AI-first companies and the future of work: As AI systems become more autonomous, what separates companies that truly transform from those that simply automate existing processes? Jade Cano speaks with François Rüf and Zeno Staub of worldreplica about why the next wave of AI is not just about productivity gains, but about redesigning workflows, leadership, and decision-making around human-AI collaboration.
JADE: [00:00:00] Right now, AI is moving at an extraordinary pace. Many businesses are still trying to answer a very simple question: how do we create value from this new technology? I'm Jade Canno, a business journalist, and I'm talking today with Francois Ruf and Zeno Staub, the CTO and chairman of WorldReplica, about what it takes to become an AI-first company.
VO: Welcome to The Clear Signal, the podcast [00:00:30] linking AI to strategy.
JADE: Francois, Zeno, welcome.
ZENO: Thanks for having us. Thanks.
JADE: I'd like to ask you, Francois, what are companies actually doing with AI today, and how much of that has to do with real transformation versus experimentation?
FRANCOIS: It's very early still for enterprises in this AI transformation, so I think we're definitely more in the experimental phase still. When a company has deeply embedded [00:01:00] AI into their core operating system, into their business model, where KPIs really depends on it, then I would say we're in full swing with transformation. But right now we see more on a task level, maybe workflow level, where companies try to start to automate first processes.
JADE: Okay, so first steps.
FRANCOIS: Yes.
JADE: And if we look across the market, Zeno, what separates companies that are getting real value from AI from those that aren't? And perhaps you can giveus an example [00:01:30] of a company doing it particularly well.
ZENO: Yeah. I would propose that we look to an industry that has already seen tremendous change through technology in order to understand where we are today and where it all could go in terms of AI.
So put yourself in the shoes, you own a small taxi fleet in the '90s 10 drivers on the street here in Europe. How did the business work? These guys were queuingat predefined places, waiting for a customer to come [00:02:00] along, waiting for a call from the central station in order to dispatch them to somewhere.
Productivity gains within the existing system would mean that you would optimize the queuing and timing of your drivers to send them to a specific place, to advise them to take the break when there is low traffic. That would be productivity gains within the existing framework. Then all of a sudden you would say, "Hey, let's think about the [00:02:30] product.
Let's actually put an app on the phone of the actual user." And that's obviously what Uber has done. So you have completely changed the product experience, and now people are working on a completely new- Business model with self-driving cars. Now, obviously, if you look at the value creation of Uber or of Waymo and the potential impact on that of the industry structure, we are talking completely different levels.[00:03:00]
And I think that's how to think about AI transformation. We are currently, as Francois said, in the experimentation phase, focusing on productivity gains within existing frameworks. Just- And
JADE: we're not making that jump to-
ZENO: Not yet ...
JADE: changing the business model.
ZENO: Now unleash your mind and say, "What could this mean if we now use this for product innovation, for completely new business models?" Or then even in the next phase, where private cars, public [00:03:30] cars could morph completely into a new common mobility system, which probably still is a couple of years down the road.
JADE: Okay. Let's go back to the beginning. Yeah. So when companies do take this step, where is it they typically see the first and biggest gains?
Is it within productivity? Are there other areas as well?
FRANCOIS: Yeah, definitely on the productivity side. AI has been around a very long time. It's before the large language models, for optimization tasks. But now I think the new layer which is coming in is mainly on the white collar work, repetitive work [00:04:00] like coding documentation, purchasing.A lot of these functions, you see a lot of gains. People start to automate some of the tasks in that role.
JADE: Okay, so that's where we currently stand today. But what's changing now is not just what AI can do, but also how work itself is structured. We're hearing a lot about agentic AI systems that can plan, act, and adapt across workflows, not just respond to prompts.
Francois, what is the [00:04:30] difference between automating a workflow and making it truly agentic?
FRANCOIS: Yeah. Process automation is also a very old topic. We have done this for a very long time. But it's basically you tell a software if then else what to do, so you tell them how to do it. Now, the new agentic way of doing it is more goal oriented.
So basically you tell the system, this is the end result that I'm expecting with some boundaries and borders, and then the system itself is making a plan how to do that. So you have [00:05:00] a way more freedom of the system to think through how the planning looks like and then a more autonomous execution.
ZENO: And I think another key change is also white collar users of technology. When they use directly their personal technology, meaning the PC and their work suite, they were used to immediate feedback You key in the formula in Excel, you get immediate response. Now all of a sudden, you set up a team of [00:05:30] autonomous agents or semi-autonomous agents who will take your order and think about it and get back to you in an asynchronous way with results that you need to judge again.
So also how you interact with technology, I think will change profoundly.
JADE: We're seeing a lot of companies trying to improve their existing systems and using AI like a layer on top of a process. At what point should companies really stop this optimization and start thinking about [00:06:00] building workflows around entire AI agent networks?
FRANCOIS: Yeah, absolutely. I think it's about the capability of the current agents, large language models, which is still an exponential curve. And I would say as a rule, if you start to see that the core piece of the work an agent is capable of doing, then probably you should look into it.And the other thing is, you have a lot of processes that are made for humans with a lot of handovers [00:06:30] and inefficiencies. And also there you can start to look into, to make this more seamless. Because you suddenly, you have someone who can work for twenty-four/seven and do the job. And then you will see big gains.
ZENO: If I may add to this, when we talk about, you ask about, how to scale-
Yes ... how to actually unleash the potential at enterprise level. Very often these pilots, these use cases start bottom up with the most [00:07:00] innovative employee, with a team that dares to do it, that is allowed to do it.
But when you then really attack the core of an enterprise, then rightfully the whole security, data, control, quality, scalability stuff kicks in. So if you really want to reach scale, you need to allow for bottom-up pilots, but at the same time, bring your AI readiness with all these enterprise [00:07:30] functionalities to a level that you then can allow scaling.
JADE: Sure. So this then obviously affects leadership. Yeah. So let's say even if the technology is there, scaling it is an issue, and it does seem like on this journey of AI transformation, at some point it top-- stops being a technology issue and becomes more of a leadership issue. What would you say are the biggest misconceptions then or that leadership has about this new technology?
ZENO: Probably that this is again, a technology that kind of [00:08:00] has an impact As one would say it from coming from the background of my former industry in the back office, that this is still something literally happening to the factory floor. And I think it's now for the first time that technology actually reaches the white collar- thinking of work. And I think that's a misconception.
And I think what we all still struggle to digest is that what is happening right now is that [00:08:30] average intelligence and aggregated average know-how is made available at almost zero marginal costs.
And if you think that through, I think the implications are huge.
FRANCOIS: And I think, through all the hype people expect super fast results. And normally if you look at the internet, for example, what kind of ecosystem completely emerged out of it. I think the long-term potential, it's very hard to understand. So don't underestimate the transformational [00:09:00] aspect of it.
JADE: And what about what it m- could mean for leadership itself? If AI systems begin to act autonomously- Yeah ... do leaders then stop leading and they, stop making decisions, and they start to think about designing systems- ...
about supervision?
ZENO: Yeah.
JADE: Does it change leadership?
ZENO: Yes and no. I would say, I believe final decisions, final responsibility will and should always remain with human beings.
On the [00:09:30] other side, good management was never telling people what to do. Good management was always creating an alignment on shared goals, and then managing scarcity or managing trade-offs. And now what will happen is that just you will get much faster on your desk the required analytics in order to actually judge scarcity or overcome scarcity and understand trade-offs.
So the job of [00:10:00] leading will again become more demanding because things will move even faster.
JADE: Francois, do you see this the same?
FRANCOIS: Yeah. I think, there, there is one issue which is coming up and this is, if models are getting really good, then they will propose decisions where you will have issues to judge whether this is a good or bad decision.
Because think about it, if you have Einstein besides you, then what are you doing? Because he says I would do this." And then can you actually follow the [00:10:30] rational yes or no? So I think this is an aspect that will come into boardrooms, decision systems very quickly.
ZENO: He's always slightly ahead of the curve.
JADE: Yeah. Although all this disruption is changing the structure of organizations themselves, it sounds like. Let's discuss what an agentic organization looks like. What challenges emerge, Francois, when humans and AI agents work together in teams, and how do you actually lead a team made up of both humans and AI agents [00:11:00] day-to-day?
FRANCOIS: I think the biggest change is coming in that most people currently, they are trained to execute the task, but the agent will do the task inthe future, so they will become also leaders because they have to delegate the work, which is a very different way of working because you're overseeing virtual employees and not executing the actual task.
And I think this is a very big change that needs a lot of upscaling inside the enterprise.
JADE: Absolutely. How does that change the [00:11:30] idea around productivity, or how does this affect compensation even when you have agents working alongside humans?
ZENO: It's still early days and academics are asking these questions, down to the slightly outlandish idea that perhaps agents need bonuses in form of tokens to optimize themselves.
Who knows where we will go? But I think one thing that will happen very soon is obviously performance reviews or productivity discussions should always have been only on [00:12:00] output and impact. But us being humans, in many cases, we also appreciated and judged the input. That was hard work. That was a difficult situation. We did our best.
JADE: We worked on a project for-
ZENO: Yes ...
JADE: so and so many months.
ZENO: Yes. Yeah. Yeah. And as Mr. Churchill once stated, "Stop giving your best. Do what is required to achieve the goal." That's not completely day-to-day human behavior. But now with agents being [00:12:30] available, doing whatever kind of input that is required, the discussion has to shift completely on output and on impact only, and I think that will be a, again a challenge also for leadership.
JADE: Sure. And also it has to do with building trust in AI within organizations. Do you have any suggestions about how to go about that?
ZENO: Yeah. There are some I would say technical framework answers to it. I'll leave that to, to, to Francois is much, much better in that in order to how [00:13:00] to design systems that you are willing to trust them.I personally believe that also the way how you interact with them is important in order to build trust. So we all know this famous book, The Hitchhiker's Guide to the Galaxy, where you build a huge project, you send it off, what is the meaning of life? You never hear for years, and then the answer comes back 42.
That's obviously nonsense. But now you can also set up an agent, ask it about [00:13:30] your market entry strategy into emerging markets. It works for five hours, and then something comes back you don't understand. You will not trust it. So you will need to design the interaction in a way that you can shape, understand, and at least has perceived control of the work plan that the agents are setting up, perhaps checking at certain milestones, and then the result will be yours, and you will trust it [00:14:00] and run with it.
JADE: Absolutely. Now, as we move from automated workflows to hybrid systems and eventually to fully autonomous ones, what will work itself end up looking like? We've touched on it a little bit, saying that we'll no longer be doing tasks- ... but we'll be more in a leadership role perhaps. But we are already seeing machines starting to outperform humans at work.
Anthropic's Mythos system has shown that it can find and exploit decade-long undetected software bugs on its own end to end. Does [00:14:30] that mark a turning point where machines are no longer just supporting work, but actually doing it better than we are?
FRANCOIS: Yeah I think, this is the big shift, actually, what's happening, because people think, it's a productivity thing, but at the end of the day, you will see, I'm coding since I'm a kid, so but, the current models are already way better coders than I ever was, so this is already superhuman for me, for myself.
So first of all, you get used to it very quickly, but [00:15:00] the question is always on the boundaries.
ZENO: Yeah, and I think we really are, as Francois said, that it, in this Turing moment where, the British scientists realized that they could not break the Enigma code machine of the Nazis with their brains alone.
They had to build a machine to fight the machine. Absolutely. Now, if your competition is using high-end AI in end consumer marketing and you still work with the smartest people, you will [00:15:30] simply lose.
JADE: You don't have a chance.
ZENO: So you have to set up yourselves that machines can counter machines where they are simply better than humans.
JADE: The more we use AI, the more trust we're going to gain into the technology. So should we be thinking of it as something that will ultimately just be part of our invisible infrastructure and no longer a differentiator, but something every company depends on?
ZENO: I think so. I'm unfortunately old enough to have seen the first internet bubble in [00:16:00] 1990, and already at that time, The Economist had a leader saying we should stop differentiating between companies that use the internet and companies that don't, because we have also stopped differentiating between companies that use the telephone and those that don't, because those that didn't are not here anymore, and this is exactly the same.
JADE: Do you think we need a whole new governance structure for AI, perhaps an additional control layer for AI agents, for example?
FRANCOIS: This is [00:16:30] becoming increasingly important. How do you govern system that are more capable than you are? This is something that currently now the industry is working on. There, there are a lot of unsolved questions around biases in in these models, which are very hard to get out.
The EU AI Act, has a risk classification, for example, for higher risk use cases in HR. If you have a bias on HR in the model and you only let the model decide, then maybe [00:17:00] you just, certain people have a disadvantage, which you don't want as a society. I think the more capable the models are, the more important will be that we have the right borders around it how to govern it.
JADE: Sure, and also how to trace accountability.
FRANCOIS: Yes.
JADE: Would you trust AI to execute a strategic decision at World Replica?
FRANCOIS: That really depends how the system is built, because what you wanna see is that the AI has access to all the relevant data. It has spent quite amount of time to reason [00:17:30] about it, and if it has done this, given the current capabilities of the model, I would say it's probably a better decision than I would do.
JADE: And Zeno, would you feel comfortable delegating that kind of responsibility?
ZENO: Not yet, but I would listen very carefully to the results of our own AI.
JADE: Excellent. Now, looking ahead, will companies in the future be structured around departments, Francois, or around workflows executed by a network of agents?
FRANCOIS: The department still makes [00:18:00] sense. I think it's an abstraction layer, which is it's not dependent on agents or not. But underneath, I think it's gonna be a much more fluid around value creation. And, currently you have all these reorganizations if things are changing, and I think, these agents will try to find optimal organizational structures all the time as an ongoing process and not siloed execution that we have today.
I think it's gonna be much [00:18:30] flatter hierarchical towards self-organization, but not fully, because you also have some legal boundaries and reports that you need to create. So there are kind of things that you have to follow as a business. But around that, I think it's gonna be quite fluid.
JADE: Where should humans remain in control no matter how advanced AI becomes?
ZENO: We should always control our destiny. We should always tell the smartest systems in the world where we actually want to go directionally. So we should [00:19:00] control what today is called the purpose of a firm. We should control its values.
We should live them, and we should just accept that the, as St- Steve Jobs has said, the PC was a bicycle for the mind, and now we have wings for our minds.
JADE: Absolutely. Thank you very much. And with that, we've come to the end of our first episode. Thank you, Zeno, and thank you, Francois, and thanks to you for [00:19:30] listening.