رونوشت کامل AI Found a 10-Year-Old Bug in the Code Behind Most of the Web

great to have you on the 00:03:47.519 --> 00:03:49.599 show. Um you joined Taktile literally 00:03:49.599 --> 00:03:53.200 last month in September and so you are 00:03:53.200 --> 00:03:55.599 coming from datab bricks and your 00:03:55.599 --> 00:03:58.480 background is primarily in working on 00:03:58.480 --> 00:04:01.280 database and data companies. So what do 00:04:01.280 --> 00:04:04.879 you bring to Taktile and why move um to 00:04:04.879 --> 00:04:08.480 decision level um in in the whole flow? 00:04:08.480 --> 00:04:11.760 Sure. Yeah. Um I guess so I think I'm 00:04:11.760 --> 00:04:13.360 joining Taktile at a really exciting 00:04:13.360 --> 00:04:16.479 time. So not only have models gotten 00:04:16.479 --> 00:04:18.400 exceptionally better in the last 12 00:04:18.400 --> 00:04:21.040 months. uh Taktile has a a platform in 00:04:21.040 --> 00:04:22.960 production on which we can build these 00:04:22.960 --> 00:04:25.360 agents and so we can me we can put the 00:04:25.360 --> 00:04:27.199 production readiness of the platform and 00:04:27.199 --> 00:04:28.720 the exciting new model developments 00:04:28.720 --> 00:04:30.160 together and build something really 00:04:30.160 --> 00:04:32.720 exciting. Uh you're right that my my 00:04:32.720 --> 00:04:34.000 background originally is in data 00:04:34.000 --> 00:04:35.600 databases. I fell in love running 00:04:35.600 --> 00:04:37.360 databases at Twitter uh we're running 00:04:37.360 --> 00:04:39.600 highcale systems there uh went into 00:04:39.600 --> 00:04:41.600 in the transition from onrem to 00:04:41.600 --> 00:04:43.680 cloud and then neon which was a 00:04:43.680 --> 00:04:46.000 Postgress startup um before being 00:04:46.000 --> 00:04:48.400 acquired by data bricks. But really the 00:04:48.400 --> 00:04:49.680 thing I enjoyed was that it was the 00:04:49.680 --> 00:04:52.000 hardest part of the stack operationally. 00:04:52.000 --> 00:04:54.800 So uh big databases are hard to 00:04:54.800 --> 00:04:56.960 maintain. Uh it's a high expertise 00:04:56.960 --> 00:04:58.400 system and and that's the the 00:04:58.400 --> 00:04:59.520 challenging thing and there were a lot 00:04:59.520 --> 00:05:00.960 of really interesting developments in 00:05:00.960 --> 00:05:02.800 databases over that time. But if you 00:05:02.800 --> 00:05:04.639 look at where the market's gone in the 00:05:04.639 --> 00:05:06.240 last sort of couple of years, it's all 00:05:06.240 --> 00:05:09.600 AI, right? Um and running AI and in 00:05:09.600 --> 00:05:12.800 particular making high quality decisions 00:05:12.800 --> 00:05:16.560 with AI is is really difficult. And so 00:05:16.560 --> 00:05:18.560 in the first half of the year at data 00:05:18.560 --> 00:05:20.479 bricks uh I spent the time sort of 00:05:20.479 --> 00:05:22.160 integrating our our technology into the 00:05:22.160 --> 00:05:25.600 company and the second half I um I was 00:05:25.600 --> 00:05:27.440 working on aentic coding and you'll if 00:05:27.440 --> 00:05:29.360 you look at their um their releases sort 00:05:29.360 --> 00:05:30.720 of around data bricks apps in the last 00:05:30.720 --> 00:05:31.840 year that's some of the stuff I was 00:05:31.840 --> 00:05:34.080 working on and as I got there it's I 00:05:34.080 --> 00:05:36.720 found that it's very easy to prompt an 00:05:36.720 --> 00:05:39.199 AI to to do something if you're willing 00:05:39.199 --> 00:05:41.520 to accept a level of error right but 00:05:41.520 --> 00:05:43.360 then the really interesting word becomes 00:05:43.360 --> 00:05:45.039 when you want to make sure that every 00:05:45.039 --> 00:05:47.680 time the customer uses the AI, they get 00:05:47.680 --> 00:05:50.639 the kind of response they expect or 00:05:50.639 --> 00:05:53.759 there's like a reliable um failure mode 00:05:53.759 --> 00:05:55.680 in which it passes out to a human team 00:05:55.680 --> 00:05:58.479 and it accepts that it's uh it's unable 00:05:58.479 --> 00:06:00.720 to make a decision for whatever reason 00:06:00.720 --> 00:06:03.199 and it reliably passes out to a human in 00:06:03.199 --> 00:06:05.520 the loop. So that's one of the the key 00:06:05.520 --> 00:06:07.440 details of of Taktile that I think is 00:06:07.440 --> 00:06:11.120 really interesting is the ability to to 00:06:11.120 --> 00:06:13.199 pass to the human team who have tons of 00:06:13.199 --> 00:06:15.600 domain and business expertise uh to 00:06:15.600 --> 00:06:17.520 record the outcome of that decision from 00:06:17.520 --> 00:06:19.600 the human team and pass that into the 00:06:19.600 --> 00:06:21.520 learning loop back into the next round 00:06:21.520 --> 00:06:23.600 of decisions. Um and so that's one of 00:06:23.600 --> 00:06:24.880 the really interesting problems that 00:06:24.880 --> 00:06:27.680 brought me to Taktile. James, um, if we 00:06:27.680 --> 00:06:30.560 were to look more on what happens under 00:06:30.560 --> 00:06:33.360 the hood, uh, in Taktile, let's say 00:06:33.360 --> 00:06:35.039 Semini, we are running a small business, 00:06:35.039 --> 00:06:38.319 we apply for a loan at Monzo who uses, 00:06:38.319 --> 00:06:40.960 uh, Taktile. What happens like if you 00:06:40.960 --> 00:06:42.960 break it down to a not non-technical 00:06:42.960 --> 00:06:43.680 audience? 00:06:43.680 --> 00:06:45.919 Yeah. So, I mean, this is probably easy 00:06:45.919 --> 00:06:48.400 if we if we go back a few years um, and 00:06:48.400 --> 00:06:50.639 we think sort of the the pre-automation, 00:06:50.639 --> 00:06:53.280 pre-AI world, applying for a loan, you 00:06:53.280 --> 00:06:55.360 submit a bunch of paper forms. those 00:06:55.360 --> 00:06:56.800 eventually land on someone's desk, 00:06:56.800 --> 00:06:59.199 whether that's as an email or PDF um or 00:06:59.199 --> 00:07:02.000 in an online tool. A human reviews that 00:07:02.000 --> 00:07:03.599 against a set of rules that are 00:07:03.599 --> 00:07:05.919 described in both uh the the kind of 00:07:05.919 -->

00:07:07.520 loan you're trying to get, the bank's 00:07:07.520 --> 00:07:10.319 own policies um and like a level of risk 00:07:10.319 --> 00:07:12.160 the the bank is willing to make in terms 00:07:12.160 --> 00:07:14.160 of how well does your application 00:07:14.160 --> 00:07:16.319 conform to those policies that bounces 00:07:16.319 --> 00:07:17.919 around inside the bank. Let's say it's a 00:07:17.919 --> 00:07:20.080 mortgage rather than a small scale loan. 00:07:20.080 --> 00:07:22.639 and a mortgage can be operated in an 00:07:22.639 --> 00:07:25.599 order of weeks at the moment. Right now, 00:07:25.599 --> 00:07:27.039 we've automated that. That's all a bunch 00:07:27.039 --> 00:07:29.680 of PDFs. We can pass those through OCR 00:07:29.680 --> 00:07:31.440 optical character resolution and then 00:07:31.440 --> 00:07:33.759 pass it to an agent. Uh we've given the 00:07:33.759 --> 00:07:36.000 policy PDF to the agent and there's a 00:07:36.000 --> 00:07:37.919 very there's a strict deterministic 00:07:37.919 --> 00:07:40.960 decision flow that the the claim will go 00:07:40.960 --> 00:07:43.199 through. That's a deterministic decision 00:07:43.199 --> 00:07:45.199 flow, but at some point in that model, 00:07:45.199 --> 00:07:47.360 it hits an a an agent that gets to make 00:07:47.360 --> 00:07:49.680 non-deterministic decisions based on the 00:07:49.680 --> 00:07:51.919 policies of your bank. Um, and based on 00:07:51.919 --> 00:07:54.080 maybe the the circumstances of the 00:07:54.080 --> 00:07:55.919 particular week. Uh, maybe the markets 00:07:55.919 --> 00:07:57.280 are good, we're willing to make a 00:07:57.280 --> 00:07:58.960 slightly more risky loan. Maybe the 00:07:58.960 --> 00:08:00.000 markets are bad, we're going to make 00:08:00.000 --> 00:08:00.960 we're going to we're going to dial the 00:08:00.960 --> 00:08:03.680 risk back down. But either way, there is 00:08:03.680 --> 00:08:05.759 a a fixed policy that's in place from 00:08:05.759 --> 00:08:08.080 the business. Then there's a controls 00:08:08.080 --> 00:08:10.560 from the business users uh that 00:08:10.560 --> 00:08:13.440 determine sort of based on today's 00:08:13.440 --> 00:08:15.759 business scenario, how do we want these 00:08:15.759 --> 00:08:17.280 flows to work? And right there in the 00:08:17.280 --> 00:08:19.120 middle of the flow is an agent making 00:08:19.120 --> 00:08:20.800 decisions based on the inputs it being 00:08:20.800 --> 00:08:22.800 given both the like the hard rules and 00:08:22.800 --> 00:08:24.639 the um the softer rules. 00:08:24.639 --> 00:08:26.080 Um and then it'll output a 00:08:26.080 --> 00:08:27.360 recommendation. 00:08:27.360 --> 00:08:30.000 James, uh by the way, congrats. Um you 00:08:30.000 --> 00:08:34.000 guys raised $110 million in June. Um now 00:08:34.000 --> 00:08:36.959 since you are a CTO at the company how 00:08:36.959 --> 00:08:38.880 are you planning to spend that money in 00:08:38.880 --> 00:08:40.719 terms of engineering and software 00:08:40.719 --> 00:08:41.680 development? 00:08:41.680 --> 00:08:44.000 Yeah um really I mean uh getting the 00:08:44.000 --> 00:08:45.519 series C is a huge accomplishment by the 00:08:45.519 --> 00:08:48.080 team. I'm really pleased for them and I 00:08:48.080 --> 00:08:50.800 think it it says a lot about about the 00:08:50.800 --> 00:08:52.720 health of the company and the success of 00:08:52.720 --> 00:08:55.040 the platform and in particular one of 00:08:55.040 --> 00:08:56.880 with the invest the main investor being 00:08:56.880 --> 00:08:59.279 Goldman. I think that says a lot about 00:08:59.279 --> 00:09:02.480 their confidence in Taktile in march 00:09:02.480 --> 00:09:04.880 into banking. So I think Taktile has has 00:09:04.880 --> 00:09:07.760 a lot of uh sort of new bank neo bank 00:09:07.760 --> 00:09:10.399 customers uh fintexs um but the future 00:09:10.399 --> 00:09:12.720 of the company is in enterprise and so 00:09:12.720 --> 00:09:15.040 with with Goldman's uh support we're 00:09:15.040 --> 00:09:17.200 sort of we're now opening doors into the 00:09:17.200 --> 00:09:18.640 largest enterprises in the world. I 00:09:18.640 --> 00:09:20.320 think one of our largest customers is 00:09:20.320 --> 00:09:22.640 Aliance. Um there's a big profile of 00:09:22.640 --> 00:09:24.160 them on the on the page but we're 00:09:24.160 --> 00:09:26.240 heading into Wall Street um in a serious 00:09:26.240 --> 00:09:28.160 way. I think there's two things we need 00:09:28.160 --> 00:09:31.279 to do to succeed in enterprise. One is a 00:09:31.279 --> 00:09:33.920 very standard enterprise journey where 00:09:33.920 --> 00:09:36.000 we we harden things sort of role-based 00:09:36.000 --> 00:09:38.399 access controls, networking limits, all 00:09:38.399 --> 00:09:40.080 those kinds of things. But the second 00:09:40.080 --> 00:09:43.360 thing is is differentiating features. Um 00:09:43.360 --> 00:09:46.240 and there what we look at in enterprise 00:09:46.240 --> 00:09:49.279 is the hard separation between um what 00:09:49.279 --> 00:09:51.279 the technical teams are looking for and 00:09:51.279 --> 00:09:53.440 what the business users need. And so we 00:09:53.440 --> 00:09:55.680 want to make sure that it's easy for an 00:09:55.680 --> 00:09:59.360 enterprise security or IT team to deploy 00:09:59.360 --> 00:10:01.279 Taktile with the hard rules that govern 00:10:01.279 --> 00:10:04.080 how Taktile should work in general, but 00:10:04.080 --> 00:10:05.600 then to leave flexibility for the 00:10:05.600 --> 00:10:07.279 business users to bring in that business 00:10:07.279 --> 00:10:09.120 context, their understanding of sort of 00:10:09.120 --> 00:10:10.640 the market and the particulars of a 00:10:10.640 --> 00:10:12.560 given week or time and to tweak things 00:10:12.560 --> 00:10:15.360 as they go. and then for them to set 00:10:15.360 --> 00:10:17.920 ranges um inside of those those 00:10:17.920 --> 00:10:20.160 parameters

and for our system to learn 00:10:20.160 --> 00:10:23.120 through operation where that slider 00:10:23.120 --> 00:10:25.440 should be inside of the rules it's being 00:10:25.440 --> 00:10:26.160 given. 00:10:26.160 --> 00:10:26.560 Okay, 00:10:26.560 --> 00:10:27.920 there we have like a reinforcement 00:10:27.920 --> 00:10:30.160 learning loop. Um and that's how we will 00:10:30.160 --> 00:10:32.160 we tweak the decision making for the 00:10:32.160 --> 00:10:32.800 customers. 00:10:32.800 --> 00:10:34.800 James, you you mentioned Allianz 00:10:34.800 --> 00:10:37.519 Partners. Uh they say that 30% more 00:10:37.519 --> 00:10:39.600 health claims now goes through you guys 00:10:39.600 --> 00:10:42.800 with no human involved. Do you see that 00:10:42.800 --> 00:10:46.320 percentage increasing to even 100% uh 00:10:46.320 --> 00:10:48.800 and what the people are still checking 00:10:48.800 --> 00:10:49.920 at the moment? 00:10:49.920 --> 00:10:52.560 Yeah, so I think it it can be brought to 00:10:52.560 --> 00:10:54.240 100% but what it really comes down to is 00:10:54.240 --> 00:10:56.480 what the business is trying to achieve. 00:10:56.480 --> 00:11:00.800 So um for claims processing in a in a 00:11:00.800 --> 00:11:03.040 regular week within SLAs's you can have 00:11:03.040 --> 00:11:05.360 a fix team working and hitting SLAs's 00:11:05.360 --> 00:11:07.680 fine. Uh what what some of our insurance 00:11:07.680 --> 00:11:09.839 customers find though is exceptional 00:11:09.839 --> 00:11:11.920 events come in uh such as a severe 00:11:11.920 --> 00:11:13.440 change in the weather uh which 00:11:13.440 --> 00:11:16.079 dramatically change claims volumes 00:11:16.079 --> 00:11:17.920 there. They have the decision about 00:11:17.920 --> 00:11:20.079 either pulling in more team members to 00:11:20.079 --> 00:11:23.440 hit timing SLAs's or by automating more. 00:11:23.440 --> 00:11:24.640 And those are the two main decisions 00:11:24.640 --> 00:11:27.120 they have to make more by dropping the 00:11:27.120 --> 00:11:28.720 quality of the checks they're doing. So 00:11:28.720 --> 00:11:30.399 you can speed up a human team by giving 00:11:30.399 --> 00:11:33.040 them less work um or by bringing in more 00:11:33.040 --> 00:11:35.279 people. Um, those are all very expensive 00:11:35.279 --> 00:11:37.040 decisions. So, if you're willing to drop 00:11:37.040 --> 00:11:38.560 quality, you make payouts when you 00:11:38.560 --> 00:11:40.079 shouldn't. If you're pulling in more 00:11:40.079 --> 00:11:41.360 team members, you have to pay for that 00:11:41.360 --> 00:11:43.600 human time. So, what we're giving is a 00:11:43.600 --> 00:11:46.480 third option is spend more tokens. And 00:11:46.480 --> 00:11:48.160 generally speaking, token spend is 00:11:48.160 --> 00:11:50.320 expensive, but it's not as expensive as 00:11:50.320 --> 00:11:53.440 people. And so, we have a a detailed 00:11:53.440 --> 00:11:55.040 sort of analysis as part of Taktile 00:11:55.040 --> 00:11:57.440 Labs. We have a paper up talking about 00:11:57.440 --> 00:12:00.399 the quality of results, the time taken, 00:12:00.399 --> 00:12:02.959 the token spend, but comparing that 00:12:02.959 --> 00:12:06.079 against human decision-m time. Um, and 00:12:06.079 --> 00:12:09.279 so in terms of both pay and timing, uh, 00:12:09.279 --> 00:12:11.120 we find that's a really valuable third 00:12:11.120 --> 00:12:14.399 option for companies to have. So, you 00:12:14.399 --> 00:12:16.399 know, sitting at 30% today, they could 00:12:16.399 --> 00:12:17.600 make the business decision to try and 00:12:17.600 --> 00:12:19.760 move that to 35%. 00:12:19.760 --> 00:12:21.920 What we give them is the ability to 00:12:21.920 --> 00:12:23.760 replay the decisions from the last week 00:12:23.760 --> 00:12:26.160 or last month assuming they wanted to 00:12:26.160 --> 00:12:28.160 automate further and to see what the 00:12:28.160 --> 00:12:30.560 outcomes of that would have been. And 00:12:30.560 --> 00:12:32.399 that's the short-term outcomes in terms 00:12:32.399 --> 00:12:34.639 of the decisions they make today. But we 00:12:34.639 --> 00:12:37.680 also take like final outcome decisions. 00:12:37.680 --> 00:12:40.079 So for instance, in a loan case, we get 00:12:40.079 --> 00:12:42.399 to see over the months whether the loan 00:12:42.399 --> 00:12:44.160 was actually repaid by the person we 00:12:44.160 --> 00:12:46.560 granted the loan to. And so with that 00:12:46.560 --> 00:12:49.040 data we say on the day of the loan we 00:12:49.040 --> 00:12:50.639 knew this much about the person and we 00:12:50.639 --> 00:12:53.360 made this decision but 12 months later 00:12:53.360 --> 00:12:55.200 we now know that that person repaid and 00:12:55.200 --> 00:12:58.000 this one this person didn't. So that all 00:12:58.000 --> 00:12:59.680 goes back into the system in terms of 00:12:59.680 --> 00:13:02.079 ramping decision quality and giving the 00:13:02.079 --> 00:13:04.399 ability to ramp automation and actually 00:13:04.399 --> 00:13:06.399 increase quality and make fewer mistakes 00:13:06.399 --> 00:13:09.040 as opposed to uh to a human team that 00:13:09.040 --> 00:13:10.480 constantly needs retraining and 00:13:10.480 --> 00:13:12.800 expansion. James, most of the things 00:13:12.800 --> 00:13:15.200 that you guys are doing um goes around 00:13:15.200 --> 00:13:18.320 decision making, right? Um and when it 00:13:18.320 --> 00:13:19.920 comes to decision- making, we know there 00:13:19.920 --> 00:13:22.800 is bias. There is always bias even with 00:13:22.800 --> 00:13:25.360 with AI. Um because again, let's say the 00:13:25.360 --> 00:13:27.839 company has XY Z policy and the uh 00:13:27.839 --> 00:13:29.839 basically AI will follow those policies. 00:13:29.839 --> 00:13:32.240 How do you guys ensure that there is no 00:13:32.240 --> 00:13:35.120 bias and everyone is treated equally and 00:13:35.120 --> 00:13:37.839 AI also I mean treats uh applicants

00:13:37.839 --> 00:13:40.079 equally based on let's say their income 00:13:40.079 --> 00:13:43.120 and etc. but will will not go to other 00:13:43.120 --> 00:13:46.240 elements that uh could uh bring the 00:13:46.240 --> 00:13:49.440 results uh eventually uh relate to bias. 00:13:49.440 --> 00:13:50.959 Yeah, that's a really interesting 00:13:50.959 --> 00:13:53.680 question and one I'm not quite ready for 00:13:53.680 --> 00:13:57.040 at day 45 at the company. Uh thinking 00:13:57.040 --> 00:13:59.680 about it though, that's a that's a I 00:13:59.680 --> 00:14:01.519 think what we've learned in software 00:14:01.519 --> 00:14:04.079 engineering is we've done a lot of the 00:14:04.079 --> 00:14:05.920 automation of the software development 00:14:05.920 --> 00:14:08.240 life cycle. What we're finding as we go 00:14:08.240 --> 00:14:10.880 from trying to um take the basic work of 00:14:10.880 --> 00:14:12.399 a software engineer, implement this 00:14:12.399 --> 00:14:14.399 picket and then produce an outcome, that 00:14:14.399 --> 00:14:16.720 part's actually quite easy. The harder 00:14:16.720 --> 00:14:19.600 part is applying higher level judgments 00:14:19.600 --> 00:14:21.360 to the quality of the work. Whether 00:14:21.360 --> 00:14:23.120 that's in terms of like longer term 00:14:23.120 --> 00:14:25.600 engineering health of the codebase or in 00:14:25.600 --> 00:14:27.120 terms of product thinking about where 00:14:27.120 --> 00:14:28.880 you want the product to go rather than 00:14:28.880 --> 00:14:30.800 going step by step, brick by brick 00:14:30.800 --> 00:14:32.959 randomly, uh you want it to follow some 00:14:32.959 --> 00:14:35.120 kind of product strategy. And and 00:14:35.120 --> 00:14:37.360 thirdly in terms of design uh where if 00:14:37.360 --> 00:14:39.279 you get your entire app designed by 00:14:39.279 --> 00:14:41.360 claude it will sort of iterate its way 00:14:41.360 --> 00:14:43.040 into something very ugly and that that's 00:14:43.040 --> 00:14:46.639 not maintainable. And so if I apply that 00:14:46.639 --> 00:14:50.000 sort of second level of review idea to 00:14:50.000 --> 00:14:52.240 my codebase what I do is I have one 00:14:52.240 --> 00:14:55.120 coding agent separate to that I have a 00:14:55.120 --> 00:14:56.959 product agent that understands our 00:14:56.959 --> 00:14:59.199 product strategy. I have a design agent 00:14:59.199 --> 00:15:01.040 that understands our design strategy. 00:15:01.040 --> 00:15:03.360 and all of them look together at work 00:15:03.360 --> 00:15:04.800 we've been doing and make their own 00:15:04.800 --> 00:15:07.040 judgments with separate context and 00:15:07.040 --> 00:15:08.880 separate uh vetoing power over the 00:15:08.880 --> 00:15:10.160 decision-m 00:15:10.160 --> 00:15:11.199 okay 00:15:11.199 --> 00:15:13.519 if a customer came to me and said how do 00:15:13.519 --> 00:15:15.199 I build a flow and make sure it has 00:15:15.199 --> 00:15:17.760 limited bias I think what I would advise 00:15:17.760 --> 00:15:20.720 them to do is build an antibbias agent 00:15:20.720 --> 00:15:23.600 and say looking at this decision history 00:15:23.600 --> 00:15:25.120 looking at these these incoming 00:15:25.120 --> 00:15:29.120 customers um are we like systemically 00:15:29.120 --> 00:15:30.560 are we making decisions that that 00:15:30.560 --> 00:15:32.480 demonst demonstrate bias or that don't 00:15:32.480 --> 00:15:34.399 and what counterveiling signals can we 00:15:34.399 --> 00:15:35.760 put into that reinforcement loop to 00:15:35.760 --> 00:15:38.000 avoid it. I think that would be a really 00:15:38.000 --> 00:15:39.760 interesting engagement and I would love 00:15:39.760 --> 00:15:41.519 to be part of that. Um I've worked on 00:15:41.519 --> 00:15:43.920 debiasing projects at previous companies 00:15:43.920 --> 00:15:46.320 um but I I that customer haven't come to 00:15:46.320 --> 00:15:48.880 me me personally yet uh but I'd love to 00:15:48.880 --> 00:15:49.920 work together if one of them are 00:15:49.920 --> 00:15:50.399 listening. 00:15:50.399 --> 00:15:53.279 As you know on our show we uh analyze 00:15:53.279 --> 00:15:56.240 the news uh within the tech and industry 00:15:56.240 --> 00:15:59.120 and um give our take or guest take on 00:15:59.120 --> 00:16:00.880 them. One of the news that caught our 00:16:00.880 --> 00:16:02.560 attention last week was coming from 00:16:02.560 --> 00:16:05.120 Plate. They launched their own credit 00:16:05.120 --> 00:16:07.519 scoring uh service last week. Do you 00:16:07.519 --> 00:16:09.199 think I know that you guys work with 00:16:09.199 --> 00:16:12.480 with plate as well? Will that create a 00:16:12.480 --> 00:16:14.399 kind of a comp competition spirit 00:16:14.399 --> 00:16:16.560 between Taktile and and plate? Uh what 00:16:16.560 --> 00:16:17.759 is your take on that? 00:16:17.759 --> 00:16:20.000 Well, I think it's always interesting 00:16:20.000 --> 00:16:22.240 when there are competitors because it 00:16:22.240 --> 00:16:23.839 means you're in a market that multiple 00:16:23.839 --> 00:16:25.920 people have identified and it it sort of 00:16:25.920 --> 00:16:28.160 validates the the the hypothesis of 00:16:28.160 --> 00:16:30.800 where you're going. Um, I'm not scared 00:16:30.800 --> 00:16:32.320 of competition. I think you can learn a 00:16:32.320 --> 00:16:34.240 lot competing for the same customers and 00:16:34.240 --> 00:16:36.240 seeing how decisions go. Uh, I think 00:16:36.240 --> 00:16:37.600 it's always good to be able to compare 00:16:37.600 --> 00:16:40.399 yourselves to competition. Um, I think 00:16:40.399 --> 00:16:44.079 there are areas in which Taktile's deep 00:16:44.079 --> 00:16:46.160 experience in decision-m remember this 00:16:46.160 --> 00:16:48.079 is a this company has been around for a 00:16:48.079 --> 00:16:49.920 while and we have a lot of C customers 00:16:49.920 --> 00:16:52.480 in production. I think that gives us a 00:16:52.480 -->

00:16:55.680 real uh compelling advantage over sort 00:16:55.680 --> 00:16:57.279 of newcomers. In particular, when I 00:16:57.279 --> 00:16:58.560 think about newcomers, I usually think 00:16:58.560 --> 00:17:01.279 about the AI startups who I think 00:17:01.279 --> 00:17:04.400 underestimate the complexity of the of 00:17:04.400 --> 00:17:06.559 business processes or the complex 00:17:06.559 --> 00:17:08.880 business needs of financial customers 00:17:08.880 --> 00:17:10.240 even though they're coming with great 00:17:10.240 --> 00:17:12.240 models like we have those models too. 00:17:12.240 --> 00:17:14.640 Um, but we also have years of context in 00:17:14.640 --> 00:17:16.640 the industry. I think that gives us a uh 00:17:16.640 --> 00:17:17.839 compelling advantage there. 00:17:17.839 --> 00:17:20.400 James, I really like the idea about how 00:17:20.400 --> 00:17:24.319 to prevent uh the bias. Um I believe if 00:17:24.319 --> 00:17:26.000 I'm sure at one point your customers 00:17:26.000 --> 00:17:28.480 will come to you and help to implement 00:17:28.480 --> 00:17:31.600 that and uh we really wish you good luck 00:17:31.600 --> 00:17:34.400 with that and good luck in your career 00:17:34.400 --> 00:17:36.720 at Taktile. Hopefully in a couple of 00:17:36.720 --> 00:17:38.400 years perhaps we will have you again on 00:17:38.400 --> 00:17:40.960 the show. Uh but today we really enjoy 00:17:40.960 --> 00:17:42.880 we really uh enjoyed the insights that 00:17:42.880 --> 00:17:44.240 you shared. Thank you for coming on the 00:17:44.240 --> 00:17:45.120 show. 00:17:45.120 --> 00:17:47.039 Great to be here. Cheers guys. 00:17:47.039 --> 00:17:47.919 Thank you James. 00:17:47.919 --> 00:17:49.679 All right everyone today we have Sadra 00:17:49.679 --> 00:17:51.919 Hosseini on the ground with us. Sadra is 00:17:51.919 --> 00:17:54.000 the co-founder and CEO of Ryft, a 00:17:54.000 --> 00:17:55.679 Manchester based payment company that 00:17:55.679 --> 00:17:58.160 helps marketplace and platforms take 00:17:58.160 --> 00:18:00.799 payments and split them between sellers. 00:18:00.799 --> 00:18:03.120 Last month, Lyft raised 20 million 00:18:03.120 --> 00:18:05.679 pounds. Sadra, welcome on the show and 00:18:05.679 --> 00:18:07.200 it's great to see you again after 00:18:07.200 --> 00:18:08.720 Fintech meet up Europe. 00:18:08.720 --> 00:18:10.160 Thank you very much for having me. I'm 00:18:10.160 --> 00:18:12.640 I'm excited to be here today. 00:18:12.640 --> 00:18:15.679 Sadra, uh, can you tell us um, a little 00:18:15.679 --> 00:18:17.600 bit about Ryft, what you guys are 00:18:17.600 --> 00:18:19.039 building and what problems you're 00:18:19.039 --> 00:18:19.919 solving? 00:18:19.919 --> 00:18:22.240 Of course. Of course. So, Ryft is the 00:18:22.240 --> 00:18:24.320 payments infrastructure for multi-party 00:18:24.320 --> 00:18:27.600 commerce. What that means is um any any 00:18:27.600 --> 00:18:29.440 business that needs split payments. Uh 00:18:29.440 --> 00:18:31.440 so, that's marketplaces as you said, 00:18:31.440 --> 00:18:35.039 digital platforms or franchises where 00:18:35.039 --> 00:18:36.960 I'll give you an example. Uh let's say 00:18:36.960 --> 00:18:39.919 Uber Eats, you payund100 for your 00:18:39.919 --> 00:18:41.919 dinner. That needs to be split three 00:18:41.919 --> 00:18:45.360 ways. So, £10 to driver, 20 to Uber, 00:18:45.360 --> 00:18:47.520 excuse me, £70 to a restaurant. That's 00:18:47.520 --> 00:18:49.440 where we sit. So, that's what I like to 00:18:49.440 --> 00:18:50.799 call commerce 2.0. You know, we're 00:18:50.799 --> 00:18:53.200 moving away from onetoone transactions 00:18:53.200 --> 00:18:55.039 where you go to Nike's website, you pay 00:18:55.039 --> 00:18:57.280 £100 for your shoes, and your payment 00:18:57.280 --> 00:18:59.120 processor would settle £100 into your 00:18:59.120 --> 00:19:02.559 account to now one to many and many to 00:19:02.559 --> 00:19:05.760 many. So that's that's where Ryft sits. 00:19:05.760 --> 00:19:09.039 Sadra, uh the story story goes that 00:19:09.039 --> 00:19:11.520 you've started Ryft because Butler, your 00:19:11.520 --> 00:19:13.600 pubing app, paid too much on Stripe 00:19:13.600 --> 00:19:17.520 fees. So uh what was expensive and when 00:19:17.520 --> 00:19:20.640 platforms move from Stripe to Ryft? Uh 00:19:20.640 --> 00:19:22.400 what is the main reason these days? 00:19:22.400 --> 00:19:24.960 Yes. Uh so Stripe is a very good kind of 00:19:24.960 --> 00:19:26.559 technical solution when you first start 00:19:26.559 --> 00:19:28.320 getting going. So if you're an early 00:19:28.320 --> 00:19:31.200 stage business, Stripe is is great. Uh 00:19:31.200 --> 00:19:33.039 and that's that's what we started. So we 00:19:33.039 --> 00:19:35.440 Bottler was a is a it was one app across 00:19:35.440 --> 00:19:36.720 all venues where you could order your 00:19:36.720 --> 00:19:39.120 food and drink within the venue. Uh we 00:19:39.120 --> 00:19:41.200 started that pre-COVID co hit. We scaled 00:19:41.200 --> 00:19:42.640 that to a million customers before it 00:19:42.640 --> 00:19:44.720 was acquired. And while we were building 00:19:44.720 --> 00:19:47.120 that business, we realized that once you 00:19:47.120 --> 00:19:49.520 go above 5 10 mil of volume, Stripe 00:19:49.520 --> 00:19:51.200 becomes a bottleneck for the platform 00:19:51.200 --> 00:19:54.400 itself. The reason for that is because 00:19:54.400 --> 00:19:56.240 Stripe has automated everything. They 00:19:56.240 --> 00:19:58.080 don't get to really understand these 00:19:58.080 --> 00:20:00.400 complex business models as as Butler was 00:20:00.400 --> 00:20:03.120 and as you see now kind of coming out 00:20:03.120 --> 00:20:06.640 with all these AI businesses. So because 00:20:06.640 --> 00:20:07.840 they don't get to understand the 00:20:07.840 --> 00:20:09.520 business, the underwriting and 00:20:09.520 --> 00:20:12.080 compliance will suddenly switch

off sub 00:20:12.080 --> 00:20:14.559 accounts. They'll pause payouts. They'll 00:20:14.559 --> 00:20:17.120 freeze accounts. So it becomes an 00:20:17.120 --> 00:20:19.120 operational nightmare for the platform 00:20:19.120 --> 00:20:21.919 itself. And then as you said there, 00:20:21.919 --> 00:20:24.880 Stripe is notoriously very expensive. Uh 00:20:24.880 --> 00:20:26.960 the reason there is they've got a lot of 00:20:26.960 --> 00:20:28.400 good merchants that they're monetizing 00:20:28.400 --> 00:20:29.679 but they also have a lot of bad 00:20:29.679 --> 00:20:31.120 merchants where they have to cover the 00:20:31.120 --> 00:20:32.320 chargebacks, they have to cover the 00:20:32.320 --> 00:20:35.039 fraud, uh the default rate and their 00:20:35.039 --> 00:20:37.120 support and operational cost is higher 00:20:37.120 --> 00:20:39.360 than if they went vertical by vertical. 00:20:39.360 --> 00:20:41.360 So that gets passed under the platform. 00:20:41.360 --> 00:20:43.280 The platform then can't monetize the 00:20:43.280 --> 00:20:45.520 transactions. That's where Ryft comes 00:20:45.520 --> 00:20:48.000 in. We we take care of all of that. We 00:20:48.000 --> 00:20:49.679 we really get to understand these 00:20:49.679 --> 00:20:52.400 business models and then we we allow the 00:20:52.400 --> 00:20:53.919 platform to monetize their own 00:20:53.919 --> 00:20:56.320 transactions in effect becoming their 00:20:56.320 --> 00:20:58.720 own stripe by using the Ryft Rails. 00:20:58.720 --> 00:21:01.440 As let's talk about the fundraising that 00:21:01.440 --> 00:21:04.480 you guys did um in one of our late shows 00:21:04.480 --> 00:21:07.679 we covered that UK fintech uh funding is 00:21:07.679 --> 00:21:09.520 at 10year low. I mean that's 00:21:09.520 --> 00:21:12.640 significant. So what did Graham House, 00:21:12.640 --> 00:21:14.400 one of your investors wanted to see 00:21:14.400 --> 00:21:17.120 before they invested in you actually? 00:21:17.120 --> 00:21:18.640 Yeah, a really good question. So, 00:21:18.640 --> 00:21:21.200 Gresham have led the round uh great 00:21:21.200 --> 00:21:23.200 investors. They've uh they kind of 00:21:23.200 --> 00:21:25.120 helped the business step into a full 00:21:25.120 --> 00:21:27.760 scaleup mode. Now, um Roit is the 00:21:27.760 --> 00:21:30.240 partner there who's uh invested in us. 00:21:30.240 --> 00:21:32.240 What they really cared about obviously 00:21:32.240 --> 00:21:35.440 there is the XC team like the CEO or the 00:21:35.440 --> 00:21:37.600 whole team around that how we how we 00:21:37.600 --> 00:21:39.840 operating, how well do we work together. 00:21:39.840 --> 00:21:42.320 One of the USPS of Ryft is the team 00:21:42.320 --> 00:21:44.640 itself that the whole team how we how we 00:21:44.640 --> 00:21:46.799 kind of move. We're very agile. were 00:21:46.799 --> 00:21:48.880 very adaptive. So they they were very 00:21:48.880 --> 00:21:50.720 happy with that. And off the back of 00:21:50.720 -->

00:21:52.640 that is obviously the traction as well. 00:21:52.640 --> 00:21:53.840 Like we we've been growing two and a 00:21:53.840 --> 00:21:55.840 half 3x year on year over the last five 00:21:55.840 --> 00:21:58.400 years. Um and the team has been 00:21:58.400 --> 00:22:00.960 outperforming uh any any of our closest 00:22:00.960 --> 00:22:03.200 competitors. So they they were very 00:22:03.200 --> 00:22:05.120 interested in Ryft. And then the final 00:22:05.120 --> 00:22:07.039 point to that is the payment 00:22:07.039 --> 00:22:10.400 sovereignty. So UK doesn't have its own 00:22:10.400 --> 00:22:13.679 sovereign payment uh platform. So you 00:22:13.679 --> 00:22:16.000 got Stripe which is US, you got um Adyen 00:22:16.000 --> 00:22:18.720 which is Netherlands. UK is missing its 00:22:18.720 --> 00:22:21.120 own version of these two businesses. So 00:22:21.120 --> 00:22:23.840 Ryft is the UK answer to Stripe and 00:22:23.840 --> 00:22:26.159 Aden. We're building the next gen of of 00:22:26.159 --> 00:22:28.400 UK payments growing globally. 00:22:28.400 --> 00:22:31.120 So 20 million um what's the plan with it 00:22:31.120 --> 00:22:32.559 now? What are they planning to spend it 00:22:32.559 --> 00:22:33.840 on first? 00:22:33.840 --> 00:22:35.280 Yeah, [laughter] 00:22:35.280 --> 00:22:37.280 I wish I could say Ferraris and Lambos, 00:22:37.280 --> 00:22:39.840 but yeah, it's going towards scaling the 00:22:39.840 --> 00:22:42.880 business. Uh the reason we raised was uh 00:22:42.880 --> 00:22:44.080 as I said we're we're growing 00:22:44.080 --> 00:22:45.760 internationally. So we're growing into 00:22:45.760 --> 00:22:48.799 the EU, growing into the US. Uh two 00:22:48.799 --> 00:22:51.360 reasons for that. So our customer base 00:22:51.360 --> 00:22:52.720 are growing. So they're growing 00:22:52.720 --> 00:22:54.320 internationally. So we we're growing as 00:22:54.320 --> 00:22:55.520 well to make sure that we can support 00:22:55.520 --> 00:22:58.880 them. But also Europe is where it's 00:22:58.880 --> 00:23:02.320 getting super exciting. Europe is um 00:23:02.320 --> 00:23:03.919 underpenetrated with these platforms, 00:23:03.919 --> 00:23:06.000 but it's growing 20% yearonear over the 00:23:06.000 --> 00:23:09.120 next five six years. uh and Europe is 00:23:09.120 --> 00:23:10.880 also very fractured. So every country 00:23:10.880 --> 00:23:12.240 has its own payment method. Every 00:23:12.240 --> 00:23:14.000 country has its own own language, own 00:23:14.000 --> 00:23:16.240 culture, but these these European 00:23:16.240 --> 00:23:18.400 businesses are growing into the whole 00:23:18.400 --> 00:23:20.559 Europe and then into the US as well. So 00:23:20.559 --> 00:23:22.960 we've raised this money to build that 00:23:22.960 --> 00:23:24.720 perfect solution for these European 00:23:24.720 --> 00:23:27.039 merchants who are then globing globally 00:23:27.039 --> 00:23:28.799 themselves. 00:23:28.799 --> 00:23:32.480 Sadra, look um in

April 2025, you told 00:23:32.480 --> 00:23:34.799 Business Insider that the Ryft uh has 00:23:34.799 --> 00:23:38.400 already broken even um and your but but 00:23:38.400 --> 00:23:41.919 your 2025 accounts at company's house 00:23:41.919 --> 00:23:46.080 show losses uh loses uh grew by 1.8 00:23:46.080 --> 00:23:48.240 million pounds. So are you guys 00:23:48.240 --> 00:23:49.760 profitable at the moment? 00:23:49.760 --> 00:23:52.320 No. So we uh we basically hit break even 00:23:52.320 --> 00:23:54.799 and then we raised our series A um and 00:23:54.799 --> 00:23:56.720 then since then we've been growing. The 00:23:56.720 --> 00:23:59.120 reason we raised series B was to 00:23:59.120 --> 00:24:01.600 obviously keep growing. Um, so we're 00:24:01.600 --> 00:24:03.520 still burning cash, but with the series 00:24:03.520 --> 00:24:06.400 B, we're going to hit profitability. Um, 00:24:06.400 --> 00:24:09.039 and if we go to raise again, it would be 00:24:09.039 --> 00:24:11.279 for M&A. There's a lot of opportunity in 00:24:11.279 --> 00:24:12.640 the market. There's a lot of 00:24:12.640 --> 00:24:14.159 consolidation happening as you've seen 00:24:14.159 --> 00:24:16.640 with World Pay and Global Payments. Um, 00:24:16.640 --> 00:24:18.320 I think a couple of other big providers 00:24:18.320 --> 00:24:20.640 are looking to emerge as well. So 00:24:20.640 --> 00:24:22.159 there's a there's an opportunity for 00:24:22.159 --> 00:24:24.720 Ryft to keep acquiring these businesses 00:24:24.720 --> 00:24:26.559 and kind of grow the customer base much 00:24:26.559 --> 00:24:29.520 more rapidly than organic growth. 00:24:29.520 --> 00:24:31.360 You you you already mentioned about 00:24:31.360 --> 00:24:32.880 international expansion. I think in the 00:24:32.880 --> 00:24:35.679 same business insider piece uh you said 00:24:35.679 --> 00:24:38.880 that Ryft would launch in US in 2025. 00:24:38.880 --> 00:24:40.559 When will US customers be able to use 00:24:40.559 --> 00:24:41.360 Ryft? 00:24:41.360 --> 00:24:43.840 Yeah, so we have uh US customers live 00:24:43.840 --> 00:24:46.480 already. We're basically building out 00:24:46.480 --> 00:24:48.559 more of the proposition. Uh the solution 00:24:48.559 --> 00:24:51.279 that we have right now is mostly online. 00:24:51.279 --> 00:24:54.400 Uh in in UK and Europe we've launched a 00:24:54.400 --> 00:24:56.559 card presence so omni channel so you can 00:24:56.559 --> 00:24:58.880 take payments online offline wherever 00:24:58.880 --> 00:25:00.640 the customer is. So we're looking to 00:25:00.640 --> 00:25:03.120 launch that in the US as well. Um and 00:25:03.120 --> 00:25:06.000 we're getting more embedded into the US 00:25:06.000 --> 00:25:07.760 ecosystem with the solution that we have 00:25:07.760 --> 00:25:10.400 there. So it's it's going to be more 00:25:10.400 --> 00:25:12.799 fledged out and and more featurerich 00:25:12.799 --> 00:25:14.960 rich product. But I think it's important 00:25:14.960 -->

00:25:18.159 to also note that we're not looking to 00:25:18.159 --> 00:25:20.799 capture a big market share in the US 00:25:20.799 --> 00:25:22.480 because even though US has a big 00:25:22.480 --> 00:25:25.279 penetration with platform 61%. It's also 00:25:25.279 --> 00:25:27.120 a very saturated market. You have the 00:25:27.120 --> 00:25:28.400 likes of Finex, you have the likes of 00:25:28.400 --> 00:25:30.240 Rainforest and obviously strip Stripe 00:25:30.240 --> 00:25:33.919 itself. Um so for us it's making sure 00:25:33.919 --> 00:25:36.480 that we as I said building that perfect 00:25:36.480 --> 00:25:38.320 solution for European merchants who then 00:25:38.320 --> 00:25:40.960 want to land in the US. So that's where 00:25:40.960 --> 00:25:42.880 we're going to we we are strongest at 00:25:42.880 --> 00:25:45.360 rather than just going purely US play. 00:25:45.360 --> 00:25:47.919 Sadra on our podcast one of the things 00:25:47.919 --> 00:25:50.159 that we usually do we cover latest tech 00:25:50.159 --> 00:25:52.400 news and we get the opinion of our 00:25:52.400 --> 00:25:54.480 guests uh on those news because we 00:25:54.480 --> 00:25:57.120 believe that they are um specialists in 00:25:57.120 --> 00:25:59.360 this. Um one of the latest news that 00:25:59.360 --> 00:26:01.520 caught our attention was Stripe uh 00:26:01.520 --> 00:26:04.000 acquiring Paraffin uh which lends money 00:26:04.000 --> 00:26:05.760 to small businesses through platforms 00:26:05.760 --> 00:26:08.480 like Door Dash. Since you are in a 00:26:08.480 --> 00:26:11.520 similar vertical, do you think uh Ryft 00:26:11.520 --> 00:26:13.679 will offer loans to sellers at one 00:26:13.679 --> 00:26:14.400 point? 00:26:14.400 --> 00:26:17.120 Yeah. So, we've uh we've been asked this 00:26:17.120 --> 00:26:19.600 multiple times over the years. The the 00:26:19.600 --> 00:26:20.720 reason that's a really interesting 00:26:20.720 --> 00:26:24.400 acquisition is Stripe as as we do, we 00:26:24.400 --> 00:26:27.039 have all the data on the on the sub 00:26:27.039 --> 00:26:28.640 account. So, the restaurants or the 00:26:28.640 --> 00:26:30.480 retailers, whoever it might be, we know 00:26:30.480 --> 00:26:32.400 how much volume they're processing. We 00:26:32.400 --> 00:26:35.360 know what's been bought and sold. So we 00:26:35.360 --> 00:26:38.080 can um allow them to have a pre-approved 00:26:38.080 --> 00:26:39.760 loan. They can click a button, money in 00:26:39.760 --> 00:26:41.279 their account, they can use that for 00:26:41.279 --> 00:26:42.960 cash flow. In the UK, it's called 00:26:42.960 --> 00:26:45.840 merchant cash advance. Um so we've 00:26:45.840 --> 00:26:47.760 already spoken to a few providers. We've 00:26:47.760 --> 00:26:50.240 got that kind of ready to go. It's just 00:26:50.240 --> 00:26:52.400 uh we have other priorities which is the 00:26:52.400 --> 00:26:54.080 international expansion before we bring 00:26:54.080 --> 00:26:56.320 a solution like that to market. 00:26:56.320 --> 00:26:57.840 Makes sense. 00:26:57.840 --> 00:27:00.720 And and other news is uh Mastercard who 00:27:00.720 --> 00:27:03.520 launched a service that payouts pays out 00:27:03.520 --> 00:27:05.120 uh both in normal money, fiat money and 00:27:05.120 --> 00:27:07.760 stable coins through one connector. Um 00:27:07.760 --> 00:27:09.679 we know that you guys have added 35 00:27:09.679 --> 00:27:11.840 currencies this month alone. Will you be 00:27:11.840 --> 00:27:15.720 offering stable coin payouts? 00:27:16.240 --> 00:27:18.400 Think me and Sam went through this uh on 00:27:18.400 --> 00:27:20.640 the panel as 00:27:20.640 --> 00:27:23.520 um there is a lot of buzz but we're not 00:27:23.520 --> 00:27:26.799 seeing requests. So obviously we're 00:27:26.799 --> 00:27:28.480 customer led. We make sure that whatever 00:27:28.480 --> 00:27:30.480 we're building is the perfect solution 00:27:30.480 --> 00:27:31.919 for our customers that we're working 00:27:31.919 --> 00:27:34.640 with. So we haven't really seen that 00:27:34.640 --> 00:27:37.039 much request from our customer base for 00:27:37.039 --> 00:27:39.360 stable coins. We're trying to figure out 00:27:39.360 --> 00:27:41.360 exactly where it fits because right now 00:27:41.360 --> 00:27:44.720 it's more um it might be beneficial for 00:27:44.720 --> 00:27:47.840 kind of crossborder uh money flows but 00:27:47.840 --> 00:27:49.520 then the problem there is the on-ramp 00:27:49.520 --> 00:27:51.600 and the offramp. You're still paying for 00:27:51.600 --> 00:27:52.880 those fees. It might be a little bit 00:27:52.880 --> 00:27:54.159 faster but you're still paying for those 00:27:54.159 --> 00:27:57.120 fees. So, we're we're holding we're 00:27:57.120 --> 00:27:59.440 holding fire. Uh we're going to see what 00:27:59.440 --> 00:28:00.799 happens with that market. I think what's 00:28:00.799 --> 00:28:03.279 more exciting is agentic payments, agent 00:28:03.279 --> 00:28:05.760 um agent eiccommerce. We're starting to 00:28:05.760 --> 00:28:07.760 see platforms kind of start requesting 00:28:07.760 --> 00:28:10.399 that, pushing for that, and that's where 00:28:10.399 --> 00:28:11.919 we're going to be focusing a lot of our 00:28:11.919 --> 00:28:14.240 energy over the next couple years. 00:28:14.240 --> 00:28:17.440 Right. Look um Sadra on our show uh when 00:28:17.440 --> 00:28:19.600 we speak to different executives from uh 00:28:19.600 --> 00:28:21.200 payments and fintech vertical they say 00:28:21.200 --> 00:28:23.279 one of the modes is to hold their 00:28:23.279 --> 00:28:26.559 license but to to hold their own license 00:28:26.559 --> 00:28:29.120 but flat pay in a Danish payment company 00:28:29.120 --> 00:28:31.120 chose to use adance license instead of 00:28:31.120 --> 00:28:33.840 holding its own you know and what what 00:28:33.840 --> 00:28:35.760 I'm trying to say that you I see that 00:28:35.760 --> 00:28:38.399 you are going uh after your own

license 00:28:38.399 --> 00:28:40.399 so do you want to have advantage on this 00:28:40.399 --> 00:28:42.799 and will continue applying for more and 00:28:42.799 --> 00:28:45.600 more licenses on different agents. 00:28:45.600 --> 00:28:49.039 Yeah. So, we definitely see that it's 00:28:49.039 --> 00:28:49.919 for [clears throat] us, for our 00:28:49.919 --> 00:28:51.679 business, it's fundamental for us to 00:28:51.679 --> 00:28:53.120 have our own license because if you 00:28:53.120 --> 00:28:54.880 don't have your own license, you're 00:28:54.880 --> 00:28:57.120 reliant on someone else like Aden, but 00:28:57.120 --> 00:28:59.200 also you're reliant on what their risk 00:28:59.200 --> 00:29:02.399 appetite is, how fast they can move. For 00:29:02.399 --> 00:29:04.880 example, when we on board a a platform, 00:29:04.880 --> 00:29:07.360 they might have 50,000 sub accounts. We 00:29:07.360 --> 00:29:09.120 can't rely on a partner to on board 00:29:09.120 --> 00:29:10.880 those uh because they'll they'll break 00:29:10.880 --> 00:29:12.480 down. We built all that all that 00:29:12.480 --> 00:29:14.720 in-house ourselves. So we need to have 00:29:14.720 --> 00:29:16.320 our own license, we have to have our own 00:29:16.320 --> 00:29:18.880 risk decision matrix to be able to on 00:29:18.880 --> 00:29:20.559 board these platforms and the sub 00:29:20.559 --> 00:29:22.960 accounts itself and then obviously if 00:29:22.960 --> 00:29:24.880 you have the license you build the tech 00:29:24.880 --> 00:29:26.880 as well. It gives you a bigger mode and 00:29:26.880 --> 00:29:28.720 it gives you a high margin as well. You 00:29:28.720 --> 00:29:30.159 don't have to rely on a partner to do 00:29:30.159 --> 00:29:32.480 everything. So for us it's fundamental 00:29:32.480 --> 00:29:34.960 that we have our own licenses. 00:29:34.960 --> 00:29:36.000 Makes sense. 00:29:36.000 --> 00:29:37.840 Well that was insightful Sadra. Thank 00:29:37.840 --> 00:29:39.120 you very much for coming. 00:29:39.120 --> 00:29:40.799 Thank you so much for having me. Really 00:29:40.799 --> 00:29:42.320 enjoyed the chat. 00:29:42.320 --> 00:29:44.320 Likewise, Sadra, we'll definitely stay 00:29:44.320 --> 00:29:46.320 in touch. May perhaps uh we'll have a 00:29:46.320 --> 00:29:49.039 chance to meet here in London. Uh if you 00:29:49.039 --> 00:29:51.279 are around uh there are many events are 00:29:51.279 --> 00:29:52.640 happening. Looking forward to it. 00:29:52.640 --> 00:29:54.799 Sounds good to me. Cheers, guys. 00:29:54.799 --> 00:29:55.360 Cheers. 00:29:55.360 --> 00:29:56.399 Bye-bye. 00:29:56.399 --> 00:29:59.200 All right, guys. Let's get back to um 00:29:59.200 --> 00:30:02.080 today's uh two breaking stories that we 00:30:02.080 --> 00:30:04.640 brought to you. First one is Barclays is 00:30:04.640 --> 00:30:07.840 cutting 5,000 jobs and it says AI is 00:30:07.840 --> 00:30:11.360 part of the reason. I thought AI um was 00:30:11.360 --> 00:30:13.600 uh bringing a lot of efficiency good 00:30:13.600 --> 00:30:15.200 things but we have to understand that 00:30:15.200 --> 00:30:17.600 with more efficiency there might be job 00:30:17.600 --> 00:30:19.600 cuts but it doesn't mean that every 00:30:19.600 --> 00:30:22.159 company is cutting the jobs only because 00:30:22.159 --> 00:30:25.200 of AI uh the may have some um role in 00:30:25.200 --> 00:30:27.520 this will go deeper into this and AI 00:30:27.520 --> 00:30:31.039 helped to find a security hole in PHP 00:30:31.039 --> 00:30:33.360 the code that runs the biggest part of 00:30:33.360 --> 00:30:35.919 the web and hackers start using it 00:30:35.919 --> 00:30:37.760 within the hours autobake so let's start 00:30:37.760 --> 00:30:40.080 with the first news sparkly is cutting 00:30:40.080 --> 00:30:43.520 about 5,000 jobs and around 5% of its 00:30:43.520 --> 00:30:45.760 staff worldwide. It announced the cuts 00:30:45.760 --> 00:30:48.559 on Thursday with its third quarter 00:30:48.559 --> 00:30:51.120 results. Uh about 1,000 jobs have 00:30:51.120 --> 00:30:53.679 already gone this year and uh the other 00:30:53.679 --> 00:30:57.120 4,000 will be gone by the end of 2026. 00:30:57.120 --> 00:30:59.600 Most are back office and support and the 00:30:59.600 --> 00:31:03.440 BBC says many cuts will fall in India. 00:31:03.440 --> 00:31:05.279 Ah, I believe this will hit the Indian 00:31:05.279 --> 00:31:07.279 economy really bad because those guys 00:31:07.279 --> 00:31:10.559 are uh India is hub for support jobs 00:31:10.559 --> 00:31:13.360 particularly and Barclay's results 00:31:13.360 --> 00:31:15.360 presentation talks about efficiency 00:31:15.360 --> 00:31:18.000 driven by AI and the banks expect to 00:31:18.000 --> 00:31:21.600 save more than 1.1 billion pounds in 00:31:21.600 --> 00:31:26.000 2027. Otabek is um this uh really AI or 00:31:26.000 --> 00:31:29.120 just cost cutting with AI label on it 00:31:29.120 --> 00:31:31.520 which is the question we ask so many 00:31:31.520 --> 00:31:33.600 times uh in our on our show. 00:31:33.600 --> 00:31:35.760 Well, I think it's partly AI. Yes. Uh 00:31:35.760 --> 00:31:37.360 many companies are seeing efficiency. 00:31:37.360 --> 00:31:41.360 The CEO of Barlay is uh Wata Krishnan on 00:31:41.360 --> 00:31:43.679 the call said AI is the icing on the 00:31:43.679 --> 00:31:46.159 cake but it's not the actual cake. The 00:31:46.159 --> 00:31:49.519 CFO and Cross said the cuts relate both 00:31:49.519 --> 00:31:51.200 to the simplification of the 00:31:51.200 --> 00:31:53.279 organization and to some of the impact 00:31:53.279 --> 00:31:56.399 that they were starting to see from AI. 00:31:56.399 --> 00:32:00.000 If you remember when we talked about the 00:32:00.000 --> 00:32:02.480 job cutting in Pipal and Coinbase uh 00:32:02.480 --> 00:32:05.519 back in May, we called it as you said uh 00:32:05.519 --> 00:32:08.480 AI whitewashing. But Barface has been 00:32:08.480 --> 00:32:11.360 going through the uh cost

optimization 00:32:11.360 --> 00:32:12.960 or cost cutting plan for the last three 00:32:12.960 --> 00:32:16.000 years. And um this is kind of a tale of 00:32:16.000 --> 00:32:17.600 it the last last year of the cost 00:32:17.600 --> 00:32:20.559 cutting program and it came out at the 00:32:20.559 --> 00:32:23.360 same time when they missed the profit 00:32:23.360 --> 00:32:26.080 target. The market was expecting them to 00:32:26.080 --> 00:32:29.519 make 2.4 billion uh pounds profit in 00:32:29.519 --> 00:32:33.919 this quarter. They did 2.3 billion. So 00:32:33.919 --> 00:32:36.480 again AI whitewashing or excuses for not 00:32:36.480 --> 00:32:38.080 missing the target that's that's a 00:32:38.080 --> 00:32:40.480 question that I think many analysts are 00:32:40.480 --> 00:32:44.559 are asking. However, Barclays do use AI 00:32:44.559 --> 00:32:47.440 in uh real business cases. Half of their 00:32:47.440 --> 00:32:50.000 developers are expected to move to 00:32:50.000 --> 00:32:51.919 Antropics quote code by the end of the 00:32:51.919 --> 00:32:55.279 year and uh they say they are already 00:32:55.279 --> 00:32:58.320 handling more than 120,000 client emails 00:32:58.320 --> 00:33:01.519 per day in their markets business. Who's 00:33:01.519 --> 00:33:04.000 actually losing the job? Do you see any 00:33:04.000 --> 00:33:06.799 push back? Uh are I mean employees who 00:33:06.799 --> 00:33:08.399 lost the job saying like you know this 00:33:08.399 --> 00:33:10.720 is AI washing um actually there's no 00:33:10.720 --> 00:33:12.240 efficiency. What's going on there? 00:33:12.240 --> 00:33:15.360 Well, we are talking about jobs in UK 00:33:15.360 --> 00:33:18.880 some and unions here are quite strong. 00:33:18.880 --> 00:33:21.120 We know that 60% of the cuts are coming 00:33:21.120 --> 00:33:23.760 to their support team and some of them 00:33:23.760 --> 00:33:27.279 are uh software developers. Unite which 00:33:27.279 --> 00:33:29.840 is the labor union in the UK said their 00:33:29.840 --> 00:33:32.080 general secretary Sharon Graham said the 00:33:32.080 --> 00:33:34.480 company's using AI as an excuse to disc 00:33:34.480 --> 00:33:37.200 to discard its loyal and long-suffering 00:33:37.200 --> 00:33:39.039 work workforce. And it's not just 00:33:39.039 --> 00:33:41.519 Barclays. Other UK banks are cutting 00:33:41.519 --> 00:33:44.000 jobs and citing AI for it as well. 00:33:44.000 --> 00:33:46.159 Standard Chartered announced that they 00:33:46.159 --> 00:33:51.440 are cutting 7,800 cuts jobs by 2030 and 00:33:51.440 --> 00:33:54.880 Lloyds are planning to cut 2 billion 00:33:54.880 --> 00:33:57.919 pounds by 2030 as well. Back in May 00:33:57.919 --> 00:34:01.039 again uh we discussed about uh JP Morgan 00:34:01.039 --> 00:34:04.720 CEO Diamond saying more AI people more 00:34:04.720 --> 00:34:07.600 AI will be replacing uh the bankers and 00:34:07.600 --> 00:34:10.720 we said if you remember if the junior 00:34:10.720 --> 00:34:13.520 roles

will be cut who will become the 00:34:13.520 --> 00:34:16.560 seniors in many years to come. Uh will 00:34:16.560 --> 00:34:19.520 hold the analyst from Quilty Chevad said 00:34:19.520 --> 00:34:22.000 the cats show how significantly AI and 00:34:22.000 --> 00:34:24.000 automation could reshape the banking 00:34:24.000 --> 00:34:26.320 industry over the next few years. I 00:34:26.320 --> 00:34:29.440 would really like to know again if these 00:34:29.440 --> 00:34:33.119 job cutings keep happening who will 00:34:33.119 --> 00:34:35.839 replace the junior roles and how they 00:34:35.839 --> 00:34:38.560 will plan to fill the senior roles in 00:34:38.560 --> 00:34:41.200 the future. I think you know what this 00:34:41.200 --> 00:34:44.079 was one of the questions uh which the 00:34:44.079 --> 00:34:48.240 audience asked um Jensen at Nvidia GTC. 00:34:48.240 --> 00:34:50.800 was almost the same question. Um and he 00:34:50.800 --> 00:34:54.079 said like yes uh we see there is some 00:34:54.079 --> 00:34:57.280 job cuts. Um there is a cycle I would 00:34:57.280 --> 00:35:00.640 say or job replacement but the thing is 00:35:00.640 --> 00:35:03.440 AI is creating more jobs too. We have to 00:35:03.440 --> 00:35:05.920 understand that and what needs to happen 00:35:05.920 --> 00:35:08.480 now. Uh this is I think simply the flow. 00:35:08.480 --> 00:35:11.760 Um it happened historically when some 00:35:11.760 --> 00:35:14.000 breakthrough happens some jobs will be 00:35:14.000 --> 00:35:16.079 lost but some more will be created which 00:35:16.079 --> 00:35:17.839 means the people who lost their job they 00:35:17.839 --> 00:35:21.040 will need to retrain themselves and uh 00:35:21.040 --> 00:35:23.920 learn more AI skills and then maybe go 00:35:23.920 --> 00:35:26.960 and find some other jobs that that that 00:35:26.960 --> 00:35:29.680 would fit their new skills uh that they 00:35:29.680 --> 00:35:31.839 um obtained. But again I would love to 00:35:31.839 --> 00:35:34.320 hear what our audience uh thinks about 00:35:34.320 --> 00:35:37.440 uh this topic uh if they are worried 00:35:37.440 --> 00:35:39.920 that AI will come after their job and 00:35:39.920 --> 00:35:43.119 they soon may lose um the job that they 00:35:43.119 --> 00:35:45.680 are doing and how they are actually 00:35:45.680 --> 00:35:48.720 preparing for machine economy for AI 00:35:48.720 --> 00:35:50.960 economy what exactly uh they are 00:35:50.960 --> 00:35:53.200 learning what skills they are obtaining 00:35:53.200 --> 00:35:54.880 please uh leave your thoughts in the 00:35:54.880 --> 00:35:56.640 comment section 00:35:56.640 --> 00:35:59.200 on to the next news uh well we discuss 00:35:59.200 --> 00:36:01.680 AI cutting shops at Barclays. Uh this 00:36:01.680 --> 00:36:04.079 week it is also found some security 00:36:04.079 --> 00:36:06.560 holes in big parts of the internet. On 00:36:06.560 --> 00:36:09.359 Wednesday, PHP released an emergency fix 00:36:09.359 --> 00:36:11.839 for two critical flaws called nicknamed 00:36:11.839 --> 00:36:15.599 PHP2Shell. PHP some as you know uh is a 00:36:15.599 --> 00:36:18.000 programming language and majority of the 00:36:18.000 --> 00:36:20.240 website on internet run on it. Uh 00:36:20.240 --> 00:36:23.200 WordPress and many other online shops. 00:36:23.200 --> 00:36:27.599 W3 tech says that about 70% 70% of the 00:36:27.599 --> 00:36:30.240 sites where it can detect the language 00:36:30.240 --> 00:36:34.079 uh run on PHP. These flaws are kind of 00:36:34.079 --> 00:36:37.359 nicknamed PHP2Shell and they let an 00:36:37.359 --> 00:36:39.599 attacker to take control of web server 00:36:39.599 --> 00:36:42.240 from anywhere. A security firm called 00:36:42.240 --> 00:36:45.359 Calif uh working on OpenAI program and 00:36:45.359 --> 00:36:47.599 the cloud security company V found them 00:36:47.599 --> 00:36:50.000 uh with the help of AI models. the 00:36:50.000 --> 00:36:52.240 details were leaked and before anyone 00:36:52.240 --> 00:36:54.560 could fix them the obviously uh 00:36:54.560 --> 00:36:57.680 attackers uh started uh to take over the 00:36:57.680 --> 00:36:59.440 websites and attacks happening within 00:36:59.440 --> 00:37:01.680 the hours. The UK's National Cyber 00:37:01.680 --> 00:37:03.920 Security Center said organizations using 00:37:03.920 --> 00:37:07.599 any version of PHP are affected. So Sam, 00:37:07.599 --> 00:37:09.440 I'm not sure if you have experienced any 00:37:09.440 --> 00:37:11.200 issues using the websites but I have 00:37:11.200 --> 00:37:13.599 seen uh many government and private 00:37:13.599 --> 00:37:16.160 websites showing the warnings or not 00:37:16.160 --> 00:37:19.520 working uh this week. So it is kind of 00:37:19.520 --> 00:37:22.400 uh concerning. So I'm not quite sure 00:37:22.400 --> 00:37:24.800 what to make of it. Uh AI finding the 00:37:24.800 --> 00:37:27.599 bugs and showing the deficiencies from 00:37:27.599 --> 00:37:29.760 one side it seems like a good thing. 00:37:29.760 --> 00:37:33.119 People are patching up their uh leaky uh 00:37:33.119 --> 00:37:35.200 kind of codes. But on the other side 00:37:35.200 --> 00:37:37.040 obviously all these hacks and takeover. 00:37:37.040 --> 00:37:38.880 So I'm not sure if it's a good news or 00:37:38.880 --> 00:37:41.599 bad news that AI is doing all this. I 00:37:41.599 --> 00:37:43.520 mean this is I think the still open 00:37:43.520 --> 00:37:46.800 question as you said um AI trying to 00:37:46.800 --> 00:37:48.640 help um let's say cyber security 00:37:48.640 --> 00:37:50.480 professionals to find the holes but on 00:37:50.480 --> 00:37:52.720 the other hand the same hackers are 00:37:52.720 --> 00:37:54.960 using those holes again to go and break 00:37:54.960 --> 00:37:57.119 into websites I didn't come across any 00:37:57.119 --> 00:37:59.920 uh warnings just yet but it was all over 00:37:59.920 --> 00:38:02.400 on my news feed that saying like a lot 00:38:02.400 --> 00:38:04.640 of

sites were taken down and the 00:38:04.640 --> 00:38:06.720 companies are working hard on patching 00:38:06.720 --> 00:38:10.079 all of those bugs and issues. So the bug 00:38:10.079 --> 00:38:12.160 actually autobex had in PHP for more 00:38:12.160 --> 00:38:13.920 than 10 years. It's not that something 00:38:13.920 --> 00:38:15.680 that happened yesterday. It was there 00:38:15.680 --> 00:38:18.400 for 10 years and AI helped actually to 00:38:18.400 --> 00:38:20.560 find and I think bring it to the 00:38:20.560 --> 00:38:22.480 attention. It usually happens. I think 00:38:22.480 --> 00:38:25.040 one of the best things about AI is that 00:38:25.040 --> 00:38:27.200 since uh sometimes the things that we 00:38:27.200 --> 00:38:29.040 are doing are too large and we are not 00:38:29.040 --> 00:38:31.280 able to pay attention to every and each 00:38:31.280 --> 00:38:33.200 detail of the things and AI could bring 00:38:33.200 --> 00:38:35.760 that to our attention and in this case I 00:38:35.760 --> 00:38:38.320 think this is what exactly happens and 00:38:38.320 --> 00:38:41.200 the thing is um attackers can use um the 00:38:41.200 --> 00:38:43.440 same models again as we said to hack 00:38:43.440 --> 00:38:46.480 into the websites we u wrote and I'm 00:38:46.480 --> 00:38:49.119 quoting frontier AI model substantially 00:38:49.119 --> 00:38:51.280 reduce the barrier turning a 00:38:51.280 --> 00:38:53.760 long-standing flow into an exploit. An 00:38:53.760 --> 00:38:56.160 exploit that is working at code that 00:38:56.160 --> 00:38:59.440 attacks the bug. TJ sireers of the 00:38:59.440 --> 00:39:01.680 center of internet security told fortune 00:39:01.680 --> 00:39:04.320 PHP2Shell shows how AI can help 00:39:04.320 --> 00:39:07.440 uncover um buried flows and turn complex 00:39:07.440 --> 00:39:09.920 exploit chains into practical attacks. 00:39:09.920 --> 00:39:12.640 What worries me the most is that Calif 00:39:12.640 --> 00:39:15.119 saying it didn't need special access. um 00:39:15.119 --> 00:39:18.160 it um its words uh were no special 00:39:18.160 --> 00:39:21.119 harness or access uh to models that were 00:39:21.119 --> 00:39:23.200 uh that weren't generally available to 00:39:23.200 --> 00:39:25.359 anyone were used. So basically they use 00:39:25.359 --> 00:39:28.480 the same available tools to everyone I 00:39:28.480 --> 00:39:30.800 think uh from their words. So uh these 00:39:30.800 --> 00:39:32.960 are the tools uh that anyone can use on 00:39:32.960 --> 00:39:35.520 September 7 autobec we covered that open 00:39:35.520 --> 00:39:39.359 AI chip 6 Astra actually reaching its 00:39:39.359 --> 00:39:42.000 critical cyber level and I I said I 00:39:42.000 --> 00:39:45.359 remember that that banks share the same 00:39:45.359 --> 00:39:47.839 open uh source code and I mean it's 00:39:47.839 --> 00:39:49.760 obvious you know there are couple of 00:39:49.760 --> 00:39:52.320 major um players uh banking software 00:39:52.320 --> 00:39:54.960 players uh that are available out there 00:39:54.960 --> 00:39:58.880 and um if one goes down one of those 00:39:58.880 --> 00:40:01.920 software providers It means number of 00:40:01.920 --> 00:40:03.520 banks will be affected. It's not that 00:40:03.520 --> 00:40:06.000 the one bank or fintech provider will be 00:40:06.000 --> 00:40:08.640 affected. We should expose major like 00:40:08.640 --> 00:40:11.440 who who is exposed uh who are those uh 00:40:11.440 --> 00:40:14.400 under the risk and what does it mean to 00:40:14.400 --> 00:40:16.079 uh for the banks for the payments or 00:40:16.079 --> 00:40:17.760 other players in our industry that we 00:40:17.760 --> 00:40:18.880 cover. 00:40:18.880 --> 00:40:20.640 Let's start with the major names. Uh 00:40:20.640 --> 00:40:22.960 some of them are WooCommerce uh that 00:40:22.960 --> 00:40:26.320 runs on WordPress, Magento which uh now 00:40:26.320 --> 00:40:29.280 Adobe Commerce. uh those guys are all 00:40:29.280 --> 00:40:31.680 built on PHP. These are already one of 00:40:31.680 --> 00:40:34.000 the biggest um platforms in the world 00:40:34.000 --> 00:40:36.320 and people actually built on top of 00:40:36.320 --> 00:40:40.320 these platforms uh their shops uh and 00:40:40.320 --> 00:40:41.920 they of course connect the payment 00:40:41.920 --> 00:40:44.000 systems to those shops and then 00:40:44.000 --> 00:40:46.560 obviously this can and those uh shops 00:40:46.560 --> 00:40:49.280 can get easily exposed and they attacks 00:40:49.280 --> 00:40:52.160 came fast autoic after the actually the 00:40:52.160 --> 00:40:54.560 incident started. Cloudflare's CEO 00:40:54.560 --> 00:40:57.040 Matthew Prince told Fortune there were 00:40:57.040 --> 00:40:59.839 tens of thousands of attacks attempts in 00:40:59.839 --> 00:41:02.720 the first 45 minutes in the first 45 00:41:02.720 --> 00:41:05.440 minutes and millions by that evening. I 00:41:05.440 --> 00:41:07.359 mean that the numbers are staggering and 00:41:07.359 --> 00:41:10.720 it's huge and help security reports that 00:41:10.720 --> 00:41:13.680 Amazon sensor saw one attempt 4 days 00:41:13.680 --> 00:41:16.079 before the fix the systems that could be 00:41:16.079 --> 00:41:19.680 at risk. You see it happened even before 00:41:19.680 --> 00:41:22.240 the news were released that that there 00:41:22.240 --> 00:41:24.800 is a hole. Someone already notes that 00:41:24.800 --> 00:41:27.359 and started there and not in a good way. 00:41:27.359 --> 00:41:29.040 Obviously it seems like they start 00:41:29.040 --> 00:41:31.599 attacking which means I mean uh hackers 00:41:31.599 --> 00:41:35.599 are um ahead of white blackhead hackers 00:41:35.599 --> 00:41:37.359 I would say are ahead of white hat 00:41:37.359 --> 00:41:41.280 hackers if you can say so and if AI can 00:41:41.280 --> 00:41:43.680 find the bugs uh that's been there for 00:41:43.680 --> 00:41:45.839 10 years and I want to know how many 00:41:45.839 --> 00:41:48.480 more out there it's very interesting and 00:41:48.480 --> 00:41:51.280 important question and as AI keeps 00:41:51.280 --> 00:41:53.920 developing we'll definitely find out 00:41:53.920 --> 00:41:56.480 what's going to happen next of today 00:41:56.480 --> 00:41:59.200 indeed I It will get worse until it gets 00:41:59.200 --> 00:42:02.560 better. Uh AI is improving efficiency 00:42:02.560 --> 00:42:06.000 throughout uh in jobs in coding as well. 00:42:06.000 --> 00:42:09.119 But obviously until we reach that 00:42:09.119 --> 00:42:11.839 efficiency, we are vulnerable and uh 00:42:11.839 --> 00:42:16.960 many ill people will uh try to use that 00:42:16.960 --> 00:42:19.520 vulnerability. Uh again the question to 00:42:19.520 --> 00:42:22.000 our viewers uh does AI make the internet 00:42:22.000 --> 00:42:24.319 safer or more dangerous? Uh and if you 00:42:24.319 --> 00:42:25.920 run a website, have you already patched 00:42:25.920 --> 00:42:28.480 it? write it in the comments. 00:42:28.480 --> 00:42:30.319 Yeah, definitely. We would love to learn 00:42:30.319 --> 00:42:32.720 more what you think where we are going 00:42:32.720 --> 00:42:35.920 in this world of AI. But uh from the 00:42:35.920 --> 00:42:38.560 talks at NVITA GTC, the world seems to 00:42:38.560 --> 00:42:40.480 be bright and we are going towards 00:42:40.480 --> 00:42:43.280 bright the future. But these news show 00:42:43.280 --> 00:42:46.880 that there is some dark sides of uh AI 00:42:46.880 --> 00:42:48.960 and how it could be used. Leave your 00:42:48.960 --> 00:42:50.960 thoughts in the comments and that was on 00:42:50.960 --> 00:42:53.040 the ground. Stay tuned for the next 00:42:53.040 --> 00:42:56.800 episode next Monday.