WEBVTT 1 00:00:00.240 --> 00:00:03.880 So think of them really as an extension of your team who can build and 2 00:00:03.880 --> 00:00:07.440 own a workflow end to end instead of just helping out. 3 00:00:08.320 --> 00:00:12.040 So founders aren't just using agents for conversation or Q and 4 00:00:12.040 --> 00:00:15.280 A. So they're running entire financial validation, 5 00:00:15.520 --> 00:00:19.160 multi agent support workflows, supply chain orchestration 6 00:00:19.160 --> 00:00:22.840 and content management without needing humans for every single 7 00:00:22.840 --> 00:00:26.640 trigger point. So for example, startups like Quanto 8 00:00:26.640 --> 00:00:30.370 use multi agent systems to process, validate and 9 00:00:30.370 --> 00:00:34.090 reconcile split payments in minutes, a job that 10 00:00:34.090 --> 00:00:37.730 used to take hours or even days. Also with agentic 11 00:00:37.730 --> 00:00:41.290 AI, when a startup lands that big customer or their user 12 00:00:41.290 --> 00:00:44.850 base grows overnight, they agents can scale to match 13 00:00:44.850 --> 00:00:48.530 demand without doubling headcount or losing efficiency. And 14 00:00:48.530 --> 00:00:51.770 that's really a new fundamental way to accelerate growth. 15 00:00:52.330 --> 00:00:55.880 Also, agents are equipped with feedback loops that improve 16 00:00:55.960 --> 00:00:59.480 themselves in production, learning from data, success and 17 00:00:59.480 --> 00:01:03.120 failure. So companies don't just need to manually 18 00:01:03.120 --> 00:01:04.840 update processes all the time. 19 00:01:10.520 --> 00:01:13.560 Welcome to startup Rad IO, 20 00:01:14.120 --> 00:01:17.640 your podcast and YouTube blog covering the German 21 00:01:17.720 --> 00:01:21.170 startup scene with news, interviews and 22 00:01:21.410 --> 00:01:22.690 live events. 23 00:01:25.010 --> 00:01:28.530 Hey guys, welcome back to our second piece in 24 00:01:28.530 --> 00:01:31.810 cooperation with aws. Together with Jennifer 25 00:01:32.210 --> 00:01:35.850 SAS is being rewritten by agents if you're building software 26 00:01:35.850 --> 00:01:39.690 today, your business model might not survive the next wave. Jennifer 27 00:01:39.690 --> 00:01:43.530 Kroon, senior Gen AI Specialist at AWS and former 28 00:01:43.530 --> 00:01:47.070 startup Country Manager and Account Manager, has 29 00:01:47.070 --> 00:01:50.870 years of experience in the startup segment, now helping 30 00:01:50.870 --> 00:01:54.550 founders to rethink SaaS from the ground up with 31 00:01:54.550 --> 00:01:58.150 agentic AI workflows. Today we'll explore how 32 00:01:58.150 --> 00:02:01.470 agents transform SaaS, what new monetization 33 00:02:01.630 --> 00:02:05.150 models will dominate, and why startups need to adapt 34 00:02:05.230 --> 00:02:08.790 fast. Jennifer Kruen is a 35 00:02:08.790 --> 00:02:12.190 Senior Specialist for Generative AI and Machine Learning at 36 00:02:12.190 --> 00:02:16.030 aws, which she drives adoption of Gen AI across Europe. 37 00:02:16.430 --> 00:02:19.470 She brings a unique mix of consulting, business development 38 00:02:20.430 --> 00:02:23.790 and startup leadership experience, having herself 39 00:02:23.870 --> 00:02:27.230 scaled as country Manager in an international 40 00:02:27.470 --> 00:02:31.150 environment and working with SaaS and B2C 41 00:02:31.150 --> 00:02:34.990 startups at AWS. This gives her firsthand 42 00:02:34.990 --> 00:02:38.030 understanding of the challenges founders face, from 43 00:02:38.030 --> 00:02:41.600 monetizing struggles to to customer adoption 44 00:02:41.600 --> 00:02:45.160 barriers. At aws, Jennifer helps startups and 45 00:02:45.160 --> 00:02:48.560 enterprises alike embed agentic AI 46 00:02:48.560 --> 00:02:52.160 workflows, prioritize high impact use cases and 47 00:02:52.400 --> 00:02:55.799 navigate change management. In this episode 48 00:02:55.799 --> 00:02:59.600 she'll share how SaaS is being disrupted by agents 49 00:02:59.760 --> 00:03:03.560 and what founders must know to build the next generation of 50 00:03:03.560 --> 00:03:07.330 business models. Jennifer, welcome back. Thank you 51 00:03:07.330 --> 00:03:10.930 Joe. Really enjoy being back on the show. It's my 52 00:03:10.930 --> 00:03:14.610 pleasure. Let's dive directly into context and 53 00:03:14.610 --> 00:03:18.050 trends here. Gartner lists agentic 54 00:03:18.050 --> 00:03:21.810 AI as a top strategic trend for 2025 and 55 00:03:21.810 --> 00:03:25.490 very likely much further down the road. From your perspective 56 00:03:25.490 --> 00:03:29.170 Working with startups why are agents more than just the hype? 57 00:03:30.130 --> 00:03:33.890 So first off, for me it's important to recognize that agentic AI 58 00:03:33.890 --> 00:03:37.570 is not just another round of automation. So traditional 59 00:03:37.650 --> 00:03:41.370 automation, as you might know, follows scripts and rules. So 60 00:03:41.370 --> 00:03:44.770 things get done, but only within tightly defined boundaries. 61 00:03:45.250 --> 00:03:49.050 Agents on the other hand, set goals, make independent 62 00:03:49.050 --> 00:03:52.690 decisions, adapt strategies, and also take action across 63 00:03:52.770 --> 00:03:56.210 complex multi step processes or with minimal human 64 00:03:56.210 --> 00:03:59.850 input. So think of them really as an extension of your team who 65 00:03:59.850 --> 00:04:03.490 can build and own a workflow end to end instead of just 66 00:04:03.900 --> 00:04:07.580 helping out. So founders aren't just using agents 67 00:04:07.580 --> 00:04:11.140 for conversation or Q and A. So they're running entire 68 00:04:11.140 --> 00:04:14.620 financial validation, multi agent support workflows, 69 00:04:14.780 --> 00:04:18.460 supply chain orchestration and content management without needing 70 00:04:18.460 --> 00:04:21.900 humans for every single trigger point. So for 71 00:04:21.900 --> 00:04:25.420 example, startups like Quanto use multi agent 72 00:04:25.420 --> 00:04:29.180 systems to process, validate and reconcile split 73 00:04:29.180 --> 00:04:32.940 payments in minutes, a job that used to take hours or even days. 74 00:04:33.500 --> 00:04:36.980 Also with AgentIQ AI, when a startup lands that big 75 00:04:36.980 --> 00:04:40.820 customer or their user base grows overnight, the AI agents 76 00:04:40.820 --> 00:04:44.340 can scale to match demand without doubling headcount or losing 77 00:04:44.340 --> 00:04:47.780 efficiency. And that's really a new fundamental way to 78 00:04:47.780 --> 00:04:51.500 accelerate growth. Also, agents are equipped with 79 00:04:51.500 --> 00:04:55.140 feedback loops that improve themselves in production, learning 80 00:04:55.140 --> 00:04:58.970 from data, success and failure to. So companies don't just 81 00:04:59.210 --> 00:05:02.810 need to manually update processes all the time. Also, 82 00:05:02.890 --> 00:05:06.090 agents don't just wait for instructions. So they spot issues, 83 00:05:06.250 --> 00:05:09.850 escalate, adapt campaigns, reorder inventory, 84 00:05:09.850 --> 00:05:13.609 or shift customer interactions on their own before problems really 85 00:05:13.770 --> 00:05:17.210 take a bigger impact. And also this 86 00:05:17.210 --> 00:05:20.010 aligns with what research shows. So 87 00:05:20.330 --> 00:05:23.610 organizations deploying agentic AI are starting to see already 88 00:05:23.930 --> 00:05:27.560 measurable business outcomes. So for instance, think about a jump in 89 00:05:27.640 --> 00:05:31.080 customer satisfaction, boosted employee productivity and 90 00:05:31.080 --> 00:05:34.760 ROI typically achieved within a short time frame. 91 00:05:35.160 --> 00:05:38.880 And gentic startups draw investor attention not just for the tech, 92 00:05:38.880 --> 00:05:42.560 but because autonomous systems deliver compounding value 93 00:05:42.560 --> 00:05:46.200 as the business scales. So this means improving margins, resilience 94 00:05:46.200 --> 00:05:49.920 and speed to market. Of course, in all of the 95 00:05:49.920 --> 00:05:53.670 light there's also shade shadow. So definitely there's 96 00:05:53.670 --> 00:05:57.430 security, governance and robust guardrails that are really 97 00:05:57.430 --> 00:06:01.190 a must have for going to production, especially when those agents 98 00:06:01.590 --> 00:06:05.430 act autonomously. So this is why. Also Gartner predicts that 99 00:06:05.430 --> 00:06:09.030 some agentic AI projects will fail early due to these hurdles. 100 00:06:15.030 --> 00:06:18.810 How do you explain the difference between classical 101 00:06:19.130 --> 00:06:22.410 SaaS automation and agentic 102 00:06:22.410 --> 00:06:25.890 workflows to founders who may not have a technical 103 00:06:25.890 --> 00:06:29.130 background? Yeah, I like really much the 104 00:06:29.530 --> 00:06:33.250 analogy of a nice recipe that you're cooking. So 105 00:06:33.250 --> 00:06:36.890 thinking Perhaps of traditional SaaS automation like a very 106 00:06:36.890 --> 00:06:40.490 smart but very strict recipe. So it operates on A 107 00:06:40.490 --> 00:06:44.090 simple if then logic. So if you use this ingredient, this will happen. 108 00:06:44.720 --> 00:06:48.560 So for instance, if a new customer signs up, then you send a 109 00:06:48.560 --> 00:06:52.200 welcome email. It's really fantastic for automating predictable 110 00:06:52.200 --> 00:06:56.040 repetitive tasks. So you set it up and it runs reliably in the 111 00:06:56.040 --> 00:06:59.680 background, but it can only do what exactly it was 112 00:06:59.680 --> 00:07:03.200 programmed to do. So whenever something unexpected happens, 113 00:07:03.280 --> 00:07:06.680 so for instance a new field pops up in the data or a step in 114 00:07:06.680 --> 00:07:10.160 the process is changing, the automation is breaking. 115 00:07:11.200 --> 00:07:14.160 The AgentIQ workflow hover is completely different. 116 00:07:14.800 --> 00:07:17.840 So instead of this rigid recipe that we just discussed, 117 00:07:18.560 --> 00:07:22.160 you can think of it as hiring a proactive goal oriented 118 00:07:22.160 --> 00:07:25.960 chef or employee. So you don't give this employee a step by 119 00:07:25.960 --> 00:07:29.720 step list, you give them a goal and then they figure out how 120 00:07:29.720 --> 00:07:33.480 to achieve it. So for instance, instead of a workflow 121 00:07:33.480 --> 00:07:37.210 that just says if a customer churns, send a survey, you might 122 00:07:37.210 --> 00:07:40.850 give the agent the goal. Retain this customer, and the 123 00:07:40.850 --> 00:07:44.650 agent can then autonomously reason, plan and take multiple 124 00:07:44.650 --> 00:07:48.490 actions to achieve this goal. So it can analyze for instance 125 00:07:48.490 --> 00:07:52.250 the customer's usage history to find a pattern. Check also 126 00:07:52.250 --> 00:07:55.250 for any open support tickets or recent interactions. 127 00:07:55.970 --> 00:07:59.770 Or also it could search the knowledge base for common issues that are related 128 00:07:59.770 --> 00:08:03.040 to their usage. It can also send and draft the 129 00:08:03.040 --> 00:08:06.360 personalized message that is offering a solution or a discount. 130 00:08:07.000 --> 00:08:09.960 And then it can send the message and monitor the response. 131 00:08:10.760 --> 00:08:14.440 And it's a huge really going away from executing predefined 132 00:08:14.440 --> 00:08:17.720 tasks to really proactively making decisions and taking 133 00:08:17.720 --> 00:08:21.400 actions. And here's another way to look at it. 134 00:08:21.400 --> 00:08:25.080 So especially for founders and business leaders. So 135 00:08:25.160 --> 00:08:28.800 traditional SaaS has also been a lot around great user 136 00:08:28.800 --> 00:08:32.440 interface and a suite of features that a human uses to get a job done. 137 00:08:33.120 --> 00:08:36.800 And the value is really in the tool itself, the dashboard, the button, the 138 00:08:36.800 --> 00:08:40.480 reporting chart. So we've really measured success 139 00:08:40.560 --> 00:08:44.400 by engagement metrics like daily active users or time spent in the 140 00:08:44.400 --> 00:08:47.920 app. Whereas Agentix SaaS flips this a lot over 141 00:08:48.000 --> 00:08:51.640 because the value is no longer about the ui, because the agent 142 00:08:51.640 --> 00:08:55.360 doesn't need a beautiful dashboard to work, it works behind the scenes. 143 00:08:55.360 --> 00:08:58.920 So talking to other services via APIs to get the job 144 00:08:58.920 --> 00:09:02.340 done. So the new metric for success is the outcome. 145 00:09:02.420 --> 00:09:06.100 So how well did the agent achieve its objective? Did it improve 146 00:09:06.100 --> 00:09:09.860 the sales conversion rate or did it reduce customer support 147 00:09:09.860 --> 00:09:13.540 ticket resolution time? So think about it like 148 00:09:13.540 --> 00:09:17.140 this. Your customers aren't really buying software to use a dashboard, 149 00:09:17.140 --> 00:09:20.780 they're buying it to solve a business problem. And that's where agents 150 00:09:20.780 --> 00:09:24.460 get us closer to that outcome based business model. So it's really a 151 00:09:24.460 --> 00:09:27.540 fundamental change from selling a tool to selling a solution. 152 00:09:31.120 --> 00:09:34.960 Selling a tool and selling a solution. I think that's 153 00:09:34.960 --> 00:09:38.680 a very important distinction. I may 154 00:09:38.680 --> 00:09:42.280 add to our audience that you shared a lot of content with 155 00:09:42.280 --> 00:09:46.080 us, including your AWS talks, where you contrast 156 00:09:46.240 --> 00:09:49.840 operational improvement versus product innovations. 157 00:09:50.400 --> 00:09:54.170 Where do agents fit best here? I would say 158 00:09:54.170 --> 00:09:57.690 that agents really fit into both the operation improvement and the product 159 00:09:57.690 --> 00:10:01.130 innovation. So they're really a unique type of AI that 160 00:10:01.130 --> 00:10:04.890 doesn't just create content but also takes action. So it 161 00:10:04.890 --> 00:10:08.650 makes them valuable both for the internal efficiency and 162 00:10:08.650 --> 00:10:12.250 also for creating new customer facing products. So on the 163 00:10:12.250 --> 00:10:16.090 operational side, agents act as proactive, goal driven 164 00:10:16.090 --> 00:10:19.530 virtual collaborators. So this is where they drive significant 165 00:10:19.690 --> 00:10:23.450 internal efficiency gains by automating complex multi step 166 00:10:23.450 --> 00:10:27.070 tasks that traditional automation couldn't handle. So 167 00:10:27.070 --> 00:10:30.830 let's take one example that is very popular also in the SaaS context. So it 168 00:10:30.830 --> 00:10:34.550 could be onboarding where an agent would autonomously create a 169 00:10:34.550 --> 00:10:38.270 series of action, drafting a welcome email, setting up access, 170 00:10:38.750 --> 00:10:42.390 scheduling introductory meetings, sending follow 171 00:10:42.390 --> 00:10:46.230 up reminders. So this really improves operational efficiency 172 00:10:46.230 --> 00:10:48.910 and reduces manual processes within the company. 173 00:10:50.040 --> 00:10:53.680 And on the product innovation side, agents are being used to 174 00:10:53.680 --> 00:10:57.080 create entirely new customer facing features that were 175 00:10:57.080 --> 00:11:00.880 previously impossible. They go really beyond being a simple 176 00:11:00.880 --> 00:11:04.600 chatbot that we've accustomed to see and it becomes a core part 177 00:11:04.600 --> 00:11:08.080 of the product. So for instance, think about a product pricing 178 00:11:08.080 --> 00:11:11.800 agent here. So instead of just providing a dashboard with data, 179 00:11:12.520 --> 00:11:15.880 this is an agentic feature that acts on behalf of the users 180 00:11:16.260 --> 00:11:19.940 to find and implement an optimal price. So it takes a complex 181 00:11:19.940 --> 00:11:23.580 goal here and coordinates with other agents for for instance a 182 00:11:23.580 --> 00:11:27.060 demand analysis, web scraping, margin calculation 183 00:11:27.380 --> 00:11:31.100 to make a decision and take an action such as updating the price 184 00:11:31.100 --> 00:11:34.780 in a database. And really agents enable a new level 185 00:11:34.780 --> 00:11:37.940 of personalized user experience and improved automation 186 00:11:38.340 --> 00:11:41.860 that can be sold for instance for customers as a core product feature. 187 00:11:46.730 --> 00:11:50.290 Let's go a little bit into what you've learned. Business 188 00:11:50.290 --> 00:11:54.050 models and use cases. You scaled a startup 189 00:11:54.050 --> 00:11:57.849 yourself as country manager from that lens, where 190 00:11:57.849 --> 00:12:01.690 do you see the biggest opportunities for 191 00:12:01.690 --> 00:12:04.970 agents in sas? So if I think about 192 00:12:05.370 --> 00:12:08.650 the time when I was in that startup as a country 193 00:12:08.810 --> 00:12:12.450 manager, so I can tell you that the 194 00:12:12.450 --> 00:12:16.250 opportunities for AI agents aren't really just about efficiency, 195 00:12:16.490 --> 00:12:20.290 they're about strategic survival and growth. So I was 196 00:12:20.290 --> 00:12:23.930 working here with Back Market, that is a French refurbished 197 00:12:24.090 --> 00:12:27.730 marketplace. And in here I was one of the first 198 00:12:27.730 --> 00:12:31.290 employees where I was focused on scaling the business in Germany 199 00:12:31.450 --> 00:12:34.250 and ensuring that profitability 200 00:12:34.970 --> 00:12:38.690 leading really to triple digit growth in just six months. And it 201 00:12:38.690 --> 00:12:42.530 was of Course as you can imagine a lot of work and I 202 00:12:42.530 --> 00:12:45.930 saw really firsthand how critical it is to get ahead of the curve. 203 00:12:46.490 --> 00:12:49.850 And if I had agents at my side I would have been a lot 204 00:12:49.850 --> 00:12:53.290 quicker because I was just on my own at the time to scale the market. 205 00:12:54.089 --> 00:12:57.610 So agents really represent a profound shift from traditional 206 00:12:57.610 --> 00:13:01.210 software and give really companies a huge competitive advantage. 207 00:13:01.610 --> 00:13:05.200 So perhaps going back on where I would have seen 208 00:13:05.200 --> 00:13:09.040 this accompany my time in the startup. So 209 00:13:09.120 --> 00:13:12.880 as you know you're always at the beginning, especially you are 210 00:13:12.880 --> 00:13:16.520 fighting for every new customer. So agents can really be a 211 00:13:16.520 --> 00:13:19.840 game changer. So instead of a one size fits all approach, 212 00:13:20.320 --> 00:13:24.120 agents can really hyper personalize the entire customer acquisition 213 00:13:24.120 --> 00:13:27.680 funnel. So they can monitor market data, analyze 214 00:13:27.680 --> 00:13:31.380 competitor moves and then draft tailored outreach emails 215 00:13:31.380 --> 00:13:35.140 for different customer segments. So if you think about a predictive analytics 216 00:13:35.140 --> 00:13:38.980 agents, it could analyze a user behavior on your website and 217 00:13:38.980 --> 00:13:42.780 score their likelihood of converting, which then allows the sales 218 00:13:42.780 --> 00:13:46.380 team to focus on the highest potential leads first. So really 219 00:13:46.460 --> 00:13:50.060 about turning data into actionable intelligence at scale, 220 00:13:50.140 --> 00:13:53.580 which is of course very essential for a growing business. 221 00:13:54.460 --> 00:13:57.740 And if I think about my role at Backmarket here 222 00:13:58.260 --> 00:14:01.820 I was working with also the building a 223 00:14:01.820 --> 00:14:05.580 supply network and identify new partners. Because we 224 00:14:05.580 --> 00:14:09.420 were operating from France to the new market, I had to also work 225 00:14:09.420 --> 00:14:13.140 with existing relationships from there and provide them also 226 00:14:13.140 --> 00:14:16.420 with data driven feedback, improve also 227 00:14:16.740 --> 00:14:20.340 their sales and also further scale the market. So 228 00:14:20.340 --> 00:14:24.060 I could imagine myself automating a significant portion of that work 229 00:14:24.060 --> 00:14:27.770 now. So, so the agents could monitor for instance the key 230 00:14:27.770 --> 00:14:31.410 account health, they could analyze also user generated 231 00:14:31.410 --> 00:14:35.010 feedback from the support tickets and social media and even 232 00:14:35.010 --> 00:14:38.730 also predict potential churn of a customer before it happens. So 233 00:14:38.730 --> 00:14:42.250 this frees up a lot of our time to focus on high touch 234 00:14:42.250 --> 00:14:46.010 strategic relationships rather than routine check ins. And 235 00:14:46.010 --> 00:14:49.210 it's not really about replacing that human element, but also 236 00:14:49.610 --> 00:14:53.450 making us then more productive and impactful so that really 237 00:14:53.610 --> 00:14:56.890 contributing also to revenue growth and customer retention. 238 00:14:57.930 --> 00:15:01.770 And lastly I think also the future as mentioned 239 00:15:02.090 --> 00:15:05.690 in SaaS is not just about kind of beautiful interfaces as 240 00:15:05.690 --> 00:15:09.210 well. AI agent don't need that. They interact directly with 241 00:15:09.210 --> 00:15:13.050 functions and data. So this means that the product itself can become 242 00:15:13.050 --> 00:15:16.410 agent first and even no UI in some cases 243 00:15:16.810 --> 00:15:20.090 with the agent serving as the primary interface for the user. 244 00:15:20.700 --> 00:15:24.220 So for instance a user could simply ask an agent to create a new campaign 245 00:15:24.700 --> 00:15:28.540 targeting our top 10 customers in Germany who haven't made a 246 00:15:28.540 --> 00:15:32.260 purchase in 90 days. And the agent could then execute the multi 247 00:15:32.260 --> 00:15:36.020 step workflow across various systems from the CRM to the 248 00:15:36.020 --> 00:15:39.660 marketing automation tool. And that doesn't only 249 00:15:39.660 --> 00:15:43.260 simplify the user experience, but also allows SaaS 250 00:15:43.260 --> 00:15:47.020 companies to focus on building a powerful interconnected back end rather 251 00:15:47.020 --> 00:15:48.710 than just a front front end UI. 252 00:15:51.190 --> 00:15:54.630 We were talking, I was wondering, should SaaS founders 253 00:15:54.790 --> 00:15:58.070 treat agents as a feature, as an 254 00:15:58.070 --> 00:16:00.790 enabler or completely new product category? 255 00:16:01.990 --> 00:16:05.750 The short answer here is it's not really. As always, a one size 256 00:16:05.750 --> 00:16:09.590 fits all solution. So let's start really with the most 257 00:16:09.670 --> 00:16:13.070 common and lowest risk approach. So integrating a 258 00:16:13.070 --> 00:16:16.760 Genai agent as a new product feature. So this is 259 00:16:16.760 --> 00:16:20.360 ideal for improving existing workflows or adding a clear 260 00:16:20.360 --> 00:16:23.440 value add without really completely 261 00:16:23.600 --> 00:16:27.240 overwhelming your business model. For example, if you are a 262 00:16:27.240 --> 00:16:30.960 CRM company, because I also work with software vendors, you could 263 00:16:31.120 --> 00:16:34.800 introduce an agent that automatically drafts follow up emails for your 264 00:16:34.800 --> 00:16:38.240 sales teams. This really provides direct, 265 00:16:38.240 --> 00:16:42.080 measurable improvement in productivity and can be a strong selling point. 266 00:16:42.350 --> 00:16:44.630 So when I was in a. Com manager I would have really loved that as 267 00:16:44.630 --> 00:16:48.310 well. So the business impact can be also 268 00:16:48.310 --> 00:16:52.070 very clearly stated because you have improved automation and 269 00:16:52.070 --> 00:16:55.830 also personalized user experience. But also you 270 00:16:55.830 --> 00:16:59.510 could always become more ambitious here. So for instance, if we think about an 271 00:16:59.510 --> 00:17:02.990 enabler agent that works to streamline entire 272 00:17:02.990 --> 00:17:06.510 processes, freeing up human workers for more strategic 273 00:17:06.510 --> 00:17:10.270 tasks. So for instance a data analyst agent 274 00:17:10.270 --> 00:17:14.040 that handles data cleanup and organization, or an 275 00:17:14.040 --> 00:17:17.880 agent that autonomously triages customer 276 00:17:17.880 --> 00:17:21.560 support tickets. So this really fundamentally changes how a 277 00:17:21.560 --> 00:17:25.280 customer team works with your software, diving 278 00:17:25.280 --> 00:17:28.560 also into a deeper level of operational 279 00:17:28.560 --> 00:17:32.400 improvement. And I would have also loved that my time in back market, for instance. 280 00:17:33.440 --> 00:17:36.880 Lastly, instead of building an agent that is living 281 00:17:36.880 --> 00:17:40.560 inside an existing product, the agent, and that's 282 00:17:40.560 --> 00:17:44.240 really the disruptive part that I see some customers of 283 00:17:44.240 --> 00:17:47.640 mine working on, is that the agent is the product. 284 00:17:48.120 --> 00:17:51.800 So this means really building a new standalone offering that 285 00:17:51.800 --> 00:17:55.600 is centered alongside the agent's capabilities. So here 286 00:17:55.600 --> 00:17:58.760 the agent isn't just an assistant, but it's really an 287 00:17:58.760 --> 00:18:02.520 autonomous system that takes a goal, for instance optimizing 288 00:18:02.520 --> 00:18:06.040 a product's price, and then reasons, plans and executes 289 00:18:06.040 --> 00:18:09.630 multiple actions to achieve it. So it might use other agents to 290 00:18:09.630 --> 00:18:13.230 analyze demand, scrape competitor websites and calculate 291 00:18:13.230 --> 00:18:16.310 profit margins before finally updating the price in the 292 00:18:16.310 --> 00:18:19.830 database. So as with all the topics around 293 00:18:19.830 --> 00:18:23.590 AI for getting started, focus on 294 00:18:23.590 --> 00:18:27.270 the business impact and feasibility of your use case. So if 295 00:18:27.270 --> 00:18:31.110 you're just starting, consider launching an agent as a feature to test the 296 00:18:31.110 --> 00:18:34.960 market and gather user feedback. But regardless of the 297 00:18:34.960 --> 00:18:38.640 path that you choose, remember that the agents work best when they can interact with 298 00:18:38.640 --> 00:18:41.320 the outside world. So really investing in a strong 299 00:18:41.640 --> 00:18:45.440 infrastructure foundation as well. And we have all the 300 00:18:45.440 --> 00:18:49.239 tools available for all different sites, kinds of builders to make 301 00:18:49.239 --> 00:18:53.000 this happen. You've mentioned already 302 00:18:53.000 --> 00:18:56.760 a few use cases, automatic follow up emails for 303 00:18:56.760 --> 00:19:00.200 sales, triage of support. 304 00:19:01.400 --> 00:19:05.040 What do you think are the most promising SaaS use cases for an 305 00:19:05.040 --> 00:19:08.880 agentic workflow? From what I observe right 306 00:19:08.880 --> 00:19:12.520 now, the most powerful agentic use cases 307 00:19:12.520 --> 00:19:16.280 in SAS today is in BI and data analysis. So 308 00:19:16.439 --> 00:19:20.200 instead of really just displaying the data, agents would actively 309 00:19:20.200 --> 00:19:23.920 transform it into actionable insights. So imagine for 310 00:19:23.920 --> 00:19:27.640 instance a multi agent system for product pricing optimization 311 00:19:28.450 --> 00:19:31.850 that we mentioned, so finding the best price. They can also monitor 312 00:19:31.850 --> 00:19:35.490 KPIs and detect anomalies in real time, so alerting you 313 00:19:35.490 --> 00:19:38.690 really to problems before they become a real issue. 314 00:19:39.090 --> 00:19:42.530 And also it is possible to generate reports in plain language 315 00:19:42.610 --> 00:19:46.090 which also makes complex data accessible to non technical 316 00:19:46.090 --> 00:19:49.690 teams. And agents can also go a step further. So 317 00:19:49.690 --> 00:19:53.130 automating data driven workflows. Think about inventory 318 00:19:53.130 --> 00:19:56.430 optimization or churn prevention by proactively 319 00:19:56.430 --> 00:19:59.870 identifying at risk accounts and recommending retention 320 00:19:59.870 --> 00:20:03.670 strategies. So for instance I'm working with a fintech startup that is focused 321 00:20:03.670 --> 00:20:07.230 on building AI to generate dashboards in natural language from customer 322 00:20:07.230 --> 00:20:10.990 transaction data. So their solution identifies anomalies 323 00:20:10.990 --> 00:20:14.390 and financial data and recently received also 324 00:20:14.710 --> 00:20:18.230 further approval for going beyond that for shaping the product 325 00:20:18.550 --> 00:20:22.350 roadmap. And if we think about marketing and 326 00:20:22.350 --> 00:20:26.170 sales, these are also really promising use cases as we've seen also previously 327 00:20:26.170 --> 00:20:29.690 with generative AI. So agents are moving beyond the 328 00:20:29.690 --> 00:20:33.130 simple content generation to orchestrating entire 329 00:20:33.130 --> 00:20:36.890 campaigns. So you can call those the digital strategist of a 330 00:20:36.890 --> 00:20:40.570 marketing team which handles a lot of the multi step processes 331 00:20:40.570 --> 00:20:44.250 with autonomy. So for instance think about a system that can create 332 00:20:44.250 --> 00:20:47.890 and send newsletters based on user behavior. This can mean 333 00:20:47.890 --> 00:20:51.470 really hyper personalized ways of creating content 334 00:20:51.470 --> 00:20:55.030 at scale, unique messages, even creating 335 00:20:55.110 --> 00:20:58.550 personalized videos for individual users based on their behavior. 336 00:20:58.790 --> 00:21:02.310 And this is really interesting for media and entertainment companies out there 337 00:21:02.790 --> 00:21:06.510 and also optimizing campaigns dynamically. So while 338 00:21:06.510 --> 00:21:10.350 really monitoring the performance and reallocating ad spend and 339 00:21:10.350 --> 00:21:13.830 real time for better roi, that's something that I used also at 340 00:21:15.030 --> 00:21:18.620 Amazon Advertising previously. That's also a great use 341 00:21:18.620 --> 00:21:22.300 case. So really automating that entire lead 342 00:21:22.300 --> 00:21:25.980 lifecycle from intent based qualification to 343 00:21:25.980 --> 00:21:29.660 automated sales handoff also saves your team actually 344 00:21:29.660 --> 00:21:33.140 countless hours on manual tasks for instance in the sales space. 345 00:21:33.780 --> 00:21:37.540 And yeah I'm also working on with an E commerce customer 346 00:21:37.540 --> 00:21:41.180 that is even using these ideas to dynamically 347 00:21:41.180 --> 00:21:44.570 generate content really disrupt the way that we are currently 348 00:21:44.970 --> 00:21:48.810 thinking about E commerce. And another great 349 00:21:48.810 --> 00:21:52.530 example that I really enjoy is also that we talked a little bit 350 00:21:52.530 --> 00:21:56.210 about the triage topic. But for instance imagine an 351 00:21:56.210 --> 00:21:59.449 agent with memory that can remember the past interaction 352 00:21:59.450 --> 00:22:02.890 preferences and issues across multiple sessions. 353 00:22:02.970 --> 00:22:06.770 So there's a couple of companies I work with that 354 00:22:06.770 --> 00:22:09.930 are also using this for a great 355 00:22:11.170 --> 00:22:14.530 impact. For instance auto resolving 8,000 out of 16,000 356 00:22:14.530 --> 00:22:18.370 customer tickets which then includes a 4% increase 357 00:22:18.370 --> 00:22:21.930 in net promoter score. And that means also that the 358 00:22:21.930 --> 00:22:25.690 AI solution processed five times more tickets than human agents in 359 00:22:25.690 --> 00:22:29.290 its first week and this success transformed their customer 360 00:22:29.290 --> 00:22:32.970 service so allowing the human agents to focus on high value 361 00:22:32.970 --> 00:22:36.340 interactions and the same ICOs with each R software 362 00:22:36.650 --> 00:22:40.410 that is looking into in product chatbots to 363 00:22:41.130 --> 00:22:44.410 also help their customers with you know, 364 00:22:44.650 --> 00:22:48.050 agentic capabilities thinking about tasks like 365 00:22:48.050 --> 00:22:51.810 summarizing CVs and analyzing employee feedback 366 00:22:51.810 --> 00:22:52.490 conversations. 367 00:22:58.890 --> 00:23:02.650 Guys, we will be back after short ad break 368 00:23:02.650 --> 00:23:06.420 talking about culture change management and an 369 00:23:06.420 --> 00:23:07.420 outlook here. 370 00:23:13.500 --> 00:23:16.060 Guys welcome back from our ad break 371 00:23:17.180 --> 00:23:20.460 being here for the Last quarter of 372 00:23:21.180 --> 00:23:24.860 2 very long and extensive interviews that Jennifer prepared 373 00:23:24.860 --> 00:23:28.580 very well. Thank you very much and let's dive straight 374 00:23:28.580 --> 00:23:32.310 in in your few what role do 375 00:23:32.310 --> 00:23:35.830 ecosystems and marketplaces play in 376 00:23:35.830 --> 00:23:38.790 scaling agent based SaaS Solutions? 377 00:23:40.630 --> 00:23:44.470 If I think about my experience with software companies for a long time, I've 378 00:23:44.470 --> 00:23:47.190 seen that they try to be a one stop shop building 379 00:23:48.070 --> 00:23:51.510 pretty much every single feature that customers could ever need. 380 00:23:52.070 --> 00:23:55.510 But as also agentic based solutions evolve, 381 00:23:55.750 --> 00:23:59.380 we are seeing also even for agents, that one small agent 382 00:23:59.380 --> 00:24:03.100 can't do it all. So manually managing these 383 00:24:03.100 --> 00:24:06.940 different agents from different vendors can lead to how I call it 384 00:24:06.940 --> 00:24:10.300 like agent coordination chaos where there's a 385 00:24:10.300 --> 00:24:13.300 conflicting logic and inconsistent customer experience. 386 00:24:14.180 --> 00:24:17.860 And the solution is to move towards a multi agent system 387 00:24:18.340 --> 00:24:21.940 which is kind of the technical heart of a marketplace. So 388 00:24:21.940 --> 00:24:25.540 imagine in that scenario that a supervisor agent 389 00:24:25.540 --> 00:24:29.280 acts like like a project manager, so he takes the complex 390 00:24:29.280 --> 00:24:32.360 problem and delegates subtasks to specialized agents. 391 00:24:33.000 --> 00:24:36.680 So in the pricing example we had earlier, supervisor 392 00:24:36.680 --> 00:24:40.320 agent might send tasks to web scraping agent together competitor 393 00:24:40.320 --> 00:24:44.120 data, another one to a demand analysis agent and the third 394 00:24:44.120 --> 00:24:47.560 one to a margin calculation agent. And this 395 00:24:47.560 --> 00:24:51.400 collaborative approach between the agent is really a natural way to build 396 00:24:51.400 --> 00:24:55.020 complex workflows just like in our usual teams. And 397 00:24:55.020 --> 00:24:58.180 marketplaces are really the logical extension of this. 398 00:24:58.820 --> 00:25:02.540 So this allows companies to easily discover and plug in 399 00:25:02.540 --> 00:25:06.340 these specialized agents. And this ecosystem 400 00:25:06.340 --> 00:25:10.140 model I believe leads us to the biggest business benefit 401 00:25:10.140 --> 00:25:13.700 which is specialization. So a marketplace 402 00:25:13.780 --> 00:25:17.460 allows everyone to focus on building one or two 403 00:25:17.540 --> 00:25:21.060 highly specialized, let's call them best in class agents 404 00:25:21.140 --> 00:25:24.870 who rather than trying to create a monolithic product that does 405 00:25:24.870 --> 00:25:27.550 everything. So this is really a 406 00:25:28.350 --> 00:25:32.030 fundamental shift that enables a more like 407 00:25:32.030 --> 00:25:35.710 how we Call it also best of breed approach. So take a retail 408 00:25:35.710 --> 00:25:39.550 company for instance. So instead of having a single 409 00:25:39.550 --> 00:25:43.310 company having to build agents for every possible e commerce task, from 410 00:25:43.310 --> 00:25:47.030 product recommendation to payment processing, you can source them from 411 00:25:47.030 --> 00:25:50.700 an ecosystem of specialized providers. So we're already 412 00:25:50.700 --> 00:25:54.020 seeing this in some cases with companies co developing 413 00:25:54.500 --> 00:25:58.300 repeatable AI solutions with partners that are addressing 414 00:25:58.300 --> 00:26:01.540 specific use cases like customer support automation. 415 00:26:02.100 --> 00:26:05.940 So this shows that partnerships and ecosystems are a really key 416 00:26:05.940 --> 00:26:09.780 way to scale successful solutions into repeatable patterns. 417 00:26:10.420 --> 00:26:13.780 And this shift also changes the business 418 00:26:13.780 --> 00:26:17.500 model, with the new standard being a usage based or value 419 00:26:17.500 --> 00:26:20.750 based model, which we discussed in an earlier episode, where 420 00:26:20.990 --> 00:26:24.590 customers pay for the outcomes the agents deliver, not just for access 421 00:26:24.590 --> 00:26:28.350 to the software, but as in all the time 422 00:26:28.590 --> 00:26:32.270 you always with a lot of agents working together, you also have to take 423 00:26:32.270 --> 00:26:35.470 into consideration governance and security challenges. 424 00:26:35.630 --> 00:26:39.310 Thinking, for instance, if a malicious tool could try to trick an agent 425 00:26:39.310 --> 00:26:42.590 into performing a harmful action. So this is a real 426 00:26:42.910 --> 00:26:46.760 risk that you need to think about early. And this is where 427 00:26:47.000 --> 00:26:50.840 the marketplace plays its most crucial role. So 428 00:26:51.080 --> 00:26:54.800 for an ecosystem to succeed, it must build on a foundation of 429 00:26:54.800 --> 00:26:58.520 trust. So this means having rigorous governance, security and 430 00:26:58.520 --> 00:27:02.280 compliance standards for all agents in the ecosystem so 431 00:27:02.280 --> 00:27:06.040 customers can confidently combine agents from different vendors without 432 00:27:06.280 --> 00:27:09.800 fear of misuse or data breaches. According to 433 00:27:09.960 --> 00:27:13.440 some discussions that I've been having is that securing 434 00:27:13.680 --> 00:27:17.360 these systems requires a new security approach that 435 00:27:17.360 --> 00:27:20.840 combines AI specific protection like guardrails with 436 00:27:20.840 --> 00:27:24.200 traditional application security controls and robust 437 00:27:24.200 --> 00:27:27.920 operational monitoring, which for instance is also a key part 438 00:27:27.920 --> 00:27:30.480 of agent core in aws. 439 00:27:33.840 --> 00:27:37.560 We've been talking about a lot about the agents, the 440 00:27:37.560 --> 00:27:41.380 marketplaces, the areas where they will be 441 00:27:41.540 --> 00:27:45.180 most able to help a startup. But 442 00:27:45.180 --> 00:27:48.740 let's talk a little bit about the people aspect, the culture and change 443 00:27:48.820 --> 00:27:52.420 management. A big theme in your AWS talks is 444 00:27:52.420 --> 00:27:56.100 creating bought in teams. What practical step 445 00:27:56.180 --> 00:27:59.420 can founders take to build an 446 00:27:59.420 --> 00:28:02.340 internal AI literacy and 447 00:28:02.980 --> 00:28:06.710 reduce the potential resistance? Yeah, 448 00:28:06.710 --> 00:28:10.510 if I think about that one, leaders for me have to really champion 449 00:28:10.510 --> 00:28:14.230 the change. So I see this a lot also in the enterprise customers 450 00:28:14.230 --> 00:28:17.910 I work with. So according also to a recent article, and 451 00:28:17.910 --> 00:28:21.750 also based on my experience, only a third of the companies actually have 452 00:28:21.750 --> 00:28:25.550 an AI strategy. So many of the companies I work with 453 00:28:25.550 --> 00:28:29.350 dive into AI out of a fear of missing out without 454 00:28:29.350 --> 00:28:32.960 really clarity on what problem they're trying to solve. So it really 455 00:28:32.960 --> 00:28:36.760 has to start with a clear purpose. So what will AI help us 456 00:28:36.760 --> 00:28:40.200 improve? Is it customer experience? Efficiency, innovation? 457 00:28:41.080 --> 00:28:44.880 And that clarity which is voiced from the top, for instance, gives the 458 00:28:44.880 --> 00:28:48.600 team the reason to care. And also what I Find very important 459 00:28:48.680 --> 00:28:52.360 is that you shouldn't treat everyone the same. So you should start 460 00:28:52.680 --> 00:28:56.400 conducting an assessment in your team. So where are the people 461 00:28:56.400 --> 00:29:00.170 currently? Are they a beginner, intermediate, advanced, and then 462 00:29:00.170 --> 00:29:03.690 you can tailor the training accordingly. So I just came out of actually a workshop 463 00:29:03.690 --> 00:29:07.410 with our internal team, so we tailor it to the different audiences over 464 00:29:07.410 --> 00:29:11.010 here. And this really role aligned approach 465 00:29:11.490 --> 00:29:15.250 ensures that you're hitting the people where they are and making 466 00:29:15.250 --> 00:29:19.010 sure that you deliver what they need. So also letting people get their 467 00:29:19.010 --> 00:29:22.690 hands dirty really with low stake experiments. So 468 00:29:23.010 --> 00:29:26.450 drafting internal mock ups, basic summarization 469 00:29:26.930 --> 00:29:30.530 and also letting people fail fast in a safe sandbox environment 470 00:29:30.610 --> 00:29:34.370 is really important because while 471 00:29:34.370 --> 00:29:38.050 also still going a step further experimentation is important. 472 00:29:38.850 --> 00:29:42.610 It also works best when you pair it with structured training. So 473 00:29:42.690 --> 00:29:46.330 making sure that you have short targeted modules that are tailored to 474 00:29:46.330 --> 00:29:49.690 different job functions. And I work also 475 00:29:49.690 --> 00:29:53.330 personally a lot with our AI champions and in the 476 00:29:53.330 --> 00:29:57.040 different teams. So these are really early adopters who love the 477 00:29:57.040 --> 00:30:00.560 tools and can show others how to apply them in context. 478 00:30:00.960 --> 00:30:04.520 So they create a lot of opportunities for peer learning, team 479 00:30:04.520 --> 00:30:08.040 sandbox projects. So we are doing that quite a lot here at 480 00:30:08.040 --> 00:30:11.520 AWS as well. They are trying out tools and 481 00:30:11.600 --> 00:30:15.040 they're also sharing back what they learned, but 482 00:30:15.120 --> 00:30:18.880 also pairing this with clear AI policies that 483 00:30:18.880 --> 00:30:22.560 also brings people clarity so they don't experiment in the dark 484 00:30:22.560 --> 00:30:26.390 or worry about violating guidelines. Because as I've seen 485 00:30:26.390 --> 00:30:29.990 it a lot of times is that people can resist 486 00:30:29.990 --> 00:30:33.790 AI because they feel it might replace their autonomy or they just 487 00:30:33.790 --> 00:30:37.310 don't trust it. So treating AI with room to adjust or 488 00:30:37.310 --> 00:30:39.990 override, seeing it really as a sparing partner 489 00:30:41.110 --> 00:30:44.910 and also sharing small but real wins in your teams, for 490 00:30:44.910 --> 00:30:48.510 instance, saying okay, I used for instance Agentic 491 00:30:48.510 --> 00:30:52.230 or another workflow to cut my reporting preparation 492 00:30:52.230 --> 00:30:56.060 time in half or, or guess what happened when our 493 00:30:56.060 --> 00:30:58.500 team tried this cool tool. 494 00:31:01.780 --> 00:31:05.620 What I see a lot of successful companies do is to celebrate 495 00:31:05.620 --> 00:31:09.220 these internal successes and showcase those broadly. 496 00:31:09.300 --> 00:31:13.100 So not really just as metrics, but really as proof points 497 00:31:13.100 --> 00:31:16.420 so you can get the flywheel spinning and get more people excited, 498 00:31:16.900 --> 00:31:20.530 making also these AI tools really accessible. So ensuring 499 00:31:20.530 --> 00:31:24.330 that employees know what's available and how to access it, that's 500 00:31:24.330 --> 00:31:27.930 also really key. So we have also an AWS Party Rock 501 00:31:27.930 --> 00:31:31.650 which has an easy way to create apps from 502 00:31:31.650 --> 00:31:35.410 scratch. You can also have any other tools. Making 503 00:31:35.410 --> 00:31:39.090 just people try this out for themselves is so much more 504 00:31:39.090 --> 00:31:42.610 worthwhile than just talking about it. And that really 505 00:31:42.770 --> 00:31:44.290 brings a lot of excitement. 506 00:31:48.540 --> 00:31:52.340 You often mentioned in like all the documentation you 507 00:31:52.340 --> 00:31:55.660 provided to us, you often mention AI 508 00:31:55.660 --> 00:31:59.500 Whisperers, what exactly do you mean 509 00:31:59.660 --> 00:32:03.420 and how can they accelerate the adoption 510 00:32:03.420 --> 00:32:05.180 inside, for example, SaaS companies. 511 00:32:07.020 --> 00:32:10.420 I would like to think about AI whispers really as a kind of 512 00:32:10.420 --> 00:32:14.180 translator or a bridge builder. So they are 513 00:32:14.180 --> 00:32:17.740 the person who is equally comfortable speaking the language of a data 514 00:32:17.740 --> 00:32:21.380 scientist, but also the language for instance of a customer service 515 00:32:21.460 --> 00:32:25.060 representative. So their primary skill is not just the 516 00:32:25.060 --> 00:32:28.900 technical expertise, but the ability to really understand how 517 00:32:28.900 --> 00:32:32.660 AI can solve real world business problems and then clearly 518 00:32:32.660 --> 00:32:36.180 communicate that to non technical teams. For instance, 519 00:32:36.900 --> 00:32:40.740 this is really important because about 88% of 520 00:32:40.740 --> 00:32:44.420 employees according to study of Gartner are non technical. 521 00:32:44.580 --> 00:32:48.300 So the Iwhisperers take the complex technical 522 00:32:48.300 --> 00:32:51.500 concepts and make them accessible. So it helps really to 523 00:32:51.500 --> 00:32:55.140 demystify the technology and shift the conversation 524 00:32:55.140 --> 00:32:58.980 from fear to opportunity. So they're really the human element that 525 00:32:58.980 --> 00:33:02.620 ensures that AI isn't just a project for the tech department, but 526 00:33:02.620 --> 00:33:06.430 a tool for everyone. And where I see this coming 527 00:33:06.430 --> 00:33:10.110 to life even more is that if you pair an AI 528 00:33:10.110 --> 00:33:13.670 whisperer with a senior leader, that is really strategic 529 00:33:13.670 --> 00:33:17.470 because it can really accelerate AI adoption across the entire 530 00:33:17.470 --> 00:33:21.150 organization. So imagine the leader that provides a 531 00:33:21.150 --> 00:33:24.870 top down vision, authority and resources that 532 00:33:24.870 --> 00:33:28.350 are needed for a large scale change and they can set the 533 00:33:28.350 --> 00:33:32.070 strategic direction and signal to the entire company that AI 534 00:33:32.070 --> 00:33:35.830 is a priority. However, the leaders might not have really the 535 00:33:35.830 --> 00:33:39.670 ground level insight into how AI can practically solve day to 536 00:33:39.670 --> 00:33:43.150 day problems. And that's where the AI whisperer comes in. 537 00:33:43.310 --> 00:33:47.030 So it provides the bottom up practical knowledge so they can 538 00:33:47.030 --> 00:33:50.430 work with different teams, identify specific pain points 539 00:33:50.590 --> 00:33:54.310 and also suggest high impact, low risk use cases that 540 00:33:54.310 --> 00:33:57.870 are perfect for a pilot. For instance. So think about reducing 541 00:33:57.870 --> 00:34:01.460 manual processes or, or like a lot of people or 542 00:34:01.460 --> 00:34:04.940 almost everyone is doing right now, speeding up software development. 543 00:34:05.980 --> 00:34:09.260 So when you combine this leader's strategic 544 00:34:09.260 --> 00:34:13.100 authority with the AI whisperer's practical underground 545 00:34:13.100 --> 00:34:16.540 knowledge, you really create a great momentum for change. 546 00:34:17.020 --> 00:34:20.460 So you have these great example proof points and 547 00:34:20.700 --> 00:34:24.380 the leader can then champion these successes from the top. And that really 548 00:34:24.819 --> 00:34:28.579 creates a powerful feedback loop that builds momentum and trust 549 00:34:28.819 --> 00:34:32.299 and also builds and bridges the gap between an 550 00:34:32.299 --> 00:34:34.819 AI vision and a real world implementation. 551 00:34:38.659 --> 00:34:42.419 I actually had to smile when I went through your material because you 552 00:34:42.419 --> 00:34:46.139 also talk about AI playgrounds. As a father of two, 553 00:34:46.139 --> 00:34:49.539 I do have a total different association with playgrounds. But 554 00:34:49.699 --> 00:34:52.800 those AI playgrounds, how do they 555 00:34:52.880 --> 00:34:56.600 encourage a safe experimentation and build 556 00:34:56.600 --> 00:35:00.200 culture of innovation? Actually I like a 557 00:35:00.200 --> 00:35:03.440 lot the analogy of the playground as you just said, from the 558 00:35:03.600 --> 00:35:07.200 children's perspective, because if we take ourselves 559 00:35:07.279 --> 00:35:10.960 back to that time, we really enjoyed to learn a lot of different 560 00:35:10.960 --> 00:35:14.600 things. So we didn't always think about a goal in mind. And 561 00:35:14.600 --> 00:35:18.000 that's also a mindset that we like to take also to 562 00:35:18.000 --> 00:35:21.620 AI because you are then experimenting 563 00:35:21.620 --> 00:35:25.380 in a sandbox. So really a risk free environment where you 564 00:35:25.380 --> 00:35:29.020 can get hands on, you're not afraid of breaking 565 00:35:29.020 --> 00:35:32.220 anything or really having any massive cost 566 00:35:32.300 --> 00:35:35.740 attached. It's really a low stakes space for iterative 567 00:35:36.060 --> 00:35:39.820 experimentation and thinking outside the box. Instead 568 00:35:39.820 --> 00:35:43.500 of just reading lots of things out there about what AI 569 00:35:43.500 --> 00:35:46.630 can do, you, you actually get to build and test things. 570 00:35:47.270 --> 00:35:50.230 So this is especially important for startups because 571 00:35:50.790 --> 00:35:54.630 AI of course isn't a magical solution for every program. So 572 00:35:55.110 --> 00:35:58.830 playgrounds really allow you to quickly test a new idea 573 00:35:58.830 --> 00:36:02.310 and see if it really has a business impact before you actually 574 00:36:02.310 --> 00:36:06.030 commit significant resources. So they're all about moving 575 00:36:06.030 --> 00:36:09.590 from a theoretical idea to a practical prototype 576 00:36:09.590 --> 00:36:13.050 quickly. And how do these playgrounds 577 00:36:13.050 --> 00:36:16.730 specifically help if you think about the startup context, 578 00:36:17.290 --> 00:36:20.810 for startups that are often short on time and money, 579 00:36:21.370 --> 00:36:24.770 AI playgrounds really provide a way to build quick 580 00:36:24.770 --> 00:36:28.250 prototypes, do a test drive of use cases 581 00:36:28.250 --> 00:36:32.090 without the need for full scale development team or complex 582 00:36:32.090 --> 00:36:35.530 infrastructure. As I mentioned, we have for instance a free 583 00:36:35.530 --> 00:36:39.290 platform that you can also try out just on your phone called plus Party Rock, 584 00:36:39.850 --> 00:36:43.490 which allows you to start with a prompt, remix an app or build 585 00:36:43.490 --> 00:36:46.890 from scratch so you can create simpler applications. 586 00:36:47.690 --> 00:36:51.090 So I've seen some colleagues using this for finding the perfect 587 00:36:51.090 --> 00:36:54.850 recipe for cooking or for more kind 588 00:36:54.850 --> 00:36:57.850 of use cases, planning a trip, finding 589 00:36:58.010 --> 00:37:01.570 apartments so there's a lot of room for 590 00:37:01.570 --> 00:37:05.290 imagination. And when the team then knows that 591 00:37:05.290 --> 00:37:09.010 they can't break the system, they're more likely to be creative and try 592 00:37:09.010 --> 00:37:12.500 new things. So you're not focused on getting the technology 593 00:37:12.660 --> 00:37:16.340 perfectly right from the start, but you are focused on what business 594 00:37:16.340 --> 00:37:19.860 value you can create and exploring different use cases. 595 00:37:20.740 --> 00:37:24.300 We had also here in our teams, reducing manual 596 00:37:24.300 --> 00:37:27.780 processes in territory, planning faster software development, 597 00:37:27.940 --> 00:37:31.740 or even creating personalized user experience for customers 598 00:37:31.740 --> 00:37:35.260 in the sales team for instance. That really allows a 599 00:37:35.260 --> 00:37:39.000 startup as well to quickly evaluate a use case based on 600 00:37:39.000 --> 00:37:42.600 what is the business impact and feasibility. And you can really see 601 00:37:42.760 --> 00:37:46.440 and test the idea to see how it could help the business and whether 602 00:37:46.440 --> 00:37:50.000 it's even possible to build. So I've been also working on the 603 00:37:50.000 --> 00:37:53.840 commercial side with kind of the, the simulating different 604 00:37:53.840 --> 00:37:57.680 shoe types and seeing how, what which ones make 605 00:37:57.680 --> 00:38:01.480 it more towards the like what what meet the, 606 00:38:01.480 --> 00:38:05.290 the needs of the the consumers. So likewise you 607 00:38:05.290 --> 00:38:08.890 can use these playgrounds as well to prioritize use 608 00:38:08.890 --> 00:38:12.530 cases and avoid really investing in projects that don't have a clear 609 00:38:12.530 --> 00:38:13.410 path to success. 610 00:38:17.170 --> 00:38:20.930 We've been talking about change management that is often underestimated 611 00:38:20.930 --> 00:38:24.450 in startups. What lessons from your own leadership experience 612 00:38:25.010 --> 00:38:28.850 would you pass on here? So I think 613 00:38:29.250 --> 00:38:33.090 the common understanding quite often that I hear from people is that 614 00:38:34.000 --> 00:38:37.640 when we think of change, we often think of a new software tool or a 615 00:38:37.640 --> 00:38:41.440 new process. But when we think about today, 616 00:38:41.440 --> 00:38:45.160 the changes are quite profound. So if we also pull it 617 00:38:45.160 --> 00:38:48.160 back to research from Gartner, by the end of even 618 00:38:49.040 --> 00:38:52.840 this year and beyond, there will be some of our virtual and hybrid 619 00:38:52.840 --> 00:38:56.240 meetings will be attended by agents, not just humans. 620 00:38:56.800 --> 00:38:59.920 And if we go a step further, we also expect that by 621 00:38:59.920 --> 00:39:03.660 2028 generative AI is really expected to 622 00:39:03.660 --> 00:39:07.500 be woven into our day to day work and some of it might 623 00:39:07.500 --> 00:39:10.540 not require that much of an oversight as it does now. 624 00:39:11.180 --> 00:39:14.780 And what is exciting also that it's not just 625 00:39:15.020 --> 00:39:18.620 for the tech teams as mentioned also it's also we have 626 00:39:18.700 --> 00:39:21.980 a large market to address for people that are non technical. 627 00:39:22.540 --> 00:39:26.340 And this means that there's already a lot of culture shift 628 00:39:26.340 --> 00:39:30.110 that is happening right now from your marketing, your 629 00:39:30.110 --> 00:39:33.110 sales team, human resources, all of these different 630 00:39:33.110 --> 00:39:36.870 departments. And if we ask ourselves then 631 00:39:37.110 --> 00:39:40.310 so if the change is so constant, why do 632 00:39:40.310 --> 00:39:43.030 startups so often underestimate the challenge? 633 00:39:43.990 --> 00:39:47.510 For me that brings me a little bit back to my time when I was 634 00:39:47.510 --> 00:39:51.270 a country manager for Germany in back market 635 00:39:51.270 --> 00:39:55.030 where I was managing change in a fast paced environment. Because 636 00:39:55.030 --> 00:39:58.710 we were really growing very fast as a company, we 637 00:39:58.710 --> 00:40:02.470 doubled ourselves within a few months and I was 638 00:40:02.470 --> 00:40:06.150 also responsible for developing, growing the business in the German 639 00:40:06.150 --> 00:40:09.630 market. So this means a lot of focus on customer acquisition, 640 00:40:09.710 --> 00:40:13.390 profitability, providing data driven feedback, 641 00:40:13.709 --> 00:40:17.070 developing action plans for our partners for the founders 642 00:40:17.630 --> 00:40:21.350 to also help us pinpoint how to improve the sales. So 643 00:40:21.350 --> 00:40:24.350 this really required me to have many different heads 644 00:40:25.560 --> 00:40:28.360 having the information, managing change, 645 00:40:28.840 --> 00:40:32.400 advocating for new ways of working, while also 646 00:40:32.400 --> 00:40:36.160 making sure that the human element is at the forefront of it 647 00:40:36.160 --> 00:40:39.960 all. So in the early days we were very focused 648 00:40:39.960 --> 00:40:43.720 on building a product and finding the market fit, especially as it was a 649 00:40:43.720 --> 00:40:47.440 new market in Germany. So the entire team quite 650 00:40:47.440 --> 00:40:51.000 often we met in a single room, we have very informal 651 00:40:51.000 --> 00:40:54.650 chats around pizza and change really happens 652 00:40:54.650 --> 00:40:58.010 organically. But as you know and as when you scale, 653 00:40:58.250 --> 00:41:02.050 that informal approach is just not happening 654 00:41:02.050 --> 00:41:05.610 as much anymore and can become also actually a major liability. 655 00:41:06.490 --> 00:41:09.450 So there's a bigger barrier then 656 00:41:10.330 --> 00:41:13.690 for effective communication, for instance from upper management 657 00:41:14.330 --> 00:41:17.930 and sometimes also an inability 658 00:41:17.930 --> 00:41:21.060 to address resistance in the company. So 659 00:41:21.700 --> 00:41:25.500 this means that also when we put this into a larger 660 00:41:25.500 --> 00:41:28.540 perspective is that a significant number of change 661 00:41:28.540 --> 00:41:32.100 initiatives fail. Because even when we look at the 662 00:41:32.100 --> 00:41:35.900 research, sometimes even as high as 60 or 70%, it 663 00:41:35.900 --> 00:41:39.300 doesn't really mean that the change itself was bad, but it's because 664 00:41:39.860 --> 00:41:43.460 the people side of the equation was in some ways also 665 00:41:43.460 --> 00:41:47.190 mishandled. And what I learned also from my time 666 00:41:47.190 --> 00:41:51.030 in startups is that people won't embrace change just 667 00:41:51.030 --> 00:41:54.830 because you tell them to. So they really need to understand the vision behind 668 00:41:54.910 --> 00:41:57.950 it and how it benefits them personally. 669 00:41:58.510 --> 00:42:02.270 So you have to shift the narrative from changes happening to 670 00:42:02.270 --> 00:42:05.870 us to actually we are shaping this change together. 671 00:42:06.670 --> 00:42:10.470 And this is especially true in my experience in startups 672 00:42:10.470 --> 00:42:14.190 that are hyper growing, like you know, from a few people to then over 673 00:42:14.190 --> 00:42:17.910 100 employees, there will be some people that you 674 00:42:17.910 --> 00:42:21.670 know are making part of that change, but some that are feeling left out as 675 00:42:21.670 --> 00:42:25.390 well because so many things are changing constantly 676 00:42:25.390 --> 00:42:29.230 as your team is scaling up. And this doesn't just 677 00:42:29.230 --> 00:42:33.070 mean new tech or processes. So you really need to focus on the people 678 00:42:33.150 --> 00:42:36.910 have to live with it every day, making, you know, sure 679 00:42:36.910 --> 00:42:40.670 that you have a two way conversation, not really a top down mandate 680 00:42:41.260 --> 00:42:45.020 involving also employees. Then early in the process identify 681 00:42:45.100 --> 00:42:48.860 change champions, which we just also shared earlier together for instance 682 00:42:48.860 --> 00:42:51.740 with the I whisperers. Those are 683 00:42:52.380 --> 00:42:55.980 really impactful people that can advocate for the new way of working 684 00:42:56.939 --> 00:43:00.300 and for a major change. One tip is also 685 00:43:00.620 --> 00:43:04.060 consider running a small pilot program with a subset of your team. 686 00:43:04.460 --> 00:43:08.220 And this allows you to test a change, gather feedback 687 00:43:08.700 --> 00:43:12.420 and make adjustments before rolling it out to the entire company and 688 00:43:12.420 --> 00:43:16.100 making sure that you have an agile approach here that mirrors 689 00:43:16.100 --> 00:43:18.380 actually the iterative mindset of startup. 690 00:43:21.340 --> 00:43:24.700 Let's go a little bit into the risk and the 691 00:43:24.700 --> 00:43:27.580 outlook Uber is currently 692 00:43:28.300 --> 00:43:32.020 leading with AI regulations. We had a 693 00:43:32.020 --> 00:43:34.620 lot of AI act content already, 694 00:43:36.090 --> 00:43:39.810 but I'm asking you, how will this shape the SAS 695 00:43:39.810 --> 00:43:43.210 models built on agents? So 696 00:43:43.210 --> 00:43:46.850 firstly, I believe that, or also what I've been seeing 697 00:43:46.850 --> 00:43:50.690 so far is that the EU AI act doesn't really treat all the 698 00:43:50.690 --> 00:43:53.690 AI equally. So as you have seen, 699 00:43:53.930 --> 00:43:57.250 there's a classification of systems into four risk 700 00:43:57.250 --> 00:44:00.970 categories. So the unacceptable, high, limited and 701 00:44:00.970 --> 00:44:04.690 minimal risks. So this means that the flexible 702 00:44:04.690 --> 00:44:08.370 structure around it means that you have to determine where the product 703 00:44:08.370 --> 00:44:12.010 lands before shipping a single feature. So if you think 704 00:44:12.010 --> 00:44:15.850 about the agentic, this means 705 00:44:15.850 --> 00:44:19.690 that you're thinking about agentic SARS for instance, used 706 00:44:19.690 --> 00:44:23.290 in consequential domains like hiring, lending or 707 00:44:23.290 --> 00:44:27.050 healthcare. So requirements here of course quite high. So you have 708 00:44:27.050 --> 00:44:30.590 to offer a total transparency, maintain up to date technical 709 00:44:30.590 --> 00:44:34.430 documentation, keep logs, ensure really explainability 710 00:44:34.910 --> 00:44:38.030 and also install mechanisms for human oversight. 711 00:44:38.510 --> 00:44:42.190 And SAS teams here will need to be able to explain how 712 00:44:42.190 --> 00:44:46.030 and why an AI or agent made the decisions and really 713 00:44:46.030 --> 00:44:49.830 be able to show this part of the system life cycle for 714 00:44:49.830 --> 00:44:53.590 audit purposes. If we go a step lower than for 715 00:44:53.590 --> 00:44:56.910 the limited risk systems, that means for instance customer support, 716 00:44:56.910 --> 00:45:00.350 chatbots or recommendation engines. So for those 717 00:45:01.470 --> 00:45:05.230 rules mostly mandate transparency. So for instance you're talking to a 718 00:45:05.230 --> 00:45:08.830 bot here, but it doesn't go as deep on documentation or 719 00:45:08.830 --> 00:45:12.550 oversight as in the previous bucket where we 720 00:45:12.550 --> 00:45:16.110 have also minimal risk AI that would be your spam filter 721 00:45:16.350 --> 00:45:19.790 which have almost no new requirements and then 722 00:45:20.030 --> 00:45:23.850 unacceptable risk. The these would be outrightly banned ways, 723 00:45:23.850 --> 00:45:27.650 for instance social scoring for the public sector. And 724 00:45:27.970 --> 00:45:31.610 if we explore this further. So for any kind 725 00:45:31.610 --> 00:45:35.330 of SAS company or not SaaS company, the first major 726 00:45:35.330 --> 00:45:39.130 challenge is mapping the agentic functionality 727 00:45:39.130 --> 00:45:42.850 into these buckets. So building an AI that touches 728 00:45:42.850 --> 00:45:46.450 hiring or financial processes. And the act also 729 00:45:46.450 --> 00:45:50.210 changes the game by making both the SaaS vendor and the customer 730 00:45:50.210 --> 00:45:53.970 so the deployer responsible for compliance. So 731 00:45:53.970 --> 00:45:57.410 you cannot just hand off the models or agents and walk away. 732 00:45:57.730 --> 00:46:01.330 So you really need to think about compliance support into the products, 733 00:46:01.650 --> 00:46:05.450 think audit logs, risk monitoring dashboards and user 734 00:46:05.450 --> 00:46:08.810 controls for then being able to do model 735 00:46:08.810 --> 00:46:12.490 oversight. And human oversight is then also a legal 736 00:46:12.490 --> 00:46:16.250 requirement. So even the most autonomous agents need a human 737 00:46:16.250 --> 00:46:19.250 in the loop architecture when dealing with high risk. 738 00:46:19.890 --> 00:46:23.650 So it also for me also is an interesting one here because 739 00:46:24.450 --> 00:46:28.210 compliance quite often is not just a legal burden, but can also 740 00:46:28.210 --> 00:46:32.050 be a market advantage. So as you have seen, surely, or 741 00:46:32.050 --> 00:46:35.650 also in one of the other episodes, there are high fines, for instance 742 00:46:35.650 --> 00:46:39.490 35 million or 7% of global turnover. So the cost 743 00:46:39.490 --> 00:46:43.310 of getting this wrong is really high. But I see also a lot 744 00:46:43.310 --> 00:46:46.870 of potential for startups who embrace compliance by offering 745 00:46:46.870 --> 00:46:50.470 security by design, robust documentation and 746 00:46:50.470 --> 00:46:53.990 explainable AI that helps them win, trust and 747 00:46:53.990 --> 00:46:57.670 unlock larger enterprise deals. So giving you an example, 748 00:46:57.670 --> 00:47:01.390 for instance startups like the Swiss startup Lakera, they 749 00:47:01.390 --> 00:47:05.230 saw the gap and build a business on it. So they create stress 750 00:47:05.230 --> 00:47:09.040 test suites that simulate real world attacks and scenarios, think 751 00:47:09.040 --> 00:47:12.720 about prompt exploits, adversarial contexts, 752 00:47:12.800 --> 00:47:16.400 and they help developers discover reliability gaps before 753 00:47:16.400 --> 00:47:20.120 attackers do. And also their platform embeds advanced 754 00:47:20.120 --> 00:47:23.919 defenses, so toxic content filters, guardrails for 755 00:47:23.919 --> 00:47:27.680 context windows or input sanitization directly 756 00:47:27.680 --> 00:47:31.280 into the agentic orchestration pipelines. So 757 00:47:31.280 --> 00:47:34.920 they pair also that automated evaluation framework with model 758 00:47:34.920 --> 00:47:38.010 alignment tools. So that agentic SaaS isn't just 759 00:47:38.330 --> 00:47:41.210 robust, but also explainable and safe by Design. 760 00:47:42.090 --> 00:47:45.930 So to sum this up is for a SaaS company, really 761 00:47:45.930 --> 00:47:49.690 building with agents is really to determine the risk classification. 762 00:47:50.250 --> 00:47:53.770 So you need to establish clear boundaries and permissions for your agents. 763 00:47:54.330 --> 00:47:57.890 So if you think about an agent that helps customers find furniture 764 00:47:57.890 --> 00:48:01.730 online, that would likely be a limited risk, but an 765 00:48:01.730 --> 00:48:05.120 agent that automates a hiring process would be high risk. 766 00:48:05.280 --> 00:48:08.760 So the AI act here places the burden on both the 767 00:48:08.760 --> 00:48:12.600 provider and of the AI system and the deployer, so the 768 00:48:12.600 --> 00:48:15.280 customer. So it's a shared journey of responsibility. 769 00:48:16.080 --> 00:48:19.920 And we also happy to help you over there. We have actually an interesting article 770 00:48:20.000 --> 00:48:23.760 where we describe also how we as AWS help on the EU AI 771 00:48:23.760 --> 00:48:27.240 Act. And yeah, you'll need to basically maintain 772 00:48:27.240 --> 00:48:31.060 detailed records of your AI system's life cycle from 773 00:48:31.060 --> 00:48:34.620 data governance to performance logs. You have to 774 00:48:34.860 --> 00:48:38.500 explain this really clearly. And we help for instance with 775 00:48:38.500 --> 00:48:41.980 AI Service card and the safety framework that gives 776 00:48:42.460 --> 00:48:45.820 customers also information on a model's intended use 777 00:48:45.820 --> 00:48:49.660 limitation and responsibilities design choices because it's super important for 778 00:48:49.660 --> 00:48:53.220 us. And yeah, as, as we've seen with this 779 00:48:53.220 --> 00:48:56.840 example, there's a lot of potential here 780 00:48:56.840 --> 00:49:00.280 to also transform compliance into real competitive 781 00:49:00.280 --> 00:49:03.480 advantage because security is a timeless principle. 782 00:49:03.960 --> 00:49:07.800 And when your customers then know your product is trustworthy 783 00:49:07.800 --> 00:49:11.240 and compliant, it really builds a foundation of trust. 784 00:49:11.560 --> 00:49:15.000 So it's not really about just avoiding defiance, but really about 785 00:49:15.000 --> 00:49:18.040 gaining market trust and also 786 00:49:18.600 --> 00:49:21.520 additional market share. So I find this a really interesting space. 787 00:49:24.710 --> 00:49:28.430 Coming to our last question here. Sitting here 788 00:49:28.430 --> 00:49:32.070 together for almost three hours. Let's give it your 789 00:49:32.070 --> 00:49:35.230 best, let's give it your last. Gartner predicts by 790 00:49:35.230 --> 00:49:38.710 2028 AI will be so embedded in 791 00:49:38.710 --> 00:49:41.990 productivity apps that oversight will be rare. 792 00:49:42.470 --> 00:49:45.590 What does this mean for SAS founders today? 793 00:49:47.120 --> 00:49:50.960 Yeah, that's a really great question here and I would really like to unpack this 794 00:49:50.960 --> 00:49:54.720 further. So imagine the day to day decisions 795 00:49:54.720 --> 00:49:57.920 in a business. So you have the invoice processing contract 796 00:49:58.000 --> 00:50:01.360 generation to help desk automation so 797 00:50:01.840 --> 00:50:05.200 handled largely looking into the future, perhaps by 798 00:50:05.360 --> 00:50:09.160 agents that set their own goals, solve problems and communicate 799 00:50:09.160 --> 00:50:12.960 with other tools or without requiring actually constant 800 00:50:12.960 --> 00:50:16.370 human checkpoints. And that's as we've seen before, 801 00:50:17.090 --> 00:50:20.930 that's one way where the industry is going. So we expect that 802 00:50:20.930 --> 00:50:24.370 one third of enterprise apps will include agentic AI by 803 00:50:24.370 --> 00:50:27.570 2028. So this means that around 804 00:50:27.570 --> 00:50:31.170 15% of all work decisions is fully on autopilot by 805 00:50:31.170 --> 00:50:34.730 then. And this oversight burn is shrinking 806 00:50:34.730 --> 00:50:37.810 and that changes everything for SaaS product design. 807 00:50:38.450 --> 00:50:42.010 And this means also that Agenti elevates the SaaS 808 00:50:42.010 --> 00:50:45.410 products from tools that automate routine processes to 809 00:50:45.410 --> 00:50:48.850 platforms that Reason, adapt and act. So 810 00:50:49.090 --> 00:50:52.890 founders now have to ask themselves. And that's what I advise in a lot of 811 00:50:52.890 --> 00:50:56.570 conversations that I go to with 812 00:50:56.570 --> 00:51:00.410 customers, is am I building a workflow or am I building an 813 00:51:00.410 --> 00:51:03.970 agent? Because you have to really decide based on, you know, different 814 00:51:03.970 --> 00:51:07.250 factors around the complexity, the cost of error, so many 815 00:51:07.890 --> 00:51:11.380 factors that you take into consideration. And also is the 816 00:51:11.380 --> 00:51:15.020 task complex enough to need reasoning or not just 817 00:51:15.020 --> 00:51:18.620 automation? And do I need also multi 818 00:51:18.620 --> 00:51:21.980 agent systems with specialized sub agents 819 00:51:22.620 --> 00:51:25.900 where central agents manage high level goals. 820 00:51:26.460 --> 00:51:30.020 And the good news here is it has never been easier 821 00:51:30.020 --> 00:51:33.860 to start building those because we have so many tools 822 00:51:33.860 --> 00:51:37.390 at our disposition that many startups are using like Landgraph Crew. 823 00:51:38.340 --> 00:51:42.020 Also we launched Amazon Bedrock Agent Core just a couple of weeks back. 824 00:51:42.340 --> 00:51:45.780 So this lets developers spin up agents that fetch data, 825 00:51:45.780 --> 00:51:49.340 execute API calls, evaluate outputs and even 826 00:51:49.340 --> 00:51:52.980 self improve via continuous feedback. So instead of 827 00:51:52.980 --> 00:51:56.740 customer support tickets bottlenecking, really a help desk, 828 00:51:56.740 --> 00:52:00.460 here the agent reads complex situations, 829 00:52:00.460 --> 00:52:04.020 it orchestrates specialized micro agents to resolve 830 00:52:04.020 --> 00:52:07.510 tasks, it verifies accuracy and delivers a 831 00:52:07.510 --> 00:52:10.670 consolidated solution with minimal human intervention. 832 00:52:11.470 --> 00:52:14.750 So I'm working also in finance with startups whose 833 00:52:15.310 --> 00:52:18.750 solution also validates invoices in minutes rather than hours. 834 00:52:19.070 --> 00:52:22.910 A single agent coordinates the data extraction, checks the supply 835 00:52:22.910 --> 00:52:26.270 status, crunches rules, using business intelligence. 836 00:52:26.910 --> 00:52:30.550 So really for SaaS founders that are pivoting towards 837 00:52:30.550 --> 00:52:33.740 agentic AI, that means products will stop 838 00:52:34.140 --> 00:52:37.460 being just helping humans, but they begin owning 839 00:52:37.460 --> 00:52:41.180 outcomes end to end. And another thing that I find 840 00:52:41.180 --> 00:52:44.900 really important in this context is that founders 841 00:52:44.900 --> 00:52:48.740 need to ask themselves should they build custom agents or do 842 00:52:48.740 --> 00:52:52.460 they want to build those off the shelf or do they want to partner 843 00:52:52.460 --> 00:52:55.980 for parties like thinking about Amazon Q 844 00:52:56.300 --> 00:52:59.660 strands. So so many different tools for different builders needs. 845 00:53:01.020 --> 00:53:04.620 And this doesn't get easier with all of these different 846 00:53:04.620 --> 00:53:08.460 frameworks and commercial solutions. Recently when I looked 847 00:53:08.460 --> 00:53:12.220 into it, so there are thousands of partners, even like over 100,000 848 00:53:12.220 --> 00:53:15.900 partners worldwide that are contributing to the agenti 849 00:53:15.980 --> 00:53:18.940 deployment, so making it really 850 00:53:19.660 --> 00:53:23.500 a huge market at the moment. So this means, as we discussed earlier, you need 851 00:53:23.500 --> 00:53:26.700 to upscale your teams, secure high quality data 852 00:53:27.360 --> 00:53:30.560 and architecture systems for flexibility, speed and security. 853 00:53:31.520 --> 00:53:35.200 And you know, as we said also earlier, the road is 854 00:53:35.200 --> 00:53:39.040 also not frictionless. So it's very likely that not all of the 855 00:53:39.040 --> 00:53:41.920 agentic projects that you are thinking about are not 856 00:53:42.640 --> 00:53:45.920 making it into production. It could be because of 857 00:53:45.920 --> 00:53:49.360 complexity, implementation, evaluation, maintaining security 858 00:53:50.480 --> 00:53:53.790 and also importantly, not all 859 00:53:53.790 --> 00:53:56.870 tasks make sense for agents. So there might be 860 00:53:57.590 --> 00:54:01.430 higher error costs, complex human judgment and regulatory 861 00:54:01.430 --> 00:54:05.230 requirements that we will see increase. Or so that may mean that human 862 00:54:05.230 --> 00:54:08.710 in the loop is required. So for those that are listening in today, 863 00:54:09.590 --> 00:54:13.270 I think that by 2028 apps will really 864 00:54:13.270 --> 00:54:16.750 embed Genti so they will win by delivering real 865 00:54:16.750 --> 00:54:20.550 autonomous impact. So faster outcomes, reduced manual work 866 00:54:20.550 --> 00:54:24.220 and new business value. So really the magic I 867 00:54:24.220 --> 00:54:27.660 see here is in transitioning from productivity enhancement 868 00:54:27.820 --> 00:54:30.140 to operational ownership. By the eye, 869 00:54:31.500 --> 00:54:35.140 we're seeing so many great developments with open source 870 00:54:35.140 --> 00:54:38.700 models, low code platforms and partner networks that 871 00:54:38.700 --> 00:54:41.900 make these agentic developments accessible for 872 00:54:42.380 --> 00:54:45.900 an increasingly large market. But of course 873 00:54:46.220 --> 00:54:49.990 make sure you stay with these main frameworks in 874 00:54:49.990 --> 00:54:53.790 mind that always hold true for software development. So you need evaluation 875 00:54:53.790 --> 00:54:57.590 metrics and responsible guardrails because as Werner Focus, 876 00:54:57.590 --> 00:55:00.830 our CEO suppose calls it, everything 877 00:55:01.150 --> 00:55:04.990 fails all the time. So you really need to have this into 878 00:55:04.990 --> 00:55:08.430 consideration. So really thinking about 879 00:55:08.990 --> 00:55:12.350 what is will my app with the overseer, will it be the owner? 880 00:55:12.590 --> 00:55:16.110 So the next few years will show how this is evolving. 881 00:55:16.770 --> 00:55:20.490 But I think that the winners will be those that make agentic 882 00:55:20.490 --> 00:55:24.210 AI not just a feature, but really a heartbeat and essential part of their business. 883 00:55:25.650 --> 00:55:29.410 I think that are really great. Final 884 00:55:29.490 --> 00:55:32.610 words Jennifer, thank you for being here, 885 00:55:33.250 --> 00:55:36.970 spending like three hours, two hours recording two 886 00:55:36.970 --> 00:55:39.970 episodes. Thank you very much. It was a pleasure having you as a guest. 887 00:55:40.610 --> 00:55:44.270 Likewise. Really enjoyed myself and and yeah looking 888 00:55:44.270 --> 00:55:47.790 forward perhaps to the next one in. A way that would be awesome. 889 00:55:47.870 --> 00:55:50.430 Have a good day. Bye bye. Thank you. Bye bye. 890 00:55:55.390 --> 00:55:58.830 That's all folks. Find more news streams, 891 00:55:59.150 --> 00:56:00.310 events and 892 00:56:00.310 --> 00:56:04.590 interviews@www.startuprad.IO. 893 00:56:04.990 --> 00:56:07.230 remember, sharing is caring. 894 00:56:08.440 --> 00:56:20.530 Sam.