#127 From AI Fluency to AI Symbiosis: How Leaders Can Redesign Work for Humans and AI Agents

Welcome to the Proteus Leader Show, a podcast for forward-looking leaders driving transformation in their organizations, within their teams, or for themselves. Your host today is a senior leader at Proteus International, a firm that partners with companies around the world to make true transformation easier through change consulting, leadership development, and executive coaching. If you're looking for fresh perspectives and practical insights, can take back to work today, you're in the right place. Enjoy the episode.

I think the real AI opportunity is not just doing the same work faster, but redesigning work altogether. Today, we're exploring what it takes for leaders to move beyond using AI as a tool and start thinking about AI as a true thought partner. Welcome to the Proteus Leader Show. Today, we are talking about this exact topic. How AI can be a true collaborator. My guest is David Schreier, Professor of Practice in AI and Innovation at Imperial College London, Managing Director of Visionary Future, author of Welcome to AI, and author of the Intelligence Capital Manifesto. David brings a rare blend of strategic insight, practical innovation experience, and a deeply human perspective on what AI means for leaders, organizations, and society. So today, David and I are digging into what true human-AI collaboration looks like, what leaders may need to redesign inside their organizations, and how to build something far more valuable than incremental productivity.

David, welcome. I'm so glad you're here. Thank you, Marie.

It's great to be here.

So in your book that I've read, Welcome to AI, What do you most want leaders to understand right now about the shift from AI as a tool to working with AI as a collaborator?

If they want to unlock the greatest potential value from artificial intelligence, leaders need to look at a significant redesign of workflow and of the organization. This is not a technology that you can simply deploy like you deploy a new word processor or spreadsheet program. It's not something where you'll get the best return on investment if you just fire people and replace them with AI. But instead, it is a transformational cognitive technology that, if you use it correctly, can give you 2 orders of magnitude better return on investment than what people are seeing today.

So when I think about a process, it has different steps. And maybe the first reaction would be to automate each step. But what you're saying is, You have to think about new steps.

So I'm going to reason by analogy, if that's helpful.

Yes.

Back when in the Industrial Revolution, when we first got combustion engines and steam power even, that replaced water as a source of power, they used to have mills that would make fabric and they have a giant wheel that's sitting there in the river and the wheel would turn and that would power the looms to make cloth.

Yeah.

When they introduced steam power, they had to actually redesign the entire factory.

Yeah.

Because the construction of the building needed to be different on the inside to take advantage of this new technology versus how things were organized when you had water as a source of power. And so that was a profound change. We have a similarly profound change, not in sort of the physical environment, but in the cognitive environment. And so when you look at that, you have to redesign what you're doing. It's not something where you get the greatest benefit by taking a piece and automating a piece and then automating another piece and automating another piece. That would be like trying to run your factory with steam power, but have it configured to run on a water wheel.

I love the analogy. And I guess the obvious question for me is, how do you go about doing this redesign? Because you need to understand what AI is capable of. How do you start doing that?

Yeah, well, part of it is equipping your people with new skills, with a different way of working, because they're not used to collaborating with AI. They might think of it like a word processor, where today we tell the word processor what to do, and sometimes it'll give us suggestions on we misspelled a word. But basically, it's we're in charge and we're dropping information in the word processor and then hitting the print button and printing out a document. Or turning it into a PDF. Instead, people have to learn to work with an AI like they work with a human collaborator.

Hmm.

When you're working with a colleague on a project together, you typically are not sitting there and dictating to the colleague and they just write it down and that constitutes collaboration. You have back and forth, you have discussion, you develop ideas. People need to learn how to work with AI that way. And that's how you get the benefits of this sort of new kind of technology. And so that in turn, if you think about the implications of that, that changes how you work. Your workflow looks different because now you've got a collaborator who's there with you all the time who has an incredibly powerful memory and incredible analytic skills, maybe not so creative and definitely not empowered with empathy. And so you each need to bring something to the conversation. And if you do that correctly, you can Really develop some interesting things.

Yes. And you make me think about leadership because this is what we teach at Proteus International is helping leaders collaborate better with their teams. But in a way, some of the people that will have access to AI, they might be earlier in their career or they haven't gotten the leadership skills yet. So, it's another challenge, you know, for some parts of the organization.

Yeah, absolutely. When we talk to leaders, and I've sat on boards and helped coach management, and I train young people who are emerging leaders, we don't tell them that the way to run a team is for you to give a series of orders and the team scurries off and executes the orders. We talk about the model of a leader who is a coach and a mentor and a source of inspiration.

Yes.

And that's how you get operating leverage off of an organization. Instead of the leader being the sole source of ideas, you unlock or empower the ideas of everyone in your organization. And so you want to start thinking about AI as a new synthetic team member.

Yes.

Right?

It has profound implications for how we manage human capital in an organization when you have these synthetic team members alongside biologic team members.

Yes.

Now, you go even further in the Intelligence Capital Manifesto, and you argue that we are not entering an AI economy, but an intelligence capital economy. I'd love to understand more that new concept. And what does that shift mean really for leaders, CEOs, boards, leadership teams?

So when we think about where things are headed, and even if we try to explain the valuation of some of these now incredibly valuable companies, we have to look beyond today's cash flow into Tomorrow's intelligence flows, and eventually those will become cash flows. But the immediate thing that we can project and measure is the intelligence flows. Those intelligence flows are not just coming from artificial intelligence. There's a reason why we called it the Intelligence Capital Manifesto and not the Artificial Intelligence Capital Manifesto, because people are themselves a source of organizational intelligence. We talk about this when we think about the services economy. We say these are elevator assets. Because they leave the building through the elevator every day. And those assets, those people have ideas inside their head. They have institutional knowledge. And that brings tremendous value to organizations. It helps shape organizational culture. It helps shape products. It helps shape market dominance. And so now in this sort of new era, we're bringing in these synthetic intelligences as well. And we're bringing in hybrid systems. where we can get people and synthetic intelligence to collaborate together into something we call symbiotic intelligence.

Hmm.

And so the first principle is to figure out how do we unlock that collaboration between people and synthetic intelligence and realize the best of both. The second is now that we have this technology, which understands what's going on and can interact with us, we can organize the firm Differently. So back in the 1930s, a management theorist named Ronald Coase came out with something called the Theory of the Firm. And the Theory of the Firm in part says we should organize in a hierarchy. It was based on the German general staff model from the military, where you have one person in charge and then you have functional specialists.

Yeah, like the pyramid.

The pyramid, yeah. But even more, you have these functional specialists. So you have the head of marketing, you have the head of sales, finance, operations, and they then have organizations under them. And so you have that pyramid structure. And furthermore, it would go on to say, if it's— the purpose of the firm is to maximize economic efficiency. And if it's more cost-effective to fire everyone inside a department and outsource it to some third-party service provider, that would be the rational thing to do.

Mm-hmm.

And if it was more cost-effective to hire the people and keep them on staff, then that would be the rational thing to do, where you have this boundary of the firm inside and outside, and you're allocating resources that way. With intelligence capital, you might say, well, actually, even though it's more expensive to keep those people on staff, we're going to keep them on staff because of the intelligence flows they are generating. We don't want that institutional knowledge and that intelligence capital to go outside our walls and be realized and harvested by the third-party service provider. We want to keep it for ourselves because for the first time, we now have an asset, intelligence, which increases in value the more you use it.

Yes.

Most assets—

You teach the models, right?

The models and the people, the systems. It's the model and the people working together in symbiosis. That unit, something we put inside what we call an intelligence capital generator. That unit increases its economics over time in sort of an up spiral, right?

Very different from the car.

Well, very different from any kind of CapEx, right? Like this phone here, as soon as I took it out of the box, it lost 30% of its value. And every day it gets less and less valuable. Like the car you brought up, as soon as you drive a car off of the lot of the dealer, it loses half of its value. These are depreciating assets.

Yes.

But intelligence is an appreciating or compounding asset, and that's very different. And so that then changes how we manage and organize to retain, harvest, and grow those intelligence capital assets.

Yes, because they learn from each other. And I guess this is that symbiotic image that you're talking about. And if you just keep one of them, you lose the value.

And so the organizing principle of this intelligence capital generator, it owns a workflow or a process.

Yep.

It operates under auditable control. So you know what it's doing. It increases its unit economics over time and it retains memory and learning. Those are the 4 principles of the intelligence capital generator. And so now when we think about how we organize our activities, We organize our activities around a whole portfolio of these intelligence capital generators rather than organizing around a pyramid on this sort of German general staff model. We organize around a dynamic portfolio of these processes that are growing intelligence capital assets.

Got it.

Imagine you have a portfolio of thousands of these things, right? And many of them will not work. If you're truly engaged in innovation, you will have a high experimentation rate, meaning in the process of discovery, a lot of them won't work.

Hmm.

The handful that do, though, operate under a power law curve, meaning that the ones that work are incredibly valuable, and that generates sort of significant return on investment for the company. So, to put numbers on it, and we can look at multiple decades of innovation research to put numbers on it. 1 in 3,000 of these things will generate massive returns. And if you manage a portfolio of them intelligently over time, over, let's say, a 5-year period, you will get a 300% return on investment.

Wow.

So, there's good experimental evidence to show that this model works. By experimental evidence, I mean, we tried it in the real world and it worked, right? And so, it is a very different way to organize, but this is not new. We've been experimenting with new ways to organize companies now for about 20, 30 years. So in various ways, Unilever, for example, has experimented with this in various ways. Valve, the software company that runs this very popular platform called Steam, is organized this way. So there are companies out there that have been trying out these new models of management. And so what's interesting is we can manage it more efficiently thanks to the latest generation of AI. And we can get more out of it because of these human-AI collaborative systems as symbiotic intelligence.

A few things I'm taking away. First of all is innovation, right? You need to get a lot of hubs of ideas to kind of get the few that will really generate the ROI. What I can't envision yet is in those companies where it's happening, you have those systems, maybe thousands of them. What happens in between them? Who coordinates all that ecosystem?

You can use AI as an orchestrator or AI and people together. You still have managers. And those human managers work with synthetic or artificial intelligence managers to stay on top of this portfolio of processes or intelligence capital generators. So, you know, what's happening in the most successful implementations of agentic AI is not that you have one monolithic AI that kind of runs everything, but instead you have specialized agents that each have areas that they focus on. And then you have an orchestrating agent that manages the other agents, right? So that's this emerging model of how we work with these things. Not that we have a uniform single AI, but rather we have many AIs that are doing different things.

Yes. And more specialized, which I believe is also Better for the energy use is to focus the tasks. That too.

Yes. You get more efficient, right?

Yes. Now, obviously this question comes up a lot. As AI agents become more capable and better, where do human judgment, ethics, context, creativity, all those elements become more important? What do we retain as humans?

So you're asking 2 or 3 questions there. I'll try and unpack it, right? So on the one hand, as we organize companies, we still need to keep humans in the loop. And we still need to embed the values and principles that humans want into the intelligence systems. So just because you've got a synthetic intelligence doing something does not mean that humans are avoiding responsibility. If you want to put it into corporate terms, and there was a notable lawsuit against Air Canada around this a couple of years ago, the board is still ultimately responsible.

I kind of prefer that view because the in the loop, I think, could be viewed in a negative way versus in control, like that the humans, they're not just being kept informed of what's happening, they're actually part of the decision-making process.

Well, they're ultimately accountable, right? If we're talking about a RACI matrix, you've still got ultimate corporate liability and accountability devolving to the board and management and the C-suite. And so with that ultimate accountability, you wanna make sure that the governance systems you put in place around these AIs and ultimately around these intelligence generators, which have human and AI, you still wanna have governance that reflects the values of the organization, compliance with law, compliance with applicable regulation. And this becomes a systems design question, right? When you're designing how you use these things, you want to do it intelligently.

Yep.

Yep. Now, we love leaving our listeners with some practical tools. And I think here the tools could be thinking about how to start building real human-AI collaboration. So just maybe a recap on if I want to do that from tomorrow and I have a team, I have a business, what are the steps I need to take?

There are a couple of first steps. One is you need to make sure your data is accessible. So a lot of companies have run into challenges over the past few years that they've tried implementing more advanced AI systems and agentic systems because their data remains highly disparate, scattered, and with poor data quality. So better data management and data governance. It's not very glamorous, but it turns out it's incredibly important to actually get value out of your AI investment.

Yeah, it's the inputs, right? This output will be as good as the inputs.

I think of it kind of like the fuel that you put in the engine, right? I mean, you may have a very shiny, beautiful engine, but if there's no fuel, it doesn't go. Or the electricity that you put into the electric motor, if you prefer to be green. And so, okay, we need that electricity to power our motor. The other sort of critical input is you need to make sure your people have the skills for working with these systems. And raising the capabilities of the human part of the organization is incredibly important.

Yeah.

Helping them acquire the new skills to work with this new approach to work is important. So for example, I was talking to the chief AI officer of a large listed company, and he was saying how it's this painful cultural shift because the way people were used to working previously is they'd show up at their desk in the morning, they'd have a list of tickets to clear. So they'd have, I don't know, 10 or 20 tickets to clear through the course of the day. And that was how their productivity was measured and their incentives were organized and their compensation. And instead, the new way of working is The AI systems are clearing 90% of the tickets, and the human is monitoring the systems to make sure they're not running into trouble and resolving the 10% of the tickets that are exceptions. Maybe that human now has a greater span of tickets that they oversee. Maybe they're seeing 50 tickets instead of 10 tickets per day.

Yeah.

But it ends up being a different approach to work and something that's challenging for the human to adapt to because it requires reframing how they think about their job.

Yeah, it's back to change management and humans are naturally resistant to change because like here they have to trust the machine, then maybe focus on the difficult tickets, which, you know, take longer to solve.

This is a change management problem. This is not a technology problem. The technology is more powerful than the tasks that we're currently assigning it to. And The resistance to change and architecture around that change are basically serving as obstacles to getting the most out of this investment.

Yes.

We talk a lot about new beginnings and a number of endings you need to acknowledge as we support companies to change and mindset change for all the individuals, right? Because you have to overcome that challenge.

It's very difficult. And I'm not surprised when I see these articles coming out where they say a big financial services company implemented AI and they only got a 17% return on investment and they were expecting a 90% or 200% return on investment. They didn't get it. What happened there? And my answer to that is they probably just did rip and replace. They probably took a human away and they dropped an AI in, and that gives you your lowest return on investment.

You didn't redesign the system as we talked about.

Yeah.

You get a higher return on investment with augmentation, which is where you keep the person and now the AI is supporting them. So that gives you 10x better than rip and replace. And you get 100x better than rip and replace if you engage in symbiosis, if you bring people and AI together in collaboration. But that latter one, that symbiosis, requires this radical organizational redesign.

Hmm.

Yep. Which is risky. This is the challenge. If you want to make that quantum leap, it requires significant change rather than incremental change.

Yes.

And what about investments? If you move to that 3rd stage, what are the investments we have to consider?

Yeah, a huge investment in change management and retraining and reskilling in addition to these investments in the technology that provides the springboard for this upside.

And back to what you were saying earlier, because a lot of those new systems are going to fail, is also being okay with that. That's part of it.

You could say failure, right? And statistically you could say 99% failure rate is expected, right? Or I like to look at what Thomas Edison famously said, which is, I didn't fail 10,000 times. I discovered 10,000 ways not to make a light bulb.

Not to do it.

Because every time he would try something out, he'd write it down in a little notebook and he'd figure out what went right and what went wrong. And then he'd try something different. And so that was a very disciplined, structured, scientific process of discovery. And so if you take this mental reframe that it's not failure, it's discovery, that can also make it more culturally acceptable to the people who are going through this process where a lot of the things they're going to try are not going to work. So I saw last year MIT came out with a white paper that everyone was very excited about that said, 95% of AI pilots fail and therefore AI doesn't work. And everyone got all wound up about this. And I look at that paper and I say, well, yeah, but if you look at the innovation literature, if anything, 95% is low. Like in 1910, there were over 500 automobile companies in the US. By 1950, there were 3 of any note. There was an extremely high Failure rate. But what that really was is a process of discovery into which ones had the best business model and the best technology that could reach the greatest scale at the lowest cost.

Yes. That doesn't mean we shouldn't use cars.

It doesn't mean we should not use cars. Part of the failure that we see today is it would be like you strap a jet engine onto a horse and then you're surprised that it's not running fast.

Yeah, I love this Thomas Edison quote. At the end of our podcast, I think it's a perfect way to end. Thank you so much, David. There's so much depth in what you shared today. I think we could talk for hours.

I appreciate the time. Thank you.

Yes. And what really stands out for me is that the biggest opportunity with AI is not simply to automate more or accelerate what already exists. It is to rethink work itself.

That's right.

And that is a leadership challenge as much as a technology opportunity.

This is a significant change. It's not an incremental change. And so we need to be very thoughtful about how we engage in it.

Yes.

Now, to our listeners, I hope today's conversation sparked new thinking about what AI could look like in your organization when it is approached with clarity, responsibility, and imagination. So to go deeper into David's work, I highly recommend to read the Intelligence Capital Manifesto on aicollab.org and his book, Welcome to AI. We also have 2 other Proteus Leadership Show podcasts focused on human AI, episode number 89 and number 115. Check the links in the show notes. For more conversations and resources on leadership and change management, visit proteus-international.com. Thank you, Erika. Thank you for tuning in. And until next time, here's to creating the leadership you truly want in the age of human-AI symbiosis.

Thanks for listening to the Proteus Leader Show. We hope you're leaving with ideas you can put to use on your team, in your organization, or in your own career. To learn more about how we support true transformation through change consulting, leadership development, and executive coaching, visit us at proteus-international.com. Until next time.