Category: Machine Learning

Source: Microsoft

Headline: Podcast: Azeem Azhar on how AI agents are transforming work

MOLLY WOOD: Today, I’m talking to entrepreneur and author Azeem Azhar. For nearly a decade, he’s published the Exponential View newsletter, which breaks down the ways technology is transforming every aspect of our life and work. He also serves on the World Economic Forum’s Global Futures Council. He joined us to share valuable insights on how we can adapt and succeed at a time when change is constantly accelerating. And now, my conversation with Azeem. Azeem, thanks so much for joining me.   

AZEEM AZHAR: Well, thank you for having me, Molly.  

MOLLY WOOD: So for people who aren’t familiar with your newsletter, Exponential View, can you give us sort of an overview of the topics that you explore in your writing and speaking and interviews and various explorations? 

AZEEM AZHAR: Absolutely. The title in a way gives it away. Exponential—it’s about fast-changing technologies. The one that matters most at the moment is artificial intelligence, so that has formed the backbone of what I’ve written about over a decade. But there are other exponential technologies as well. So, what’s happening in the new energy system with the cost of solar panels falling exponentially, the same is happening with batteries. It’s happening in the worlds of biology, where gene sequencing and genomics and proteomics are getting exponentially more accessible, and I try to bring all of those together through my own framework, which is about why these technologies get cheap, what happens when they get cheap, and how does that then manifest itself first in business, then in the economy, and finally in society. 

MOLLY WOOD: Having seen so many of these disruptions, what’s the high-level advice about how to adapt your business and your culture?  

AZEEM AZHAR: The reality is that there is no rule book. And I think one of the challenges for any business person is that they’ve been able to operate in a world where there has been a rule book and they’ve been able to get that rule book from business school, they’ve been able to get it from a textbook or a dummy’s guide, which is normally the place I turn to. But what happens in a world where there isn’t a rule book because everything is being made up as we go along? Some of us remember that, because if you were early in the internet you will absolutely remember that simple things that we take for granted today, like being able to count the number of visitors on your website, were really hard technical and product problems which had to be invented by dozens of companies around the world. So the thing that really matters is the capability to learn, and that learning has to come from actually experiencing the technologies. At some point with the internet, you didn’t need to know about a stack of technologies from tcp ip to ftp to dns to http into a whole set of other acronyms that may mean nothing to listeners. You could just go to a SaaS provider and say, provision me an online store. But it took us about 15 years to get there. Where we are with this AI change is that we are at those early years and it’s not clear to me at what point everything stabilizes sufficiently that you can just, you know, download a manual or buy a book and figure your way through it.  

MOLLY WOOD: You wrote an interesting piece that I want to ask you about in January about contrarian ideas about GenAI in the workplace, and you sort of started with point one right now, which is, you know, that we are really only scratching the surface of what’s going to happen to work here. Let me start by asking you, why do you think this is such a big deal and that, in fact, we’re only scratching the surface?  

AZEEM AZHAR: It’s a huge deal because the technology of GenAI is kind of magical. You can talk to your computer and it talks back to you in quite sensible ways. I can have a half-baked thought and I can speak into my phone and the large language model will turn it into a structured outline. I can take that structured outline, I can post it back into the large language model and say, write me a research report on this, and it will go off and do that. That is absolutely at the heart of the cognitive work that drives most of the value in most companies in the world. And the technology is coming into a world that is ready for digital technologies. So 30 years ago, when the internet showed up, you had to do a lot of infrastructural work, you had to teach people how to use the internet, you had to move processes online. Over the last 30 years, companies have gone through a process of, you know, business process reengineering, transformation, digital transformation. Everything is now digital. And so this new technology, which has got this magical component where I can just talk to it and it can talk back to me and do quite sophisticated things, is now also available like that [snaps fingers] at the snap of the fingers, and you know, Microsoft has demonstrated that, it’s done it. It’s put Copilot and a whole load of other AI tools in the hands of probably hundreds of millions of workers in a matter of a couple of years. So the combination of a really powerful, easy-to-use technology onto the desks of loads of workers, I think kind of creates a completely new and unparalleled situation.  
 
MOLLY WOOD: So one of the evolutions that business leaders are struggling to keep up with is that they’re just getting a handle on the capabilities of AI assistants—like Microsoft Copilot—but now they’ve got to wrap their heads around the potential of Copilot plus agents. 

AZEEM AZHAR: One of the things I would say is that the speed with which people are adapting to assistance is pretty remarkable. And I think historians in a few years will be able to look back and give us accurate data as to whether it’s quicker than, say, the internet, which I think it does feel like it’s quicker than the internet. What is an agent rather than an assistant? Well, in an assistant, we sit in a world where, effectively, there’s a kind of query and response between me and the AI system. I might say, improve the phrasing of this letter to my lawyer and send that in and it will go off and improve the phrasing and send me the results. With an agent what I can start to do is have it undertake a more open-ended multistep task that may actually have a goal. And what that’s doing is it’s taking me out of the loop in those intermediate tasks. I’ll give you an example of one agent system that I use. This is a system where I want to essentially access a brain’s trust to improve the quality of messaging in something that I might be sending out. I will have four different AIs. One of those AIs acts as a moderator and the other three act as members of a focus group. And I can take my material and send it to the moderator and say, please have the focus group criticize and improve this until they all agree that we’ve got something that scores 10 out of 10 on how compelling it is. I will put that query in, the moderating agent will run that process, it’ll take three or four minutes, it’ll cost me 10 cents, maybe 15 cents. And at the end of it, it’ll come back saying, right, this is much more compelling messaging, as these particular agents agree. Now, each agent has a persona. This one is like a 45-year-old marketing manager, and this one’s like a 37-year-old early adopter of technology, and so on. We’ve given them those personas, and that process runs in of itself, and what comes out at the end, works about half the time, is often more compelling in some sense of what went in. So that’s an example of using an agent-based workflow where if I hadn’t done that, I would have been clicking and pressing and copying and pasting from tab to tab to tab, and guess what? I would have made lots of mistakes and got very bored.  

MOLLY WOOD: Right. You’ve said, for example, that in the future we will have hundreds of agents working on our behalf. How does that reshape business?  

AZEEM AZHAR: Well, let’s go one step at a time then. So I’ve given you an example of using a set of agents to construct a virtual focus group that helps me from time to time. What I do when I look at using AI is I’m really interested in using AI to improve the high-value tasks I undertake. Some other people prefer other approaches like using agents to schedule their meetings for them. But, you know, that for me is kind of a low-cost task. I don’t really need to automate it. If I can get help on the things that really drive my business, that’s where I want the help. So that’s one example of a task, the agent model. Another example is using the tools to do really, really detailed research. What I can do with deep research is I can ask it a particularly tough question that might be about market dynamics, or it might be about a technology area that I’m interested in investing in. Deep research will turn that into a research question and go away for between five minutes and, in one case, 75 minutes to produce an annotated multi-thousand-word report with references to academic and mainstream news sources, and it’s really, really quite impressive. I would say it is about as impressive as having a couple of junior analysts working together for a couple of days, so really impressive. But you would never trust it, right? You always work on what the juniors have produced. So I think that gives you a flavor. But then you still ask me this final question, what does it do for business, right? I think that’s really the big question.  

MOLLY WOOD: Fundamentally, what we keep coming back to is it’s a question of leadership, adaptation, and adoption. You know, how do leaders get into the mindset of playing both offense, in terms of unlocking new business opportunities and delivering value to clients, and also defense, like reducing costs and making sure that everybody understands how to use these tools well.  

AZEEM AZHAR: So what I go and tell CEOs of big companies is, no one built a great business by cutting costs. What’s really interesting is what you can do to do more and to deliver more, and that’s a choice you make, and as a CEO you might say, my shareholders would just rather me cut costs. And if that’s the case, that’s the decision you should go off and take. If you want to do the latter, then a lot of the changes will actually come from frontline employees, because they are the ones who deal with the reality on the ground every day. They’re the ones who know which part of the existing processes work, which ones no longer work. You need to get their insight on what the potential of the technology is. The other side is what the CEO knows, understands, and, most importantly, feels. Do they feel this is going to be a radical breakthrough technology? Because if they don’t, they will only ever sign off a checkbox on a slide presentation from a consultancy. They will never really believe and drive their team forward. Now, I have personal experience of this, because in my first job when I worked as a journalist, and to put the Guardian newspaper in the UK online, the deputy editor who went on to become the editor, Alan Rusbridger, felt and believed the internet was going to be transformative to the media business. He felt and believed it in 1994. And so I’ve been really lucky for my first experience to not have to push water uphill. It’s hard, I think, to make a radical change without a sense of belief and a sense of intentionality. I think what you can do as a leader is you can get buy-in. You can say, look, I’ve bought into AI and we’re putting AI in customer service, we’re putting AI in fraud detection. That reminds me a little bit of the very first car manufacturers at the turn of the 20th century, who bought into electricity by hanging a pendant light in the workshop so workers could work an extra hour or more. But the person who believed in electricity was Henry Ford, and he realized, with electricity, you could build cars in a completely different way through a production line. 

MOLLY WOOD: That is an excellent analogy. Okay, now tell us how that looks in a business deploying AI as a light bulb versus deploying it to automate a factory.  

AZEEM AZHAR: Well, I think with the light bulb example, that’s probably where most companies are. You know, you’re using the AI to improve customer service or ticket response, and you’re measuring it by cost cutting. The question is where are you delivering more at a higher quality, but at constant cost to your customers? And that’s only possible because you have got that sense of real belief in the technology. We’re early days yet to find really good examples of that. There are a few in digital finance that are emerging. There are, of course, the AI-native companies that are building the tools themselves, who, in a sense, have bought their own dog food. And I think there are small firms. I think Exponential View, the work that we do, is entirely AI-native now. And we wouldn’t be able to do it with the team we have if we didn’t have the tools that we use.  

MOLLY WOOD: So we’ve sort of kept this conversation to this specific disruptive technology. Of course, there are lots of other disruptions happening. And so I wonder how you think AI will help leaders navigate other changes—economic uncertainty and supply chain disruption and intensifying competition and strategic and global issues and, you know, you could keep going with this list. But at some point you have to stop.  

AZEEM AZHAR: I think there’s a simple model here, which is that all of the issues that you’ve raised are problems that are first and foremost cognitive problems and they’re knowledge problems. In other words, you have to orient yourself. What’s really happening with our supply chain? What’s the root cause of this problem? How might we trace back the dependencies between that cause and the issue it’s facing? Those are all analytical questions that rely on data gathering, and they rely on that sort of second order analysis. And AI tools are really, really good at helping people do that. So for every strategic problem that a business has, you should be able to take these new generative AI tools and help you with your identifying the root cause with your strategic planning, with your scenario analysis. And so I don’t see how you can address these given the growing complexity of the world without some kind of help. And of course the kind of help that often does this—the strategy consultancies, the academics—they’re rare, they’re overworked, and they’re expensive.  

MOLLY WOOD: Leaders have an opportunity to be a different kind of reactive, the way that you’re describing this, right? The worst thing is for someone to react without information. And what you’re saying is we now live in a world where there is no reason for you to be able to do that. You have all the information and the help that you need to react in a smarter way.  

AZEEM AZHAR: You have all the information, and you have the abilities to process large amounts of unstructured information and come out with real insights to help you act. Of course, then, acting on it is still complex. You have to persuade your leadership team, you have to find time to figure out whether you really believe the decision that you’re about to take. So those human dimensions and human social dimensions still exist. But what it really means is that the cognitive knowledge component of this question, the scenario planning and hypothesizing, is something that can take place quickly, cheaply, and frequently.  

MOLLY WOOD: Okay, well, speaking of knowledge, you’ve written about how knowledge is different from data and that we should not use those words interchangeably. Can you dig into that a little bit? What’s the difference?  

AZEEM AZHAR: Data is, in my mind, the smallest, lowest level unit. A piece of data, you know, it’s useful in so far as from aggregations of that you can get a lay of the land. Where old AI systems used to operate before generative AI, they were really good at helping us understand patterns in data, so, understanding patterns that might say, there’s fraudulent behavior happening here. But what generative AI allows us to do is it allows us to synthesize across many, many different domains. And so you can take data as we often do, we take web analytics data from our websites and we’ll throw them into one of the LLMs and say, tell us what the most important changes in behavior on our website over the last three months have been. And what you’re then starting to do is get that higher-order analysis that you would normally have asked your web analytics team to produce for you. They can now actually do much, much more with it. Companies have been very, very data rich, but they’ve probably not had the capacity to turn those into knowledge-driven decisions to actually change what they do. We’ve become very good at optimizing an existing sales funnel, not asking how could that funnel be radically better. And I think that process, which is a bit more creative, it’s a bit more strategic, becomes a little cheaper and more accessible now that we have the tools that we have at our fingertips.  

MOLLY WOOD: What are some of the most unexpected ways that you’ve used AI in your work and, if you’re comfortable, in your personal life?  

AZEEM AZHAR: So my one power tool is that when I’m driving back after dropping my daughters to school, I will dictate my random thoughts into one of these LLMs and tell it to order them for me. And so when I get to my computer at my desk 20 minutes later, I’ve got a structured set of to-dos, but often at quite a great degree of granularity that I can sometimes just copy and paste straight into an email. So that is my work one.   

MOLLY WOOD: Wait, wait, tell me more. So you’re like, I know I have to do this stuff, I need to blah, blah, blah, and I need to do this, and I need to email this person, I need to follow up on this. And then you say, like, can you prioritize these for me? I need an example of this because I am doing this immediately.  

AZEEM AZHAR: Okay, fine. Right. So what I do is exactly that. It is word salad coming out of my head, it’s completely disordered. It’ll be three points about the proposal I’m sending out, then it’ll be a couple of things about a contract, then I’ll go back to the proposal and I will say to the LLM, reorganize that so it makes sense. Put it in bullet points in a structured way, thank you—I’m always very polite to these systems—and it’ll go off and do that while I’m driving, and then I’ll get out of the car and I’ll go to my desk, and it’ll be ready for me to act on. And I get it out of my head.  

MOLLY WOOD: Amazing work hack. Are you comfortable sharing any things that you do with your normal life? For example, I just asked AI to help me plan my son’s 18th birthday party, I’m sorry, kiddo. I just need a little advice here.  

AZEEM AZHAR: Oh, congratulations to you and to your son. I actually have found this as the first technology in 25, 30 years that has given me time back. I get so much done, so quickly. I get tired, the machines don’t get tired. And I have had time for hobbies and reading and not being at my desk, even when I get really busy. So that, I suppose, is the way it’s impacted my personal life.  

MOLLY WOOD: Are there opportunities and challenges for AI at work that we did not touch on? What have we missed?  

AZEEM AZHAR: You know, I think the big challenge is going to be that it really breaks the assumptions and the boundaries of what someone’s job is. And so in organizations that are not fluid learning organizations, which is the case for many, it’s going to really, really challenge the nature of a person’s work, why they work, how they work, and where that handoff to the next person is. And I think that’s really going to be a question that companies will struggle with and wrestle with. And it’ll be a few years before we know what the answer is.  

MOLLY WOOD: And then, finally, if you fast-forward three to five years, what do you think will be the most profound change in the way we work?  

AZEEM AZHAR: That’s an almost impossible question because the speed of change is really, really dramatic. What I would hope is that we are able to rehumanize many aspects of work and allow people to spend time in areas which humans really enjoy, and that will maybe in the social dimension, it may be in the creative or strategic dimension, or it might also just be in the dimension of ownership and getting things done. If these tools do enable that degree of productivity, that might be how work gets reshaped. I mean, getting there will be complicated because there’ll be a lot of transition, there’ll be a lot of companies that will fail, there’ll be companies that succeed. But we’d hope that the work will end up being more human and less mechanistic. 

MOLLY WOOD: I love that. Better humans, thanks to machines. Azeem Azhar, thank you so much for the time.  

AZEEM AZHAR: It’s my pleasure, thank you.  

MOLLY WOOD: I think we can all agree, that was a great way to kick off this new season of WorkLab. Thank you all for joining us, and keep checking your feeds. We have more fascinating guests on the way with actionable insights that can help leaders develop an AI-first mindset, reorient their business for an era of abundant expertise, and maximize the ROI of AI. If you’ve got a question or a comment, please drop us an email at worklab@microsoft.com, and check out Microsoft’s Work Trend Indexes and the WorkLab digital publication, where you’ll find all our episodes along with thoughtful stories that explore how business leaders are thriving in today’s new world of work. You can find all of that at microsoft.com/worklab. As for this podcast, please, if you don’t mind, rate us, review us, and follow us wherever you listen. It helps us out a ton. The WorkLab podcast is a place for experts to share their insights and opinions. As students of the future of work, Microsoft values inputs from a diverse set of voices. That said, the opinions and findings of our guests are their own, and they may not necessarily reflect Microsoft’s own research or positions. WorkLab is produced by Microsoft with Godfrey Dadich Partners and Reasonable Volume. I’m your host, Molly Wood. Sharon Kallander and Matthew Duncan produced this podcast. Jessica Voelker is the WorkLab editor.