:23:12,080 And this is the question that gets asked most. So let's be a little bit concrete. 349 00:23:12,470 --> 00:23:15,854 Basically, if I'm trying to make it as simple as possible, 350 00:23:15,937 --> 00:23:20,398 there are roughly three ways to share an agent from the simplest to the most 351 00:23:20,481 --> 00:23:25,121 involved. The simplest method will just to create a shared folder with your own 352 00:23:25,204 --> 00:23:28,289 tools, like the tools that your company already owns. 353 00:23:28,372 --> 00:23:33,132 And then each person just point their own tool to the shared agent and the folder 354 00:23:33,215 --> 00:23:37,795 will include instructions and potentially how to store new information as part 355 00:23:37,878 --> 00:23:40,090 of the way the agent overall behaves. 356 00:23:42,470 --> 00:23:46,320 But as a stepping stone to building shared agents and team level agents, 357 00:23:46,404 --> 00:23:50,745 that can be a very good start, especially if all of your team members are already 358 00:23:50,829 --> 00:23:55,171 using similar agentic tools or agentic tools that can point to folders as sources 359 00:23:55,254 --> 00:23:58,612 for their information. So that's the first and simplest method. 360 00:23:58,696 --> 00:24:02,437 The second way that we can do that is using a vendor ready-made agent. 361 00:24:02,520 --> 00:24:04,841 And we're increasingly seeing more and more. 362 00:24:04,924 --> 00:24:09,102 And we believe that we will continuously see more and more in the coming weeks 363 00:24:09,185 --> 00:24:12,190 and months of the year. We'll talk more about it later. 364 00:24:12,470 --> 00:24:16,675 But the way this works is that the vendor hosts it and you configure it. 365 00:24:16,758 --> 00:24:20,367 And in this category, we have many very recently famous tools, 366 00:24:20,450 --> 00:24:24,000 including CloudTag. OpenAI has their ChatGPT workspace agent. 367 00:24:24,083 --> 00:24:27,871 Copilot has their own offering that can be used like that notion. 368 00:24:27,954 --> 00:24:32,337 And many others are already offering shared spaces with agents that you can 369 00:24:32,421 --> 00:24:36,864 work together on. And it's just a matter of you configuring their specifics. 370 00:24:36,947 --> 00:24:41,509 And lastly, and that's, of course, the most sophisticated is an agent that you 371 00:24:41,592 --> 00:24:45,926 host, That's something that you run. It can be, for example, 372 00:24:46,009 --> 00:24:50,973 an open source agent like OpenClaw or Hermes on your own servers or on a list 373 00:24:51,056 --> 00:24:56,347 cloud. Or some companies are even building their own custom harnesses specifically 374 00:24:56,431 --> 00:24:59,625 for these needs. So that's the most sophisticated, 375 00:24:59,708 --> 00:25:04,934 but obviously has the most technically demanding requirements as well as the most 376 00:25:05,017 --> 00:25:10,177 freedom to build around that. So that's the three broad strokes options of where 377 00:25:10,260 --> 00:25:15,651 these team agents can live. On top of the decisions of how to build or the tools, 378 00:25:15,734 --> 00:25:19,370 there are two additional questions that come with them. 379 00:25:19,730 --> 00:25:24,072 The first question is who can see each person's conversation with the agent. 380 00:25:24,155 --> 00:25:27,624 Maybe it's only the people who are conversing with the agent. 381 00:25:27,708 --> 00:25:31,584 Maybe it's the entire channel or everyone in the session. Over here, 382 00:25:31,667 --> 00:25:35,835 there is a lot of differences between the different tools. In some tools, 383 00:25:35,918 --> 00:25:39,503 everybody can read every chat. For example, in Claude in Slack, 384 00:25:39,586 --> 00:25:43,987 everyone in the channel sees what the agent does and what the agent converses 385 00:25:44,070 --> 00:25:46,050 with others and can also steer it. 386 00:25:49,730 --> 00:25:51,992 sometimes a design decision by the vendor. 387 00:25:52,075 --> 00:25:54,672 Sometimes it's something that you can configure. 388 00:25:54,756 --> 00:25:58,525 And the second question is what it learns stored and who can read it. 389 00:25:58,609 --> 00:26:01,541 So an ideal agent is not one that is obviously frozen, 390 00:26:01,624 --> 00:26:04,668 but one that has a lot of memory and learning on the go. 391 00:26:04,751 --> 00:26:08,800 And then the question, where is this learning being stored and how does it 392 00:26:08,883 --> 00:26:11,480 happen? Is it something that happens per person? 393 00:26:11,564 --> 00:26:14,942 And then the agent evolves just from its interaction with you? 394 00:26:15,026 --> 00:26:16,310 Or is it per channel or 395 00:26:19,730 --> 00:26:23,718 and the agent evolves with everybody in public. And of course, 396 00:26:23,802 --> 00:26:28,972 one thing never costs customers. So do check and choose and tell the team before 397 00:26:29,056 --> 00:26:33,832 the first real task, what's the status with the team agent that you built? 398 00:26:33,915 --> 00:26:37,790 Because this is where a lot of trust can be gained or lost. 399 00:26:38,150 --> 00:26:42,406 And my rule is to pick the simplest option that two people will actually use 400 00:26:42,489 --> 00:26:46,574 this week. And for our customer agent, that's probably a channel agent or 401 00:26:46,658 --> 00:26:49,886 a simple like a tool agent because four functions need it. 402 00:26:49,969 --> 00:26:54,397 And if I will make it overly complicated and people will need to understand how 403 00:26:54,480 --> 00:26:58,336 to connect to that versus just going into a Slack or a Teams channel, 404 00:26:58,420 --> 00:27:02,733 it's not going to work. So in our case, that's probably going to be the right 405 00:27:02,816 --> 00:27:07,392 solution. this decision, I want to move arguably to the most important decision. 406 00:27:07,476 --> 00:27:11,000 And that's what it knows. And I think if you've built any agent, 407 00:27:11,083 --> 00:27:15,058 the recipe will sound very familiar. What's different for a team is that 408 00:27:15,142 --> 00:27:18,159 this is the moment the team agrees on the ground truth. 409 00:27:18,242 --> 00:27:21,428 How the work actually gets done, which definitions we use, 410 00:27:21,511 --> 00:27:24,415 which versions of the pricing policy is the real one. 411 00:27:24,499 --> 00:27:27,290 And I think that that conversation is worth having, 412 00:27:27,373 --> 00:27:30,530 even if you never ship the agent, because in most teams, 413 00:27:33,530 --> 00:27:38,074 And it's also where team agents get harder because the moment knowledge is shared, 414 00:27:38,158 --> 00:27:42,081 then you have more contributors and more contradictions and more places 415 00:27:42,165 --> 00:27:46,371 for something important to fall through. So the process around the knowledge 416 00:27:46,454 --> 00:27:49,587 matters even more than it ever did in your private agent. 417 00:27:49,671 --> 00:27:52,691 So you need to pay careful attention here. And ideally, 418 00:27:52,775 --> 00:27:54,750 you should go to these four stages. 419 00:28:03,530 --> 00:28:08,176 where information already resides. And I want AI do a lot of the heavy lifting 420 00:28:08,260 --> 00:28:11,097 in terms of aggregating and collecting the data. 421 00:28:11,180 --> 00:28:14,955 So for our customer agents, what I would do is I'll make sure that sales 422 00:28:15,039 --> 00:28:19,380 and solutions and success and delivery, they all will contribute their own piece. 423 00:28:19,760 --> 00:28:23,874 And then I want you to refine. I want you to merge the information coming 424 00:28:23,957 --> 00:28:26,806 from different sources, surface the contradictions. 425 00:28:26,889 --> 00:28:31,348 I'm sure that you will find five versions of the truth and then date everything 426 00:28:31,431 --> 00:28:35,775 and keep out what should never be shared. Of course, that includes passwords, 427 00:28:35,859 --> 00:28:39,857 notes about people, one customer's details in another customer's space, 428 00:28:39,941 --> 00:28:42,272 and so on. And then we have to approve it. 429 00:28:42,355 --> 00:28:45,204 Each piece should be signed off by whoever owns it. 430 00:28:45,288 --> 00:28:47,760 And the agent's owner puts it all together. 431 00:28:49,760 --> 00:28:52,412 goes tail very quickly, so you have to maintain. 432 00:28:52,495 --> 00:28:56,970 You need to decide what the agent may add to its own memory based on its working 433 00:28:57,053 --> 00:29:01,414 experience and what person has to review first when it goes into the knowledge 434 00:29:01,497 --> 00:29:05,915 and the memory. And we want to make sure that one person's definition of what's 435 00:29:05,998 --> 00:29:09,360 last year quietly becomes everyone's without any agreement. 436 00:29:09,890 --> 00:29:14,150 And of course, put the upkeep on a schedule because you want the agent 437 00:29:14,233 --> 00:29:19,237 to propose updates regularly and a person needs to review and approve the updates. 438 00:29:19,320 --> 00:29:24,263 And this is really the place to be very, very diligent and disciplined because it 439 00:29:24,346 --> 00:29:28,916 can totally make or break your team agent if you haven't done a good enough 440 00:29:28,999 --> 00:29:33,507 and self-sustaining process around acquiring and maintaining and verifying 441 00:29:33,590 --> 00:29:36,010 the knowledge that the agent taps into. 442 00:29:39,890 --> 00:29:42,300 very quickly fix anything that goes wrong. 443 00:29:42,383 --> 00:29:46,753 It can create a lot of havoc in your company if your team agent is not well 444 00:29:46,836 --> 00:29:51,146 educated enough on what matters. The fourth decision is what it can touch. 445 00:29:51,229 --> 00:29:55,954 And this is where team agents differ from the private ones because your own agent 446 00:29:56,038 --> 00:29:59,917 acts as you. and a team agent acts for many people. 447 00:30:00,000 --> 00:30:04,420 Of course, there are many security 101 that you need to apply here. 448 00:30:04,590 --> 00:30:09,443 Those that apply to any agent definitely stick for the four rules that are specific 449 00:30:09,526 --> 00:30:13,785 to, like, I'm going to just stick to the rules that apply to team agents. 450 00:30:13,868 --> 00:30:17,591 The first thing that you have to decide is whose access it uses. 451 00:30:17,675 --> 00:30:22,468 And you have three options. You can use the access or to act as whoever is asking. 452 00:30:22,552 --> 00:30:26,572 So it only sees what the person that was asking the question can see. 453 00:30:26,656 --> 00:30:31,568 And that's probably the safest choice, but sometimes the most complicated to execute 454 00:30:31,651 --> 00:30:34,090 unless the vendor already did it for you. 455 00:30:34,590 --> 00:30:39,105 Probably the right one when people on the team have different access levels. 456 00:30:39,188 --> 00:30:43,764 The other option that you have is to have its own account set up with exactly 457 00:30:43,847 --> 00:30:48,605 the access the job needs. And that's right when the whole team works on the same 458 00:30:48,688 --> 00:30:52,175 shared material. And lastly, it can use one person's login. 459 00:30:52,258 --> 00:30:56,712 Only ever do that for read-only and non-sensitive material because everyone 460 00:30:56,796 --> 00:31:00,403 who talks to the agent effectively gets that person's access. 461 00:31:00,487 --> 00:31:03,270 So I would not recommend to go down that path. 462 00:31:04,590 --> 00:31:08,759 probably do for our customer agent is to act as the person asking because sales 463 00:31:08,842 --> 00:31:12,742 success and delivery see different things in the CRM and in other systems. 464 00:31:12,825 --> 00:31:16,510 So I don't want to have the agents responding to them with information 465 00:31:16,593 --> 00:31:20,816 that they shouldn't be able to see. So that was the first rule on what the agent 466 00:31:20,899 --> 00:31:24,140 can see. The second rule is to decide who can ask it. 467 00:31:24,330 --> 00:31:28,650 And when the agent has its own account, everyone who can talk to it can use 468 00:31:28,733 --> 00:31:33,464 the account. And if you put an agent with access to the pricing sheet in a channel 469 00:31:33,547 --> 00:31:37,280 of 40 people, a few contractors among them, then all of a sudden, 470 00:31:37,363 --> 00:31:39,770 all 40 can now get the pricing by asking. 471 00:31:40,170 --> 00:31:43,171 The agent knows more than some of the people who can reach it. 472 00:31:43,254 --> 00:31:46,104 So decide who can talk to it with the same care you give. 473 00:31:46,260 --> 00:31:50,679 to what it can see. Okay, so that's something that happens very regularly when 474 00:31:50,763 --> 00:31:55,124 people don't pay attention. I also want you to decide where the answer lands. 475 00:31:55,208 --> 00:31:58,992 This one runs the other way. The agent uses the asker's own access, 476 00:31:59,075 --> 00:32:01,878 and the asker has every right to ask the question. 477 00:32:01,962 --> 00:32:06,496 The problem is that the answer shows up in a shared space in front of people who 478 00:32:06,580 --> 00:32:09,729 don't. So this is something that is happening right now. 479 00:32:09,812 --> 00:32:13,680 If you will look at the documentation of Claudine Slack, it can use 480 00:32:16,260 --> 00:32:21,595 and after the person approves and the anthropic documentation currently notes 481 00:32:21,679 --> 00:32:24,736 that it doesn't consider who else is in the channel. 482 00:32:24,820 --> 00:32:28,973 So if I approve to use my connectors and fetch all the information that I am 483 00:32:29,056 --> 00:32:33,599 permitted to see, and now the information is thrown at the channel where others can 484 00:32:33,682 --> 00:32:37,947 see that, that's the reality currently with the existing cloud implementation. 485 00:32:38,030 --> 00:32:42,406 So if it's sensitive, the answer has to go to the person who is asking privately 486 00:32:42,489 --> 00:32:46,586 and not in a shared channel. And lastly, keep record of who asked for what. 487 00:32:46,669 --> 00:32:51,240 And that's an important logging because when an agent works under its own account, 488 00:32:52,200 --> 00:32:56,288 agent did it, right? And you want to know which person asked so you can backtrack 489 00:32:56,371 --> 00:33:00,098 and make sure that there are no unexpected behaviors. And everything else, 490 00:33:00,181 --> 00:33:03,754 like starting with the list access and having a person approve anything 491 00:33:03,837 --> 00:33:06,174 that can't be undone is the same for any agent. 492 00:33:06,257 --> 00:33:08,407 So I'm not giving you security one-on-one. 493 00:33:08,490 --> 00:33:13,313 Okay, lastly, last decision and the one that will make your team agent live beyond 494 00:33:13,396 --> 00:33:15,430 its first week or the first month. 495 00:33:15,720 --> 00:33:18,562 That's the full operating manual here. Of course, 496 00:33:18,645 --> 00:33:23,218 the multiplayer sprint has a much more comprehensive way of thinking about it, 497 00:33:23,301 --> 00:33:27,533 but These four points are what makes or break the agency in practice. 498 00:33:27,617 --> 00:33:31,905 So the first thing is one owner. Anyone on the team can handle the work, 499 00:33:31,988 --> 00:33:33,810 but I want to have one person. 500 00:33:33,990 --> 00:33:37,192 or a very small group of people who owns the priorities, 501 00:33:37,275 --> 00:33:41,298 maintain the agent and decide when two people ask for opposite things, 502 00:33:41,382 --> 00:33:44,173 how to evolve the agent knowledge or feature set. 503 00:33:44,256 --> 00:33:48,455 They also decide on standing instructions and so on. So that's one thing. 504 00:33:48,539 --> 00:33:52,738 The second thing that I want to mention is the clear rules of engagement. 505 00:33:52,821 --> 00:33:55,436 I want you to tell people how to work with it, 506 00:33:55,520 --> 00:33:58,252 what it does and doesn't do and what it can see, 507 00:33:58,336 --> 00:34:02,970 who can see their conversation with it and everything that we discussed so far. 508 00:34:03,990 --> 00:34:08,359 We want to have clear indications of how to correct the agent when it's wrong. 509 00:34:08,442 --> 00:34:12,926 And I think that people trust the agents that they're using and the team agents, 510 00:34:13,009 --> 00:34:16,921 the more they know what it's learning and what's the learning process. 511 00:34:17,005 --> 00:34:21,145 The next thing I want you to do is to put decisions where it can see them. 512 00:34:21,229 --> 00:34:25,230 So a team agent only knows. what's written down in a place that it can reach. 513 00:34:25,314 --> 00:34:29,232 And if your team decides things in private messages and hallway conversations, 514 00:34:29,315 --> 00:34:33,060 the agent will never hear about them. And part of running it smoothly is, 515 00:34:33,300 --> 00:34:37,587 is to have it be able to tap into what's happening in real time in the team and 516 00:34:37,671 --> 00:34:41,958 to make sure that the team decisions and the team ongoing day to days are being 517 00:34:42,042 --> 00:34:46,661 learned by the agent itself. And lastly, I want you to keep watching because you will 518 00:34:46,744 --> 00:34:51,032 probably start with a small, ideally, you should start with a small pilot group 519 00:34:51,115 --> 00:34:55,320 and keep a few test questions that you can rerun whenever something changes. 520 00:34:55,680 --> 00:34:59,880 But I also want you to just monitor because we know that things change very 521 00:34:59,964 --> 00:35:04,449 frequently. So it's not just about having a proper process for whenever you want 522 00:35:04,533 --> 00:35:08,333 to introduce a new model or a new tool or a new knowledge or changes 523 00:35:08,416 --> 00:35:12,788 in instructions, but also just to monitor that everything is working properly. 524 00:35:12,871 --> 00:35:17,357 And of course, in some cases, we would want to retire the team agent if it's not 525 00:35:17,440 --> 00:35:20,841 behaving properly as we expected. We covered a lot of ground. 526 00:35:20,924 --> 00:35:23,780 And if I need to pull it together before we close. 527 00:35:25,680 --> 00:35:29,978 you that team agents matter for everyone, even if it will take you a while until you 528 00:35:30,062 --> 00:35:33,943 will actually be building one, and that building them properly is what makes 529 00:35:34,026 --> 00:35:37,073 the difference. Of course, not every agent should be shared. 530 00:35:37,156 --> 00:35:41,037 A private agent or shared knowledge and skills with private agents or a team 531 00:35:41,120 --> 00:35:44,793 agent, these are all valid options and should be used where appropriate. 532 00:35:44,876 --> 00:35:48,705 We talked about the team agents that are coming in four kinds, the experts, 533 00:35:48,788 --> 00:35:51,240 the common work agent, the bridge and the chief 534 00:35:55,680 --> 00:35:57,480 a lot about how to get it right. 535 00:35:57,600 --> 00:36:00,603 And we also talked about three signs on when to wait, 536 00:36:00,686 --> 00:36:03,340 whether it's because taste beats the standards, 537 00:36:03,423 --> 00:36:07,533 whether because nobody can own the knowledge or when it doesn't simplify 538 00:36:07,616 --> 00:36:11,958 the work. And once you're building, we went over the five decisions in order 539 00:36:12,042 --> 00:36:14,695 of what it does, where it lives, what it knows, 540 00:36:14,779 --> 00:36:18,888 what it can touch and how do you run it. And if you take those with you, 541 00:36:18,971 --> 00:36:22,640 you have what you need in order to get the team agent properly. 542 00:36:27,600 --> 00:36:30,336 First of all, we have the multiplayer AI sprint, 543 00:36:30,419 --> 00:36:34,623 which is free and will walk you through a team activity of in four weeks, 544 00:36:34,707 --> 00:36:38,911 configuring everything that we discussed, some of them in greater detail. 545 00:36:38,994 --> 00:36:43,727 If you want to learn how to properly build seriously team agents and agent rosters 546 00:36:43,810 --> 00:36:46,311 and how to do that in the best possible way, 547 00:36:46,395 --> 00:36:50,951 we have another cohort of the executive agent leadership that starts on October 548 00:36:51,035 --> 00:36:54,500 5th and we'll be happy to see you with all of our builders. 549 00:36:57,600 --> 00:37:01,698 you go and build agents for teams and rosters of agents and so on, 550 00:37:01,781 --> 00:37:06,702 we also have the executive catch-up that helps you become best-in-class AI user 551 00:37:06,785 --> 00:37:11,980 before you go and build those agents. And lastly, everything is changing. Odds are 552 00:37:12,210 --> 00:37:14,837 that every week we will get a relevant release. 553 00:37:14,920 --> 00:37:18,757 And by the time you hear this, maybe already something was released. 554 00:37:18,841 --> 00:37:23,197 But I do think that everything that we cover today holds no matter what chips 555 00:37:23,280 --> 00:37:27,804 next, because When an agent is built properly and the decisions are made right, 556 00:37:27,887 --> 00:37:30,519 it's orthogonal to any specific tool or feature. 557 00:37:30,603 --> 00:37:34,932 It's the business decision and the team standardization that matters much more 558 00:37:35,015 --> 00:37:39,458 than the tools that will help make it better by design the more releases we will 559 00:37:39,541 --> 00:37:43,418 have. And if I need to make some predictions for the rest of the year, 560 00:37:43,501 --> 00:37:47,970 so I think that we will see more and more formalization of what we just covered 561 00:37:54,210 --> 00:37:58,490 permissions, ownership, and so on, as well as like we can always trust 562 00:37:58,573 --> 00:38:03,352 the practitioners to share many of their learnings in the public eye so we can 563 00:38:03,436 --> 00:38:08,526 learn from many of the other AI and like frontier individuals and companies and see 564 00:38:08,609 --> 00:38:10,230 how it's working for them. 565 00:38:10,470 --> 00:38:14,682 That's it. Awesome. Great stuff, Nufar. I have a few things that I want to lob out 566 00:38:14,765 --> 00:38:17,824 there, discussion style, just as we close out. First of all, 567 00:38:17,908 --> 00:38:21,752 I guess the question, you know, you gave four archetypes of different types 568 00:38:21,836 --> 00:38:23,250 of agents that you've seen. 569 00:38:23,400 --> 00:38:27,985 Do you see, are any of them more common starting places than others for teams 570 00:38:28,068 --> 00:38:30,713 that you've observed? I think that in theory, 571 00:38:30,796 --> 00:38:35,260 your definition of an agent that lives between individuals or between teams 572 00:38:35,343 --> 00:38:38,897 sounds the most attractive, but it's the hardest to execute. 573 00:38:38,980 --> 00:38:43,019 So I think that's actually the ones that will, from what I'm seeing, 574 00:38:43,103 --> 00:38:46,111 are not the first to go for. And I've seen various, 575 00:38:46,195 --> 00:38:51,204 very successful versions of the first one of the expert agents that help create more 576 00:38:51,287 --> 00:38:52,560 redundancy in a team. 577 00:38:53,400 --> 00:38:57,479 sense and relieve some of the burden on those bottlenecks within the company. 578 00:38:57,563 --> 00:39:00,290 So those have seen a ton of implementations already. 579 00:39:00,374 --> 00:39:03,264 And I think the more the tools make it more accessible, 580 00:39:03,347 --> 00:39:07,318 the easier those will be to build. So those are probably the lowest hanging 581 00:39:07,402 --> 00:39:11,140 fruits. That's funny because that's exactly where my head goes. 582 00:39:11,223 --> 00:39:15,090 I'm super attracted to the ones that I think are most difficult to build. 583 00:39:15,180 --> 00:39:19,806 I've also seen a lot of, you know, very simple implementation of that expert one, 584 00:39:19,890 --> 00:39:20,797 which, you know, 585 00:39:20,880 --> 00:39:26,045 we talked about for a long time without even identifying it as this sort of team 586 00:39:26,128 --> 00:39:31,147 agent is just basically the agentified team internal knowledge hub, you know. 587 00:39:31,230 --> 00:39:35,208 policies around whatever, like early dismissal, who knows? You know, 588 00:39:35,291 --> 00:39:39,568 that just the company database of information that you can access through 589 00:39:39,651 --> 00:39:44,525 the chatbot instead. That's been something that companies have had a ton of success 590 00:39:44,608 --> 00:39:48,586 with as just an early, easy, fast use case right from the beginning. 591 00:39:48,669 --> 00:39:51,811 Another question that I have for you is how, you know, 592 00:39:51,894 --> 00:39:56,409 I'm sure that a lot of folks are sitting there wondering how much they should 593 00:39:56,493 --> 00:39:59,927 invest in. in building these sorts of things. 594 00:40:00,010 --> 00:40:06,162 for GrokBot or Microsoft Copilot or one of these sort of core tools that they might 595 00:40:06,245 --> 00:40:09,257 be using, OpenAI, Anthropic, just drop 596 00:40:09,340 --> 00:40:12,377 the sort of native version of this. And there's clearly some indications 597 00:40:12,460 --> 00:40:16,017 that they're thinking in this way. I think Claude Tag being the best example so far, 598 00:40:16,100 --> 00:40:16,457 but... 599 00:40:16,540 --> 00:40:21,126 You know, is this one where the value of digging in at this stage is going to be so 600 00:40:21,209 --> 00:40:25,683 you understand the theory and the ways to customize when better tools come around 601 00:40:25,766 --> 00:40:28,860 in the future? Or how do you think about that tradeoff? 602 00:40:29,110 --> 00:40:33,483 I think the heavy lifting is always going to be the configuration and the knowledge 603 00:40:33,567 --> 00:40:39,329 curation. I would select the one tool that is adjacent the most to your existing tool 604 00:40:39,412 --> 00:40:40,187 ecosystem, 605 00:40:40,270 --> 00:40:45,859 and figure out how to implement all the rest and if like new better improved 606 00:40:45,942 --> 00:40:50,745 tools, come to play will be ready because you already agree on the ground truth, 607 00:40:50,828 --> 00:40:54,046 on the do's and don'ts of these agents, on the use cases. 608 00:40:54,129 --> 00:40:57,058 So even if you at first implement them very naively, 609 00:40:57,141 --> 00:40:59,400 that's going to have your future ready. 610 00:40:59,830 --> 00:41:04,395 As for going and building these on like a competing product or your own like team 611 00:41:04,478 --> 00:41:05,970 level harnesses and so on. 612 00:41:06,250 --> 00:41:10,897 If you have the chops and you can do that easily and you have the justification, 613 00:41:10,981 --> 00:41:15,155 you can. But I'm not sure that I would have spent my energy now on going 614 00:41:15,238 --> 00:41:19,058 and building my own version of Claude Tag or similar when we were, 615 00:41:19,141 --> 00:41:22,577 I think both of us agreed that all of these are coming and, 616 00:41:22,660 --> 00:41:25,491 will probably be made very accessible and very smart. 617 00:41:25,574 --> 00:41:29,724 And your moat is probably in everything that these companies cannot tap into. 618 00:41:29,807 --> 00:41:33,847 Awesome. Well, thanks as always for another great Operators Cut and excited 619 00:41:33,931 --> 00:41:36,020 to have you back soon. Thank you. Bye.