What does AI mean for your job?

Joseph Marks [J.M]: So I thought we could begin this one by polling the audience: who here has used AI for their job today? Or, if you got up too early to come here today, yesterday? Just a show of hands.

A third. Maybe half. Half to a third. 

Okay, and how many would say that AI has fundamentally changed the way their workday or their workweek looks? What's the show of hands here? We got one, three. Okay, good. It'll be a better panel. 

So let's ask basically that same question. We'll start with you [Niles Bachtler] here.

So what does your workday or your workweek look like now, and how is that different from what it was a couple of years ago?

Nils Bachtler [N.B]: I don't think that my workweek in general looks significantly different. But what has changed is the way we operate, the way we move, how fast we can take decisions, how we delegate work, what to expect from the team. All in all, I think the workweek is still as busy and tiring as it was. 

But for us, as Klivvr, we're still a very young company, and being in the credit business, we've been using AI from day one, so nothing has really changed significantly, other than that we're very, very excited [about] how fast we can move, how fast we can make decisions. Actually, we're making much better decisions than we used to in the past, and I think it just helps us accelerate our business faster — so grow faster and actually perform much, much better.

It also, I think, has a huge impact on our team, on their productivity and on their future. So all in all, I think the impact is huge, and it's been helping us a lot.

J.M: Nader, what does your workday or your workweek look like now?

Nader Soliman [N.S]: For me, the biggest change that I've noticed is that I spend significantly less time gathering information, collecting information, collating information, and significantly more time interrogating that information. 

Five years ago, a big part of my week was reading reports, looking at market data, looking at market insights, preparing for meetings, for client meetings, for product decks. Today, AI has reduced that cycle significantly, and it has allowed me a lot more time to actually focus on what really matters from a leadership perspective: applying judgment, making better business decisions, comparing different strategies, spending more time with clients, spending more time innovating our products to help our clients achieve what they're looking for.

And for us, and for me specifically, I don't look at AI as a decision maker, but rather as a force multiplier. Basically, it helps me get to that first 70% much faster. But at the end of the day, the ultimate outcome is still my accountability and my responsibility.

J.M: Youssef?

Youssef Rizk [Y.R]: I've been on two kind of opposite ends of the spectrum on how to use AI. On one hand, I'm the CTO of Wondercraft, which is a startup backed by YC [Y Combinator] and [we’ve got a] video studio, and there, by virtue of being a very young company, you can completely abandon how things used to be done and reinvent them from scratch.

So our entire company runs on AI. Our engineers spend their entire day on Claude and Codex building software, which is easy enough in software, bugs automatically come in, fixes are filed, we review them, but not to the extent that anyone used to previously. Our marketing is done with AI, and I think the real value here is just the unlock in experimentation that you can do in that regard. Our SEO is done the same way; sales is done that way. And so from that perspective, AI has basically transformed the company: what you used to need 10 people to do, you can now realistically do with around two to three people who fully own the planning of the function that they're in.

On the other hand, we work with Windfall, where we help companies implement AI solutions. And at that level, I've mostly seen AI usage, which I would say is 90% of how people use it day to day, limited to what I would call “read-only applications”, where, as Nader [Soliman] said, you are able to just connect your data. And I even think that's a very advanced use case. I think most use cases are "summarize this email, translate this email and send it on." And I think that's a huge accelerant, because it fundamentally changes what you used to spend your time on. 

But I do think that the thing that fundamentally makes AI transformative is if you really think about it in this framework: this AI system is intelligence plus sensors plus actuators.

So you need sensors, which is the data, basically how the AI reads the state of the world, and then actuators, which are ways for it to interact with the state of the world. And once you have that, you're able to create this loop that either runs by itself or with some mild interrogation from someone sitting above it, to actually notice a very material impact on how your business runs.

J.M: That's an interesting distinction between how AI is used in one firm versus the other. 

How do you get from number two to number one? I just want to throw that out to the rest of the group. 

At what pace do you think the average organization in Egypt is going to get to where Wondercraft is now, and what are the pitfalls they have to look out for?

Y.R: I'm happy to start. I think Wondercraft has a huge benefit in being a young company. We started at the start of AI, so we're very AI-native, kind of by construction. I think it's very hard to take an existing business and flip it, effectively do that transformation, because they're not all set up that way.

Wondercraft is a software company, so it's quite easy to do that. I think if you're a construction company, or something in financial services, especially in Egypt, where the data literacy isn't as high as it is perhaps in the Western world, it's a lot harder to do that. So I do think that the way to start is to at least build these read-only views, which are already a huge accelerant, because you're able to actually understand the way your business works a lot better and derive insights that you would previously have had to hire people for, to go build dashboards for you. It's error-prone and so on. But now you can just declaratively say, "Hey, I want that." I think once you have that, you've effectively built two-thirds of what you need to get to this full loop. Now, the last thing you need is to give it the arms and legs to actually interact with the world.

J.M: Yeah, Nader, you can't move as quickly as Youssef can, right?

N.S: Actually, we are. We always look at Visa as the very first fintech established. 

For us, AI is not a new topic. We've been using AI for more than 30 years. Our first machine learning model was deployed in the '90s. In the past decade, we've invested more than USD 3 billion in AI and data, so it's not a new topic to us.

And in my opinion, AI literacy is as important as digital literacy is today. At Visa, we have access to almost every agent you can think of out there. But it's not just about the access. It's not just about the technology. It's about building the literacy, teaching people how to use it. As Youssef [Rizk] was saying, a lot of people just use it as a chat: "Summarize this email," or "Summarize this report," or "Send me this email." But that's not the benefit of AI. AI should be used for much more than that. 

And at Visa, we have a big focus on that. We have more than 1,200 AI champions across the world who help all our employees learn how to use AI, the different use cases and applications of AI. We have a lot of demos, we have a lot of hackathons.

And it's not just about AI or using AI, right? That's not the objective. You need to really understand what the problem is, or what you're trying to solve for, first, and then how AI can be used and leveraged to help you on that. And then you can start working on it. The objective is not to use AI. The objective at the end of the day is to deliver a better product or a better outcome much quicker, much faster, more efficiently. And to allow you time to apply your human judgment, your empathy, to be able to deliver these better results in a much shorter time.

J.M: Nils, we're talking about use cases. Can you talk about use cases you guys are doing with credit?

N.B: Sure. As I said, we've been deploying AI since day one. First, we started using models to improve our underwriting process, our collection strategies, fraud prevention models and all of that. And recently, we actually started launching our AI agents, internally and customer-facing. And the feedback has been amazing, because now it's not only debt.

When we used to use models, those models were basically just improving the existing workflows that our team members put in place. But now, with those agents, they've actually started creating new workflows, working in a way that none of our team members actually thought about: deploying new policies, new collection strategies much, much faster, deploying new customer acquisition campaigns. Everything has been done with those agents much, much faster than what we used to do. The only thing is, in the critical areas, with underwriting and collection and all of that, it still goes through a human approval process, but we're predicting that in the very near future, those agents will be able to deploy that on their own.

The other interesting part is that last April, we launched our customer-facing AI agent, which is Klivvr AI. It's on the app. The feedback also is amazing, because with managing credit, there is the internal part, with the underwriting process, and then there is the customer-facing part, which is just as important: customer awareness, explaining the product, understanding what they have to do, the do's and don'ts and all of that. 

For example, last month Klivvr AI probably answered north of 40,000 customer inquiries, which would usually have gone to our call center. So now we're deploying an agent at our call center, which will most probably improve the experience and get the customers the answers much, much faster.

We're also using it for collection. There's always been a big concern about how collection works in Egypt with the customers. So now, in the future, deploying our own collection agents should, A, improve our performance and, B, improve the experience for the customers, because they're more aware and they know and understand the consequences if they don't repay.

J.M: What is it you said that the AI is going to do on its own without human approval?

N.B: In general, deploying customer acquisition campaigns, collection calls or messaging to the customers, customer journeys, or sending them promotion offers, information. All of that is going to be done without an approval. And hopefully, we're currently testing and simulating and comparing the performance of the agents to what our teams are currently doing, and we're expecting that in the near future, there will be decisions that will be taken automatically by the agents.

On the other hand, we're also building agents as guardrails. So there will be agents in governance; it's not only on business growth. We'll have agents as well controlling all of that. And there will always be a human layer overlooking all of that, monitoring all of that and seeing the performance.

J.M: What did the thought process look like when you decided — obviously there's oversight at a higher level — that they can fly on their own?

N.B: Basically, it's the ability of those agents, A, to process data much, much faster, with much, much better insights. And with our data, the experience of those agents grows much faster than what the teams have. 

The thing is, when you hire someone, for example a risk officer or a credit officer, he comes with past experience from a different time, from a different institution. The world in general is evolving faster than people can cope with. So I think those — our experiences — that the agents will be much more efficient in deploying. There are no weekends. They don't go home at five or six. So I think overall, the effect on productivity is going to be much, much higher.

And from a business perspective, the risks of taking those decisions are going to be less than the value it's going to create for the shareholders and for our companies.

J.M: Nader, talk about agentic buying and Visa's role in that.

N.S: Actually commerce is, in my opinion, one of the most interesting applications of AI. I was just talking to a friend at the break. I was explaining to her what agentic commerce is all about, and she was saying, "That's very scary." I was like, "That's actually very interesting." 

Most of us, if not all of us, use AI today in commerce, whether we're looking for a new pair of running shoes, or comparing different phones, or planning a vacation. We all use the agent to search for options, compare and do all that. I personally used it, last summer — we were planning our summer vacation — and what typically used to take weeks of research took me a couple of hours. But a year ago, the magic stopped there. For us to convert that into a reality meant that you'd have to go through all the different links and make the payment yourself.

Today, this is changing with agentic commerce. What agentic commerce promises to do, and what is available today, is that you can actually complete the whole cycle and finalize the transaction with a click of a button. So now the agent can do the purchasing for you. And that's what Visa Intelligent Commerce is all about.  

But our business runs around trust, right? Our core business is around trust. And we need to make sure that this happens with the trust in place: the trust of the consumers, the trust of the merchants, the trust of the issuers. And everyone thinks that the payment part is actually the challenge, or the hard part. It isn't. Actually, the payment is the easiest part. 

The biggest challenge in this whole story is the intent. With agentic commerce, we're introducing a new player in the ecosystem, which is the agent. Previously, we had the cardholder, the merchant, the banks. These were the players. Today, we have a new player in that ecosystem, which is the agent, and the intent becomes extremely important. 

In traditional commerce, the most important question was, "Is this the real cardholder?" Today, we have additional questions that we have to answer. Is this a trusted agent? Is the agent authorized to perform this transaction? Is the intent of the customer aligned with what the agent is trying to solve for? These are all questions that have to be answered.

And that's why at Visa there's a lot of focus and a lot of product development going into that intent part, because that's the key challenge that we have to solve for, to make sure it's answered correctly. 

And then from the other side, when you look at the merchant side, merchants have typically always blocked any bots, right? On any website, you block the bots. Now, if a merchant today decides to block all kinds of agents, that means they can actually lose genuine business. And if they decide to allow all agents, that's going to increase the risk of fraud. So today, it's very important for the merchant to be able to distinguish between the legitimate agents and the fraudulent ones. And that's why we've also—

J.M: Is that working? Are they distinguishing?

N.S: Yeah, that's working. We have the Trusted Agent Protocol that we've developed, which helps the trusted agents, the ones we actually trust to perform the transaction as per the intent of the customer, communicate in a secure manner with the merchants. So the merchant can also be 100% sure that this is a trusted agent, legitimately acting on behalf of the customer, to be able to perform that transaction and complete it in one click.

J.M: It's interesting you use the word "trust", which is slightly different from security. We talked about security before. Trust, as I think about it, is the consumer trusting that this stuff is going to work. 

And in this environment of everyone reading about Hugging Face, reading about agents out of control, is trust something you guys are going to struggle with, to get consumers to really trust AI tools, Youssef?

Y.R: I think it depends on how far along the adoption curve you are. And at the moment probably. 

I mean to your earlier question about at what point you just let it go on without supervision, I think there's a very concrete answer for that, which is if you're implementing AI solutions, you need to have evals [evaluations]. You need to really measure what the baseline of the performance of your current solution is, and then compare that to the agent. If it's better, then you lose nothing really by replacing it. And I think at the moment, it's not really quite there yet to be comparable to whatever the analog alternative [is], if you will.

I do think that if you just project that trend two, three, maybe fewer years into the future, without being too tactical about what the specific technological improvements will be, AI is here to stay, and we're going to trust it, and it's going to perform absolutely amazingly, way faster than you expect. And I think people already are, perhaps worryingly, too cavalier about letting AI into all aspects of their life. Sure, at the moment that poses risk. But I do think that there are enough market players in the ecosystem that it's kind of self-regulating, and that the trust will actually emerge, and agents will be a very, very common part of your life.

J.M: You think three years till the AI world? Does anyone disagree with that?

Y.R: I don't even know if I’d quote that myself. Faster than you imagine, though.

J.M: All right. Youssef, to stay with you, we had an interesting conversation about how, two, three, whatever it is, years down the road, AI is going to change the way people live, the way you do your work, the way you exist in the world. Can you talk about that?

Y.R: I think there are a few things to unpack here. First of all, it depends a little bit on what time frame we're looking at. In the short term, a lot of the work will be around "How do I transform my business into something that AI fundamentally helps?" And I think it goes back to a lot of the earlier conversations from today about AI being a shift in business model, not just a technological shift. 

So I think you're going to do a lot of this work too. I laid it out earlier: it's sensors and actuators. Easier said than done. What are the sensors? Where's the data? Do a lot of the work to integrate that. How do I let it write back into the world? Also not very clear what that means and how to implement that. So I think there's going to be a lot of that work.

Once you have these systems that run in a loop and do things with or without human intervention, depending on how good they are, then I think your role becomes a lot more of an executive that manages. I think everyone has heard this before, and it's probably quite true, that we're just going to become planners and orchestrators instead of necessarily the people doing it. 

Which is interesting, because at that point — there was always an implicit equilibrium between how fast humans can do things and how fast we can plan to do things. And the planning was always lagging, because it was the execution that was hard. When you suddenly invert that, where execution is incredibly quick, planning can't really catch up. And planning is oftentimes a virtue of just waiting: right place, right time, you get an idea. Planning isn't as methodical or systematic as some of the execution tasks.

And so what I think that means, the theory I have at least, is that you end up with a lot more effort going towards applying AI in new places, where either the planning itself is a little bit more systematic, or you can build agentic systems that don't plan, but experiment so much that they can just exploit it when they find something that works. And generally, maybe fine-tuning or using some of the existing leverage you have as a business to come up with new business areas. 

I'm more optimistic about the future of businesses, and that they're going to do more with AI, not less.

J.M: I want to close the way we did before, asking you guys [about] the one thing that excites you the most about AI and the one thing that scares you the most. Nils, we'll start with you.

N.B: What excites me the most is, I think, the benefits to humanity and the markets, which are huge and could be compared to the Industrial Revolution. So I think the effects and the productivity will outweigh the risks and all the concerns that people have in the long term. I'm not really concerned about huge unemployment and all these things. 

The thing I'm most concerned about, though it might be a very minor issue, and I see it with my kids in school and all that, is that the education system still needs to catch up, because I think kids right now are not learning as much as they should, and this could have an effect in the future. But all in all, I'm actually very, very optimistic.

J.M: Optimistic, pessimistic? Nader?

N.S: Optimistic with reservations, if I may say so. What excites me the most? I'd say it's actually not one; it's two. 

One is that AI today [has] democratized capability. So now individuals, professionals, small businesses have access to capabilities that before required a lot of time, effort and money. And I think that's great for businesses. I think that's great for economies. I think there's a lot of opportunity coming in that. And as someone who works in product, I'm very excited about the significantly shortened time between ideation and execution. We can now execute much faster.

What worries me, or one thing that I think we really need to talk about more and focus on more, is the speed gap between innovation and governance. And this is very important, because technology continues to innovate and evolve in a very fast manner, but governance needs to evolve as quickly as the technology does. I've spoken a lot about trust. Again, our business all revolves around trust. We need to make sure that trust and governance keep up to speed with the innovation, because usually the difference between the innovations that scale and the ones that don't is not the technology itself; it's how much people actually end up trusting it.

J.M: Youssef, quick.

Y.R: I'm excited about the fact that I think it makes building fun. I think most people probably have a knack for building something. And I think AI removes that initial friction you have of having to get started. You can just ask it to do something. And sure, it's probably limited to the digital world for now, but it will very quickly expand into the physical world. I think you can already design something in CAD with AI and send it to get actually built pretty quickly. I do think that's pretty exciting.

What makes me a bit worried is that, going back to what I said earlier, there was an implicit assumption that there is only this much a human can do. And I think that permeated into a lot of businesses and into the fabric of society. So we are fundamentally breaking that contract, and there are a lot of things that will break because of it. Perhaps innocuously, software engineering, which is probably the main thing that AI has, I think, solved: it used to be that a human could only write this much code, and therefore it was easy to review code. Now you review hundreds of changes a week. The review is now the bottleneck, and it's quite difficult.

I think it's the same thing: you could assume there was a limited set of attackers [cyber attacks], and they need to sleep at the end of the day, they need to eat. And now that's not really the case. They are relentlessly resourceful about hacking you, and they will not stop until they hack you. There is a near-infinite swarm of agents and entities that can do this. 

And not just businesses, but I think businesses, governance, government, everything wasn't really prepared for.