Cognitive organizations are revolutionizing how businesses operate. Technology thought leader, entrepreneur, and author Kenneth Corrêa, whose groundbreaking book Cognitive Organizations was published by MIT, reveals the profound impact of AI on productivity, skill-building, and workflow management. He shares real-world examples and insights into how businesses are currently leveraging AI agents to automate tasks and optimize processes.

Discover the critical balance between human intelligence and artificial intelligence in this rapidly evolving landscape. Kenneth also candidly discusses the challenges and opportunities leaders and individuals face in adopting these transformative technologies, emphasizing the need for continuous learning and strategic experimentation.

Check out the full series of Career Sessions, Career Lessons podcasts here or visit pathwise.io/podcast/. A full written transcript of this episode is also available at https://pathwise.io/podcasts/kenneth-correa/.

—

Watch the episode here

Listen to the podcast here

The AI Revolution And How It Is Creating Cognitive Organizations, With Kenneth Corrêa

Welcome To Cognitive Organizations: The AI Revolution Begins

I’m J.R. Lowry. This is Career Sessions, Career Lessons, which is brought to you by PathWise.io. If you’re ready to take control of your career, join the PathWise community. Basic membership is free. My guest is Kenneth Corrêa, a technology thought leader, entrepreneur, and author of Cognitive Organizations, which was published by MIT in 2024. In this conversation, we’re going to be focusing on the concept of cognitive organizations, and more broadly, how AI is reshaping the world of business. We’ll also touch briefly on Kenneth’s broader career journey. We have lots to cover, so let’s get going.

—

Kenneth, welcome. Thank you for joining me on the show.

J.R., thank you for having me. It’s a pleasure to share a little bit about this topic. We’re going to talk about AI, which is a hot topic.

That was certainly an appeal for me. As somebody who has worked in the technology space for most of my career, it’s hard to keep up with that because there is so much happening. Apart from just the technology implications, there are also broader implications for how people think about skill-building and their careers. I want to make sure that we get into that. Before we get into that and the book that you’ve written, Cognitive Organizations, give us a quick background on you.

I come all the way from Brazil. That’s where I’m at right now. I don’t live in São Paulo or Rio, the most famous cities in South America. I’m on the border of Brazil and Paraguay. That’s the West of Brazil, right in the heart of South America. You’re going to say, “How are you talking about AI being from this place?” I had a very early contact with technology. I learned to program and type before I learned to handwrite. I had an early computer because of my family.

Kenneth is not a Brazilian name. I have origins in America. I have a grandmother from New Jersey who had six children. One of them is my mom. We were always very connected to whatever was happening in technology. Having had this contact, when I went to university, I studied management. I liked both those topics, technology and management. Further in my career, I came to realize that those are converging areas. The largest companies in the world are technology companies. You can even imagine a company working in this era of information, or the cognitive age we’re going to talk about, without imagining the use of technology every day.

I own a company called 80 20 Marketing. It’s a 95-people company. We develop marketing and tech solutions for the largest companies in the world operating in South America. All the way from Rio Grande in Mexico, down to the bottom of Argentina, there are companies, our clients, and our employees walking around, making things happen, launching products, capturing data, creating insights, and developing solutions on top of those insights.

To your point about being in the Western reaches of Brazil on the border of Paraguay and not in São Paulo or Rio, the reality is that in this day and age, skills can be anywhere. Knowledge can be anywhere. The world is completely connected with a few exceptions, but you’re a living proof that work can be anywhere. Client bases can be anywhere. This idea of creating this almost fully remote, dispersed workforce, if you manage it, it’s doable. I think it’s harder to manage them, but it can work.

The World Is Flat: Location-Agnostic Work In The Digital Age

Years ago, Friedman published The World Is Flat. Maybe not in 2015, but in 2025, we can say that it’s easy to be in touch and to have your fingers on the pulse of whatever is happening everywhere. You must know where to look and who to follow, but I’m pretty sure we can be very well connected. That doesn’t mean that being in business hubs like London, New York, or Silicon Valley, where I returned from, makes a difference when you want to deploy that knowledge, when you want to build a company, and when you need talent, for example. That’s still a lot different. As for acquiring knowledge, I can completely agree. The world is now a little bit flat. That’s not the flat Earth theory, but it’s easier. It’s more in equilibrium.

I remember reading that book. I was reading it on vacation with our family. We’re sitting by the side of the pool. I want to say it was on a trip to Florida, probably one of the times we took our kids to Orlando, to Disney World, and all of that, hanging out at the pool. The kids are having fun. I’m sitting there in the shade, trying to avoid getting too much sun and reading what was a pretty dense book. I remember reading it on one of those trips. You do a bit of teaching, too, right?

Yes, I am. I also teach emerging technologies at a business school. It’s called FGV. It’s the leading business school in the country. I travel everywhere. Brazil has 27 states. It’s half the states of the United States. I’ve been to 21 of those. I’m teaching emerging tech. My topic is a moving target. As of right now, I’m teaching multi-agent systems. How can you connect multiple AIs to develop a workflow to deliver a product or a service? Two years before that was generative AI. Two years before that was the metaverse.

Two years, even before that, was blockchain. I can get back to big data. If I go two years, the mobile revolution and e-commerce, every cycle of around two years, there’s this new next big thing, something that is going to solve every problem for humanity, is going to cure cancer, and is going to solve poverty and hunger, which never happens. I don’t know why there’s a hype cycle that Gartner has been promoting for some time now. All that technology will reach the peak of inflated expectations. That’s where the multi-agent systems are right now.

Unpacking Cognitive Organizations: The Genesis Of The Book

Use that as a way to get into talking about cognitive organizations and the book on that topic that you wrote in 2024. What was the tipping point? Was it this emergence of agentic AI, or was there something else that led you to write the book?

Being with the finger on the pulse and also experimenting with this emerging tech, I was able to see a vision of AI agents, and especially the idea of multiple agents. For example, Salesforce can call a customer service chatbot an AI agent because it’s an AI customer service agent. That’s not what I’m talking about. I’m talking about the idea of getting the org chart of the company or a flow chart of processes, and then using both human and AI agents to do atomic parts of a process as a whole.

I had a vision back in December ’23. I was doing my first test with a piece of software called AutoGen by Microsoft. It didn’t go very well. I ran into an open source alternative called CrewAI, which I found out was developed by another Brazilian, but I found that out later. It’s called CrewAI, where you can create a crew of AI agents. The software will orchestrate the communication between those agents. It is in the same way that a company has a boss. This boss supervises three different professionals, and then you need somebody to do your market research, so you call the researcher. You need a salesperson, so you call the salespeople. The same orchestration that a boss does can be done by software and can be done with the aid of artificial intelligence. Those were my first contacts.

All year of ’24, I was deploying these solutions for companies, which are my clients. I have clients in the health sector. I have clients in the telecom business. I have clients in the meatpacking and agricultural business. I call those guys. Usually, the CTO, somebody in connection with technology. I said, “There’s this new tech here. I think it can help you solve problem A or problem B.” Sometimes, they call me back, and then we get it up and running.

I’m seeing not only the idea happening, but especially learning the downfalls of the technology itself. You have the positive and good side, but you also have the downside on the other side. When I saw all that happening, I said, “This is a new way of doing business. This is a new way to manage a company.” The idea that we’re going to manage both human beings, which we have been doing for some time now, and AI agents deserves a chapter in the history of management thinking. That’s why I call it the cognitive organizations.

I use the date of November 2022. That’s when ChatGPT was launched. That’s the milestone for a new era. As a professor in Brazil, a business professor of emerging tech, I do in-company training. I’m always connected to these big companies and seeing how the coming of the large language models, the Copilot for Microsoft, Gemini for Google, and ChatGPT itself, is changing productivity with numbers that are very hard to ignore. If you were talking about a 1% or 2% productivity boost, that was average. We’re talking about an average of 20% and 40% productivity. It’s something that is changing how business operates around the world.

Let me play back, because in a way, the difference between a cognitive organization and the way that you’re describing it, and a company that’s using tech and digital, is this idea of weaving together the processes that a machine can now do and the process that a human still needs to do, but putting the workflow management over that. This is your point from a minute ago about it’s more than the chatbot that Salesforce uses, which is a point solution agent.

Kenneth Correa: The orchestration a boss performs can now be done by software, aided by artificial intelligence.

It’s taking a bunch of those things that perform different functions, stitching them together with the human processes, and having the workflow all managed for you. This is where Gen AI, robotic process automation, and things like that get taken to the next level. Now, you’ve got much more broadly spoken automatic workflow management. You can easily see how, in labor-intensive and knowledge-intensive processes, you could cut out 20% to 40% of the effort involved in doing them.

The Meatpacking Metaphor: AI Agents And The Human In The Loop

Sometimes, we’re going to see the discussion. People will say, “It’s going to terminate this job, so people will get fired and all that.” When I’m talking to real companies with real problems, I don’t see them wanting to let everybody go. That’s not the way that they approach. Let me tell you a short story that will elucidate. I have a client. It’s a meat-packing factory, the largest player in the world. They’re the largest in Brazil. They have 30% of the North American market, selling cattle meat.

Those guys have a lot of different plants for processing the meat. They wanted to create a new factory. This was supposed to be their flagship factory with all the things like technology, sensors, analytics, big data, and all that. They bought the most advanced equipment available, some from Germany, some from China, and some from the United States. They put all that together in a factory, press play, and start running. The sponsor, one of the people who paid for that project that wanted the flagship, comes into the place and says, “This looks like every other factory. There’s nothing different here, and I paid an extra $20 million to get something different in this space.”

The CTO called me and said, “Kenneth, we had a problem.” I said, “What happened?” “We have all the machines available, but we don’t have the data scientists available.” What happened in Brazil is that whenever people started learning how to develop and write code, they started learning how to analyze data and generate insights out of data. With ChatGPT, they can speak English or at least write emails in English a lot more easily. They’re all working for American companies. They’re all working for European companies. We have a lack of talent in that space in the country.

I say, “I know this piece of tech. It’s called multi-agent systems, where we may try to create agents that will do whatever you would need the data scientists to do.” I’m not talking about replacing anyone. I’m talking about being able to have a deliverable as a part of a workflow done by these large language models. I like that you mentioned RPA because robotics process automation is something that we have been doing for some time now. It’s all about clicks and filling out forms. With AI agents, we go a step further because they can understand context. It can access databases and information.

What we did was create four different agents. Three of those specialize in each of the machinery. For example, one of those was the conveyor belt. With the conveyor belt, you know the weight of the meat that’s being carried out. This data is being generated by the machine. It does all by itself. They have one specific type of database, the Oracle database. They have this specific encoding for the data files they’re generating.

You jump to another machine. It’s a shadow of light. You can know the color of the meat, which is important for quality assurance. This machine with a big light bulb runs a SQL database, another piece of tech. The encoding is not the same as the conveyor belt. The last one is weighing the pieces of meat because you need to have the right weight at that very point in time. What we did was we trained one AI to be a specialist in each of those machines. Every fifteen minutes, the fourth agent, which is the orchestrator, will communicate with those three agents. What the agents do is they connect to the database, read the data, and say, “This is inside operational limits, top and bottom.”

If everything is all right, there’s nothing you need to do. Let’s say the conveyor belt specially says, “There is something off here. I see that the weight is 8% larger than it should be.” It sends a message to the orchestrator. We connected the orchestrator. Everything here is 100% AI. We connect this orchestrator to the human in the loop, to the human engineer. It’s a floor engineer. That’s how they call it. It is the person who is on the floor of the factory, taking care of everything, and sends a message like, “It’s 8% off.” The human in the loop decides and takes the call if he’s going to act or not, on top of that information. This is the workflow where you have AI agents doing whatever AI does best, which is analyzing lots of data to compare it to parameters.

You have the human in the loop, the human being, doing whatever humans do best, which is making a judgment call about what to do. The process is not interrupted, so it can fix the machinery. It’s all computer, code, and all that, but you can have this information coming from the agents. One special information, this was pulled off in two weeks. We’re not talking about a six-month project. We’re not talking about developing this code that becomes hard to maintain. This is very quick and fast to implement. It is now doing this part of the job right. This is the situation where I’m getting calls from companies wanting to implement multi-agents in their companies.

Co-Pilot To Autonomous: Navigating Levels Of AI Integration

I want to come back later into the discussion about what this means in terms of re-skilling and all of that. Just dive a little bit deeper into the ways that people can apply these agents. You talk in your book about copilot, delegating, and autonomy. It would be helpful to give the audience a bit of background on what the differences are there in terms of how much you’re turning over to an agent to do for you.

Step number one would be the copilot. It’s what they call the augmenting tools, where you have a person still in charge of every activity they usually have to make for their service. They are implementing different AI tools. I have a list of AI tools that we already implement and use every day. There’s a list of tools that augment people’s work. We’re talking about ChatGPT, which is amazing. In emails, you don’t need more proofreading because they’re reading crystal clear, plain grammar. It is amazing in that sense. We have brainstorming sessions for marketing companies where you can talk to these assistants. You can get ideas back from what you’re doing.

You have image creation tools. Whenever you need to illustrate stuff and whenever you need to create ads, we do that a lot in the marketing space. You can use these tools to do that faster. You don’t need to follow the whole process of having to do everything by hand. This is the first step that I see lots of companies doing. The most aggressive companies were doing that back in ’23, when ChatGPT was out and everybody went looking for those tools. There is a difference in the rhythm of adoption.

A lot of companies in 2025 are still asking me, “Do you think we should let our employees use ChatGPT?” I’m like, “I don’t know. You should force them, maybe. You should have been doing that for a long time now. You should have already implemented those tools, so you would figure out where they’re needed for the next step, where you start to create some automations.” Whenever we talk about AI, if you think of where AI came from, we always think of sci-fi. We think of robots and AI pilots running everything, but it’s not where we’re at.

I know that OpenAI has been promising the idea of AGI, the artificial general intelligence, a piece of software that will be able to understand every situation and every scenario, and make the best decision. That’s not how we technologists look at it because what we see in practice in real life is that those are very useful but limited systems. They have what they call a narrow intelligence. If you are to feed it the correct information, you’re going to be able to process that, analyze, and make a decision, for example. Usually, that’s not how companies operate. We usually operate on top of missing information.

We have to fill in blanks and gaps, which humans do fairly easily. For computers, it’s not good, even with the large language models there. This next step will be the idea that you start integrating parts of the process where you get computers and humans working side by side. Generally, broadly, that’s the stage we’re at right now. Companies began to realize that it’s inevitable, or at least it would be very stupid for the company not to integrate these tools. We’re talking about productivity gains, as I mentioned before, of 20%. Let’s use 20% as a number. If a person works five days a week, 20% is one less day a week to do the same work they needed to do before.

This is something very hard to ignore. Companies are looking into how they can automate this process, combining that with human beings. Still, we’re talking about delegating the decision to the human in the loop. There is a third step, where it is autonomous. J.R., as of my experience, I’m not seeing autonomous systems work the way we have been promised. Every time that I leave it 100% for AI, two hours later, everything is nuts. It’s crazy. If you use ChatGPT for a while, you will see that it hallucinates, which is a technical term, but a very vivid term to explain what happens. It creates information that is not true. Sometimes, it’s not even based on anything that happened as a fact.

You imagine that in a workflow. If one of the agents makes a mistake or hallucinates, this message is sent to the next agent. This agent is not trained to check if this is true. It will decide based on top of what it received from the third agent, and then, following on and on, the decisions go completely haywire. I’m not seeing these autonomous systems work, or else we’d already have something that Sam Altman has been promising. He said back in ’23 that he knows that there will be a unicorn, a billion-dollar valued company, consisting of only one person. In reality, I know he has a conflict of interest because he’s selling his technology to promise that, but it’s how I see it. The delegation is the right spot we’re at right now.

The flaw with that idea is that if one person could create it, it’s not worth $1 billion. We’re in this intermediate stage. We’re moving beyond the point of having these point things that are doing chatbot, robotic process, automation into more interconnected things. We haven’t, in most instances, fully turned over the keys to them. It feels like, though, as it happens with a lot of other technologies, we’ll get there eventually.

It is probably in the not-too-distant future, at the pace things seem to be going, where you say, “I’ve got complete trust in this process, with an orchestrator and these individual agents to run this process.” The downsides of them getting it wrong are pretty minimal. It feels to me like the only thing that may hold us back from complete autonomy is legal concerns. If something goes haywire and you don’t have a human in the loop anymore, you’re exposing yourself to some form of legal risk because you’ve let the machines run rampant, if you want to think about it that way.

Being autonomous is not a black and white thing. It’s shades of autonomy or levels of autonomy, like an autonomous car. I was in San Francisco using Waymo, Google’s autonomous vehicle. You call it. Use your phone, call on an app, a car arrives, no driver, and you get in. It takes to wherever you need to go. You get down, and it goes to the next trip. When you look at it, it seems 100% autonomous. It’s in San Francisco. There’s an office by Waymo where there are human beings looking at images in the camera and looking at data being generated.

Not in any trip that I took, but sometimes, they will assume control of the car. Sometimes, they will ask the passengers to get out of the car as well, because there are still human beings as part of that process. It’s not 100% autonomous in that sense and for legal reasons as well because it’s a technology still under development. Getting back into the company and the environment, the business space, when we’re designing the system, we always look at it from where AI, computers, technology, systems, and logic will fit best, and where human beings will fit best. There’s this large Swedish company called Klarna that does customer service for a lot of different companies around the world.

They went all in, 100% automated. They took a step back. They found out the hard way, painfully, that 20% of the work still needs to be done by human beings. Having the ability, the empathy, and judgment calls that have to be made are a lot better when you put a human being in the loop. That means that 20% is the exception. It’s the case that runs out of the norm. In 80% of the cases, an AI agent, a chatbot assistant, solves 100% of the problem. It’s always good to design, and that’s why I call it the cognitive organizations, a combination of human and AI, human intelligence and artificial intelligence, because that’s when you’re choosing which is best for each scenario that we’re working with.

Kenneth Correa: It’s always good to design cognitive organizations as a combination of human and AI—human intelligence and artificial intelligence—because that’s when you’re choosing what’s best for each scenario you’re working with.

The Cost Of Inaction: Why Companies Can’t Afford To Wait

You’ve got companies that are doing this well. You’ve got companies that are not doing this well or not doing it at all. You make the point, and I can quote other people as well who made the point, that being behind here is bad because you may never get caught up. What are the ones who are doing right? What are the ones who aren’t doing that they should be doing?

When we talk about big tech, like Google, Microsoft, and OpenAI, there’s a topic that is not a big tech company, but it is a Silicon Valley company. In this space, everybody is doing the best they can to try to keep ahead of the LLM race. This is a scenario for those very specific companies. When we jump into the other companies or the realm of more regular companies, which are tech companies sometimes, but not exactly tech companies, what I see is that not doing anything has never been so expensive.

Doing it right or doing it wrong is still better than not looking at it. That’s the first thing that I would like to mention. There’s a logistics company called C.H. Robinson, one of the largest third-party logistics providers in the world. They operate multiple naval. They use trains, trucks, roads, and all that. Those guys automated the process of quoting. As a logistics provider, they will get phone calls or emails from clients asking if they can get a product from Portland and take it to Orange, California.

How much would it cost to make this move? When are you available and all that? This is a process that had to be done by human beings 100% of the time. They will need to get a call from the client, then call the truck drivers around the country, get a quote, define a price, call back the client, negotiate, get the best price, and call back the truck driver. They realized that this process was very mechanical in that sense.

It was people taking phone calls and making notes. They were able to automate it. It took them more than a year to put that into production. Some companies will quickly go into adopting the technology, looking at it, studying it, and creating prototypes. This will not necessarily come to the consumer. C.H. Robinson is a case where they were quick at it, but they saw the downsides. They were like, “I don’t know if we’re doing that or not.” When they were able to get a part of that process, which is getting the requisites, the phone call from the clients, and talking to the truck drivers, or so, then they were able to implement them.

They’re doing 3,000 phone calls a day with their AI system. That’s approximately 10% of what they do. They managed to get 10% done by automated processes. It’s going to be easy to jump into 20% and then 30%. I don’t know what is going to be the percentage will be at the end. This is an example of a Fortune 500 company that did a good job, but it still took them a year to get it applied. With every change in an organization, there are a lot of things to learn. Quickly, as we start learning. That means as quickly as we start making mistakes, that’s how fast we’ll get to getting it right.

From Decision Makers To Chief Experimenters: A New Leadership Paradigm

You described at one point the idea that leaders in these companies are going to need to shift from being decision makers to chief experimenters. To your point, you’re not going to always get it right on the first try. You’re going to need to have that iteration. It comes back to some of the things that were in the lean startup, minimum viable product, and minimum lovable product. I think you called it at one point in the book.

It is to get something good enough that you can learn from and then rapidly iterate from there. Companies have struggled with doing agile development. This is agile development not just on tech, but on the whole business processes. It feels like that will be difficult for a lot of companies to make that mindset shift. What are you seeing in terms of what it takes to get there?

Jack Welch has a phrase that I love. He coined this phrase. He said, “If things are happening outside of the organization faster than they’re happening inside, you’re in trouble.” It is the scenario where we’re at. You started our conversation by saying that a lot is going on, and it’s exactly like that. I’m studying it a lot. I’m trying to get a hold of what’s going on, but I’m missing 99% of everything. I’m trying to be up to date.

The idea of trying and being close to the conversation to look for the successful use case and learning from the failures, that’s the way to go. It’s not 100% certain. A hundred percent of the time, you will not get it right the first time. It’s never like that. I’m talking to companies all around the world, even using the Copilot and even using Google’s Gemini, it’s hard. For the legal area of the company, it may work perfectly. You talk to the people at operations, and it’s terrible. For example, in 2024, those AI tools, Copilot tools, were very good with text, but they were horrible with numbers.

Whenever the engineering area or architects were using the tools, they said, “This is crap. This is not useful at all.” After this last Google Gemini update, Google Gemini is amazing with numbers. You can provide CSV files and spreadsheets. You can even print your dashboards. It’s able to capture information from the CSV files and from the data files. We’re talking about a tool that’s useful for those departments. How will a company find that out? They have to be testing. They have to be trying to implement that.

If the posture of the company is “Let’s wait,” because a lot of companies will go, “Let’s wait until it gets more stable. Let’s wait until it’s perfect,” it’s going to cost them a lot. I don’t know any sector that has no competition. If you are in a scenario where you have no competition, maybe you can go 20% or 40% lagging behind, but that’s not the case for most companies. You’re in very competitive environments. Anything that you do not do, but your competitors are all doing, is going to get you back in that race.

One of the things that makes this hard with any technology is this. You referenced big data and the metaverse, and you didn’t mention VR, but we could throw that one in there as well. It is all these things that have come over. ChatGPT came out. Within three months, there were a million different startups out there saying, “We’re harnessing AI to do X,” which is a thin veneer on top of ChatGPT. I guess my sense would be, if it’s a thin veneer on top of ChatGPT, I don’t need it because I can go do that myself.

The challenge for companies is that when you need to go a bit deeper than that, and you need some particular technology, and there are hundreds of companies out there, you can burn a lot of calories with the wrong partner. To me, that’s the hardest thing about this space. In any new technology, if you’re a decent-sized company, there will always be hundreds of companies lining up at your door who want to get in and talk to you about how their technology can help you.

“Just let us do a proof of concept. We’ll only charge you X amount of money for the proof of concept,” but the thing you burn is time. That’s the thing that makes it particularly hard. “We can do it in two weeks.” It turns into a month. It turns into three months. It turns into six months. At some point, you have to say it’s not working, but at that point, you’ve burned time. In my own experience, that has always been one of the hardest things about trying to bring in new technologies. It is figuring out, “Am I working with the right partner, or did I pick a dud?”

I completely agree. That’s the current scenario we are in. The way the company could approach it to try not to lose that time, even if they have to change providers or change partners, is to be agnostic between those different models. What happens in the LLM race is that Google Gemini 2.5 Pro is the best in most benchmarks. I don’t know. Maybe while we’re talking, there’s a new release. I know GPT-5 has been promised for some time now, and then it’s probably going to crush everything. Grok by X released version 4. It’s not a lot, orders of magnitude better, and better than version 3 that they had before. You have to be able to move between those models easily.

The second thing that is a mindset change, which is very important, is that we’ve worked with companies in a time where software was considered IP, intellectual property. You write code, you hire a company, and then you want to own that code because it’s your code and all that. We are at a stage of software development, where it is so easy and so fast to deploy new products that we started not caring a lot about the code. The database, the data, is important. You have to have 100% control of that. Being able to change software, to change from Salesforce to HubSpot or Microsoft Dynamics, has to be an easier transition when you focus on getting the data right. This change between data versus software is a way to go to try to alleviate a little of what’s happening that you described perfectly.

Kenneth Correa: We are at a stage of software development which is so easy and fast to deploy new products that we started not really caring.

I don’t think we’re anywhere near that, though. You think about how hard it would be to leave Salesforce, to move off of any big system, SAP, Salesforce, any of them, onto their competitors, Workday or whatever. It takes a long time because those things are typically heavily customized. They’ve been bandaided together with all of your internal systems. In terms of rolling out these new things, the AI will be ahead of where the data is. What AI does is prove to you that you’ve got all sorts of data issues. It proves that much more quickly than a human could.

Below that is all your underlying systems infrastructure, the clunkiness of it, and the difficulty of changing it. To me, if you think about trying to hive off these pieces and take an entire process, the workflow, the data, the underlying infrastructure, and hive it off so that you can fully make it autonomous, it’s hard to do that because the software industry has set itself up for high switching costs. It makes it hard to undo that data. Databases are getting better.

There are conflicts of interest. Companies are trying to get you hooked on their product and don’t want anything open about that. I spoke with Bosch, for example, a global company in the automotive sector. They took a data-first approach in the sense that they have providers and solutions for dealing with data, but every software that they acquire has to be placed on top of their data lake. It’s a data-first approach in the sense that the software has to adapt to the data, not the data has to adapt to the software. That’s a good paradigm to begin with.

Talk about bringing the processes to the data as opposed to the data to the processes. It’s a similar construct. Let’s talk about some of the people implications. Let’s start with leaders. We talked a little bit about the importance of being willing to try, experiment, and learn. What are some of the other things that this shift is going to mean in terms of what it takes to be a successful leader in the future?

Personal AI Adoption: Leaders, Get Your Hands Dirty!

First, it is impossible to see the potential, to see how 20% increase in productivity gets real, if you’re not using the tools yourself. What I try to talk to business leaders all around the world is that you have to get your hands dirty. It’s not about asking for the IT guy to learn all about it. It’s not getting your assistant to do stuff for you. You have to apply it to your everyday life. By saying every day, I mean not only your professional life. It can be in your personal life as well.

Get a phone, install ChatGPT or Copilot, whichever of the tools, and then you get to a refrigerator. It’s a Sunday. You don’t want to cook anything, but you have to eat. You can take a photo of the refrigerator and say, “With these ingredients, what could I cook in less than five minutes?” You then get a reply. Whenever you see these responses that are customized and tailored to your photo, the ability to understand an image, find instructions, and suggest to you what to cook will show you how this could be implemented. Getting close to the tools is a very important first step.

The second one, and it’s something that we covered here in the conversation, is remember that you always have humans in the loop. I got a lot of phone calls from executives. They call me and say, “Kenneth, I need AI to solve this problem.” What they’re trying to solve is unsolvable. It’s not something that you have like, “Let me get that AI for myself and then let’s solve it.” It’s not like that. You have to think of activities. You can prioritize the activities that you’re going to try to make autonomous or try to implement AI.

I do recommend that companies look for the most boring and repetitive tasks in the company. We’re not trying to make an automation to do the strategic plan for next quarter or for next year. You may have accounts receivable and accounts payable. You have operations that you need to fill in data every day. There are people typing stuff in the CRM tools, or you have to download files from the bank and upload them to another. Those are the tools that you should look into automating because they are easier to do.

Remember that the results that you get are directly proportional to how many times that task happens, because sometimes, it will take you 40, 80, or 100 hours to automate a process. If you’re automating something that happens once a month, how much time are you getting back? It happens 5,000 times a day. You then get your 100 hours back easily. It is looking at it from a task perspective, not a job, and prioritizing whatever you want to automate in that sense.

The third thing would be to be open to the possibility of mistakes because these generative AI tools and these generative AI solutions are still not perfect. You’re not going to get it right from the start. We said that before. Also, you have to deal with the fact that mistakes will be made. You cannot compare AI to perfection. You should compare AI to non-AI. If your employees are getting a 90% performance, and you get 91% with AI, you’re already fast. Do not compare it to 100%. Compare it to the 90%, which is the process you had earlier.

I have a funny story about that. I work with a big law office. Brazil holds 70% of the work-related lawsuits in the world. Seventy percent, one country, that’s Brazil. Those guys work in that space, and they wanted to use AI because it’s a volume. They work with the largest companies and with the government sector in Brazil. They wanted to automate that. They have 600 lawyers working in the company. They create this system.

Whenever there’s a new suit that they have to work with, they will input or upload a file with all the data from that suit. The system will see which judge it is, what the court is, what the size of the ask is for that process, and who the parties are. Based on that, it creates a copy. It’s called an initial, the first response from that company that was sued. From the 600 employees, they were all using the system, and they made changes. They get a draft. They work on the draft. “Let me change that. I didn’t like that wording or those phrases.” They made changes.

Some of the employees were lazy. They didn’t take the time to mess with the text. Whatever the system provided, they just sent it in. What happened is that when we started comparing the performance, the lazy guys had the highest performance because they were not touching what the AI wrote before. Those very proactive but not so performative lawyers were doing a disservice to whatever the AI was creating. This is the importance of measuring and having a KPI so you can measure the results that you’re having from the start. You develop the solution, thinking about how you are going to measure and how you are going to compare it to whatever was done before.

In my own company, I told you, 95 employees. We are a creative economy company, so we write code, we create content, and we analyze data. We are doing a lot of tests that AI can help with. What I did first, a little bit before ChatGPT was out, we were already using these tools because when I saw Generative AI, my head exploded. I knew this was going to be a straight shot into my own business. I went on to learn about it. Now, I teach all my employees how to use those tools.

The tech-savvy guys were fast to use it. Other people were a little bit more conservative. They didn’t like it either. No problem. I taught them all. We have a time sheet in my company. We always know how much time every person is spending on each project for each client. I started comparing the time to complete a task between people who were using AI and people who were not using AI. I have a 46% difference between people who use AI compared to people who don’t use AI. This is the boost that we can get.

Economic Dislocation: Reskilling And The Future Of Work

It’s hard to see how this doesn’t create economic dislocation because you’ve got your meatpacking example, your law firm example, and your own company’s example. When you’re driving that much productivity improvement, one of two things is going to happen. Either people are going to lose their jobs, or they’re going to need rapid reskilling. One isn’t good. The second one is hard because we’ve already seen it in the past X decades.

As work has been mechanized by robots and as it’s been offshore to different countries, people who get left behind are struggling to keep pace, even with what’s happened so far. I have a hard time seeing how this doesn’t exacerbate that challenge for people who are living on the margins of being able to keep pace with what’s going on in the world. This makes it worse. I would love to get your view because I’m having a hard time not being worried about the near-term fate of the world.

I met with two of the three economy novels from 2024. They wrote Power and Progress. They touch exactly on that topic. Remember, the Nobel Prize is a political prize. It touched exactly on that point because the committee saw that AI is advancing very fast. Whatever you just described is already happening. It’s not something that will happen. It will happen even more. It will be more intense, but it is already happening. When you compare a person who loses her job because of automation, and now, she’s out of the market, it’s hard to get that person back.

If I change the focus to my own company, I see it happen inside my own company. This is a company where we have a culture of innovation. Whenever we selected people in the past, we selected the ones who could use and adopt new technologies and new tools because those tools are changing all the time. Even in my close scenario, I can see that difference in my own company. The way we try to solve it is not a perfect solution, but we increase and intensify training. As you said, it is something very hard to do.

We had to let some people go because some people were very negative towards the technology. They didn’t want to use it at all. At first, it was easy for me to say, “No problem. Don’t worry. This part of the team will use it. This other will not use.” Once the 46% number is out, it’s harder to ignore it. As a company, we’re trying to increase profits. We’re trying to grow the company. We’re trying to get better margins, and this is a direct impact. The genie is out of the bottle.

This AI revolution is already happening. What I’m trying to do as a business person is train my team. What I try to do as a professor is, as you could make sense of that, to help people understand that this is going to be the status quo. The way to deal with that is you will have to always be learning. This is something that at the company level is hard. At the society level, I have no idea how to solve it. I’m sorry. Even the Nobel Prize winners, I don’t think they know it as well. This is an equation we’re yet to figure out how we’re going to deal with because it’s happening fast.

The Future Is Now: Embracing AI For A Competitive Edge

I’m working with companies that are using the humanoid robots, not the Tesla, because Tesla is just promising, but we cannot see them all. Unitree Robotics from China, for example, has these humanoid robots. Figure from Silicon Valley has humanoid robots inside a BMW factory in Germany, already doing parts of the work. It’s not only AI, but it’s getting physical. This is going to become the common reality faster and faster. It is what it is, and we’re trying to deal with that.

I was going to ask you a question. I’ll let you answer it in a second, as it relates to the last question about final advice you would give to people who want to make sure that they don’t become irrelevant in an AI-augmented world. At the end of the day, it goes back to what you’re saying. You’ve got to commit to learning. If you don’t make peace with the fact that the genie is out of the bottle, that the pace is happening quickly, and you let yourself get left behind, you’re like a little person version of what we talked about earlier in the conversation about companies that figure that this is just a passing fad or something like that, and don’t want to get involved. You’re going to get left behind. It’s going to be hard for people to keep pace.

We saw the revolution of the internet. I’m pretty sure that a lot of journalists, for example, would say, “I don’t like the internet. I don’t like the information that’s coming out of there. I don’t like Wikipedia or whatnot.” In 2025, it’s very hard to think of a journalist working without the internet. Generative AI, in that sense, is what the internet represents. What I think, and in some ways I hope, is that people will acknowledge that faster, or else we’re going to have a lot of journalists trying not to use the internet.

We’ve lived through the internet era. We go back to the late ’90s. We’ve managed to see a lot of these changes come into the fold, the internet, the web, making e-commerce sites, turning into social media, the Apple rise, the shift into mobile devices, and all of these things. We’ve still managed to keep it to a point where it hasn’t blown up the economy. I guess the question is, do we keep on rolling in that way, or is there a blow-up coming? It’s hard to tell, but ignoring it is certainly not going to be the way out of it.

As a management professor, I have a lot of students who are in the market out there. Some of them are displaced by these automations and AI. What I try to talk to the people that I get in contact with is that you have to understand that there are two sides to the story. You have to choose which side you’re in. I know that’s not a society solution because there’s not enough space for everybody, but at least for the ones thinking to get out there, to get their head out of the water, that’s what I’m trying to teach them.

Get closer to technology. Start using it in your personal and professional life. Implement it in your workflow, on your day-to-day. Prioritize a task that you yourself can automate as well, because a lot of companies are still blocking the use of those techs. Say you cannot use ChatGPT. I see a lot of employees. Nowadays, everybody brings their own device. They’re using it on their phones. I see that happening. That’s what I try to say. Save yourself. Get to the lifeboat that’s left. That’s how I’m trying to frame it.

I am not sure we’ve ended on the happiest note, but we will call it there, being mindful of time. Thanks for doing this with me, Kenneth. It will certainly be a thought-provoking conversation for anybody. I appreciate you taking the time.

Thank you very much. Thank you for having me again.

Sure thing. Take care.

—

Thought-provoking conversation there. I want to thank Kenneth for joining me to dive a bit into how AI is creating cognitive organizations, and particularly how this notion of agentic AI will take what we’ve seen so far with Gen AI and some of the other tools to another level. There is a lot to think about as it relates to our career journeys. That’s what this show is all about. As a reminder, it was brought to you by PathWise.io. If you’re ready to take control of your career, you can join the PathWise community. Basic membership is free. You can also sign up on the website for the newsletter. Follow us on LinkedIn, Facebook, YouTube, Instagram, and TikTok. Thanks.

Important Links

• Kenneth Corrêa’s Website
• Cognitive Organizations
• 80 20 Marketing
• The World Is Flat
• Salesforce
• CrewAI
• ChatGPT
• Microsoft Copilot
• Gemini
• OpenAI
• Grok
• HubSpot
• Power and Progress
• Unitree Robotics
• Figure
• PathWise on LinkedIn
• PathWise on Facebook
• PathWise on YouTube
• PathWise on Instagram
• PathWise on TikTok
• PathWise on Twitter

About Kenneth Corrêa

Kenneth Correa is a technology thought leader, entrepreneur, and author of “Cognitive Organizations,” published by MIT in 2024. As the founder of DataNeural, an AI consulting firm, and Head of Strategy at 80 20 Marketing, Kenneth is dedicated to helping organizations harness the power of AI agent networks to achieve transformational results. With over 25 years of experience and a global presence spanning Brazil, the UK, the US, and beyond, Kenneth specializes in bridging cutting-edge AI innovations with practical business applications. His clients include Fortune 500 companies and leaders in healthcare, finance, and technology. Through his consulting, workshops, and international keynotes, Kenneth equips businesses with frameworks to drive productivity and stay ahead in the AI era.