New Era: How Artificial Intelligence Is Rewriting the Rules of Business and What You Need to Do About It

"The question is not whether machines will think. The question is whether you will think about what to do with them."

Table of Contents

• Chapter 1: The Shift Nobody Saw Coming (And Everyone Felt at Once)
• Chapter 2: The Companies That Got It Right, and the Ones That Didn't
• Chapter 3: Your Playbook for the AI Age

Chapter 1: The Shift Nobody Saw Coming (And Everyone Felt at Once)

Let me tell you about a Tuesday morning in November 2022 that changed everything.

On that morning, a research company called OpenAI quietly released a product to the public. They did not buy a Super Bowl ad. There was no celebrity spokesperson. No flashy launch event with smoke machines and a keynote speaker in a black turtleneck. They just posted a link, and people started clicking.

Within five days, one million people had signed up. Within two months, the number was one hundred million. For context, it took Netflix three and a half years to reach that milestone. Instagram took two and a half years. ChatGPT did it in sixty days.

That product, of course, was ChatGPT. And whether you were one of those early users or you heard about it from a colleague or a teenager at your kitchen table, you probably remember the first time you actually used it. There was something uncanny about it. Something that made the back of your neck feel a little strange. You typed a question and received a response that sounded, well, like a person. A pretty smart person, actually. One who had apparently read an enormous amount of books, articles, research papers, legal documents, code, poetry, and Reddit threads.

The reaction from the business world was immediate and messy. In boardrooms and on earnings calls, executives who had never said the words "large language model" in their lives suddenly could not stop saying them. Venture capital poured into AI startups with a speed that felt almost frantic. Journalists declared that every industry was about to be disrupted. Stock prices for companies with even a tenuous connection to AI spiked. Other companies, the ones seen as vulnerable to AI disruption, watched their valuations dip.

But here is what is worth noticing about that initial reaction: most of it was noise.

Not all of it. Some of the reaction was genuinely important. Some of the fear was warranted and some of the excitement was, too. But a lot of what happened in those early months was the business world's version of a panic response. People saw something they did not fully understand, recognized that it was significant, and started shouting. That is a very human thing to do. It is also not particularly useful.

What this book is about is the signal underneath the noise. What is actually happening with AI, why it matters, and most importantly, what you can do about it. Not in some vague, inspirational sense. In a specific, practical, roll-up-your-sleeves sense.

To get there, though, we need to slow down and understand what this shift actually is. Because if you misread the nature of a change, you will almost certainly misread how to respond to it.

This Is Not the First Time We Have Been Here

Every generation gets at least one technology that feels like it breaks the rules. That feels fundamentally different from what came before. And every generation's first instinct is to either overestimate how fast the change will happen or underestimate how deep it will go. Sometimes both at once.

Think about the internet. When it first became accessible to ordinary people in the mid-1990s, the predictions were extraordinary. Some said it would make physical retail obsolete within a decade. It didn't. Some said it would eliminate newspapers overnight. It didn't do that right away either. But the people who dismissed it as a fad for tech enthusiasts made an even bigger mistake. Because while the timeline was wrong, the direction was not. The internet did eventually transform retail. It did eventually gut the newspaper industry. It did eventually change how we work, shop, socialize, learn, and fall in love.

The pattern with transformative technology tends to follow what some researchers call an S-curve. Things move slowly at first while the infrastructure gets built and people figure out the use cases. Then there is an acceleration phase where adoption explodes and applications multiply faster than anyone can track. Then things plateau as the technology becomes embedded in the fabric of everyday life, and people stop thinking of it as technology at all and just think of it as the way things work.

We are somewhere in the early-to-middle section of that S-curve with AI right now. The infrastructure has been building for years. The compute power, the training data, the algorithmic breakthroughs, all of that happened largely out of public view. The release of consumer-facing AI tools like ChatGPT, Midjourney, and others was not the beginning of AI. It was the moment the S-curve started bending upward. It was the moment ordinary people could feel the acceleration with their own hands.

And that matters for how you should be thinking about this. We are not at the end of the story. We are not even close. The applications we are seeing right now, as remarkable as some of them are, are early versions. Rough drafts. The equivalent of the first websites, which were mostly just digital brochures, compared to what the internet eventually became.

What AI Actually Is (Without the Jargon)

Let's be honest about something. A lot of the coverage of AI is confusing because it treats the technology as a single, monolithic thing. "AI" gets used to describe everything from the algorithm that decides what to show you on Instagram to a system that can diagnose cancer from a medical scan. That is a bit like using the word "vehicle" to describe both a child's bicycle and a Boeing 747. Technically accurate, practically misleading.

For the purposes of this book, we are mostly going to be talking about a specific category of AI that has become genuinely important to businesses in the last few years. That category is sometimes called generative AI, and sometimes called large language models, and sometimes called foundation models. The labels vary but the basic idea is the same.

These systems are trained on vast amounts of text or images or other data. Through that training, they develop the ability to generate new content: text, images, code, audio, video. They can answer questions, summarize documents, write emails, create marketing copy, translate languages, write software, analyze financial data, and do dozens of other things that until very recently required a human being with specific skills to do.

What makes these systems different from the AI we have had for decades, the AI that runs spam filters and product recommendation engines, is that they are remarkably general-purpose. You do not have to train them on a specific task. You can just talk to them. You can say "write me a cover letter for this job posting" or "explain this legal clause in plain English" or "find the inefficiencies in this business process" and they will give you something useful.

That generality is the key thing. It is what makes this a different kind of shift from previous waves of automation. Previous automation was narrow. A robot on an assembly line does one thing extremely well and cannot do anything else. An AI system that processes insurance claims does that and nothing else. These new systems are different. They are capable across a huge range of cognitive tasks. And cognitive tasks, the things that involve reading, writing, analyzing, reasoning, and communicating, are exactly what most knowledge workers spend most of their time doing.

This is why the business implications are so significant. Previous waves of automation primarily affected physical labor and narrow, rule-based cognitive tasks. This wave touches the kind of work that white-collar professionals do. That is a much bigger category.

The Three Waves of AI in Business

To understand where we are now and where we are going, it helps to think about AI adoption in business as coming in waves. Not every company will experience these in exactly the same order or on the same timeline, but the pattern is useful.

The first wave is efficiency. This is what most businesses are doing with AI right now, or at least what they are starting to do. They are using AI to do existing tasks faster, cheaper, or better. Customer service teams are using AI to handle routine inquiries. Marketing teams are using it to generate first drafts of content. Developers are using it to write and debug code. Lawyers are using it to review contracts. Finance teams are using it to analyze data. The underlying work is the same. The tool has changed.

This wave is genuinely valuable. Companies that execute well on efficiency gains can reduce costs significantly and free up their human talent for higher-value work. But it is also the least interesting wave from a strategic perspective, because it is available to everyone. If your competitor can do the same efficiency improvements as you, they are not a source of sustainable competitive advantage. They are just the new cost of doing business.

The second wave is transformation. This is where companies are using AI not just to do existing things faster, but to do things they could not do before. This might mean offering personalized services at a scale that was previously impossible. It might mean developing new products that are powered by AI capabilities. It might mean creating entirely new business models. The companies that reach this wave are not just faster versions of what they were before. They are genuinely different.

This is the wave that creates competitive separation. A company that transforms its operations and offerings with AI does not just have lower costs. It has capabilities that competitors without AI adoption simply cannot match.

The third wave is reinvention. This is the furthest out and the hardest to see clearly from where we are sitting today. In this wave, entire industries get rebuilt around AI capabilities. The assumptions that define how an industry works, its cost structures, its competitive dynamics, its relationship with customers, get replaced. Some industries that look dominant today will look very different in ten years. Some business models that seem unassailable will not survive. New categories of business that do not exist yet will emerge.

Most businesses reading this book are in wave one, beginning to move into wave two. Wave three is on the horizon. The strategic question for leaders today is: how do you execute wave one without losing sight of wave two, while trying to anticipate wave three without being paralyzed by uncertainty?

That is a hard problem. But it is the right problem to be working on.

The Labor Question Everyone Is Afraid to Ask

Let's address the thing that is in the back of a lot of people's minds when they think about AI in business. The job question.

Will AI eliminate jobs? Some of them, yes. Almost certainly. Any honest assessment of what these systems can do has to acknowledge that some tasks currently performed by humans will be performed by AI instead, and that some roles that are heavily composed of those tasks will shrink or disappear.

But the relationship between technology and employment has always been more complicated than the "machines take jobs" story suggests. The industrial revolution eliminated certain kinds of work and created enormous numbers of new kinds of work. The computer revolution did the same thing. Spreadsheet software made many bookkeeping tasks automated and yet the number of people working in finance and accounting grew in the decades after spreadsheets arrived, because the new tools opened up new capabilities that created new demand.

Will AI follow the same pattern? Probably, at least in part. But there are also legitimate reasons to think this wave is different in some ways, because of that generality we talked about. Previous technologies eliminated specific, narrow tasks. This one is capable across a broad range of cognitive work. The adaptation will likely be faster and more disruptive than previous waves.

Here is what we do know. The research on which types of jobs are most exposed to AI disruption is actually fairly consistent across multiple studies. Jobs that involve highly routine cognitive tasks, processing standard documents, answering predictable questions, generating standardized content, face the most near-term displacement risk. Jobs that require deep interpersonal judgment, physical dexterity in unstructured environments, and genuine creative problem-solving in novel situations are less immediately threatened.

But the more interesting question, from a business strategy perspective, is not which jobs disappear. It is what happens to the jobs that remain. And the answer, based on what we are already seeing, is that they change substantially. The professionals who thrive in the AI era are not the ones who ignore these tools. They are the ones who figure out how to work with them, how to direct them, how to check their work, and how to focus their uniquely human capabilities on the parts of problems that AI still handles poorly.

This means that talent strategy in the AI era looks different from talent strategy before it. The skills that matter are changing. The way you train people needs to change. The way you evaluate and reward performance needs to evolve. We will get into specifics on all of this, but the fundamental shift in mindset is this: your job as a leader is no longer just to hire people with the right skills. It is to build a team that can learn, adapt, and combine human judgment with AI capability in ways your competitors cannot easily replicate.

The Data Foundation You Cannot Ignore

Here is something that does not get enough attention in the popular conversation about AI: the business that wins with AI is not necessarily the business that uses the most AI. It is the business that has the best data.

Underneath every AI application is data. The general-purpose models like GPT-4 or Claude were trained on enormous publicly available datasets. They are impressive. But what makes AI genuinely powerful for a specific business is when it can be applied to that business's proprietary data. Your customer interaction history. Your internal knowledge base. Your product usage data. Your operational metrics. Your sales call transcripts.

When you combine the reasoning capabilities of a sophisticated AI model with the specific knowledge embedded in your proprietary data, you get something that a competitor who is also using AI but without that data foundation cannot simply replicate. Your data is your moat.

This is not a new insight in business strategy. Data as a competitive asset has been discussed for years. But AI changes the economics of what you can do with data. A dataset that previously required an army of analysts to extract insights from can now be interrogated by a much smaller team using AI tools. Patterns that were invisible before become visible. Decisions that were made by gut feel can be made with evidence.

But you have to have the data in the first place. And for many organizations, this is where the real work is. Not in deploying AI tools, which has become relatively easy, but in making sure that the information generated by your business is being captured, organized, and made accessible in ways that AI can actually use.

We will come back to this in detail later in the book. For now, just hold onto this principle: before you ask "how do we use AI," ask "how good is our data?"

The Speed of It All

One more thing worth sitting with as we set the stage here. The pace of change in AI is unlike anything most industries have experienced before.

This is not hyperbole. The rate of capability improvement in AI systems has been extraordinary, and the competitive dynamics of the industry are driving continued rapid investment. The model that seemed impossibly capable in 2022 looks modest compared to what is available in 2024. The tools available to businesses today are meaningfully more powerful than the tools available eighteen months ago. And the trajectory suggests this pattern will continue for at least the near-to-medium term.

What this means practically is that the planning horizon for AI strategy is compressed. In most business strategy work, you might think about a three-to-five-year plan with reasonable confidence in the underlying assumptions. With AI, the environment changes fast enough that a plan made with today's capabilities might need significant revision when new capabilities emerge in twelve months.

This does not mean planning is useless. It means the nature of the plan needs to change. You need a clear direction and set of principles that can guide decisions even as specifics evolve. You need organizational capabilities, the ability to learn, experiment, and adapt, not just a fixed roadmap. And you need to be paying attention to what is happening in the field, not just executing a plan made six months ago.

The businesses that navigate this era well will be the ones that build learning into their operating model. Not learning as a platitude, but learning as a genuine capability. Testing things. Measuring results. Changing course when the evidence says to.

That might sound obvious. But it runs against the grain of how a lot of established businesses operate. Large organizations, by necessity, have built structures and processes designed to execute reliably and at scale. Those structures are good at many things. They are often bad at rapid learning and course correction.

The organizational challenge of the AI era is partly a technology challenge and partly a culture challenge. And culture, as any experienced leader knows, is much harder to change than technology.

What This Book Will Give You

By the time you finish these three chapters, you will have a clear picture of what is actually happening with AI in business, not the hype and not the dismissiveness, but the grounded reality. You will have seen, in detail, what companies that are getting this right are actually doing. You will understand the mistakes that have tripped up companies that tried and failed. And you will have a concrete framework for building your own AI strategy, one that is appropriate for your size, your industry, and your current capabilities.

This is not a book for people who want to become AI engineers. You do not need to understand how a transformer architecture works to make good strategic decisions about AI. What you need is a clear mental model of what these systems can and cannot do, a realistic picture of what execution looks like, and enough strategic framework to guide action in an environment that is genuinely uncertain.

That is what we are going to build together.

Let's start by looking at the companies that have already been through this. The ones that made bold moves and won. The ones that moved carefully and still found a way forward. And the ones that waited too long, or moved wrong, and paid a price for it.

Because the best teacher for what you should do is almost always what someone else has already done. The landscape ahead of us is new, but the human patterns in how organizations respond to change, those are old. And once you can see them clearly, you can make better choices.

That is the promise of this book. Not certainty. Nobody can promise you certainty right now. But clarity. And in a moment this confusing, clarity is worth a great deal.

Let's get into it.

The Mindset Shift That Makes Everything Else Possible

Before we move on, there is one final thing I want to address in this opening chapter, because without it the rest of the book will feel like a list of tactics without a foundation to stand on.

The businesses that are doing well with AI are not uniformly the ones with the biggest budgets or the most technical talent. Some of them are mid-sized companies that you have never heard of. Some of them operate in industries that do not seem glamorous or technologically sophisticated. What they share is a particular mindset, one that I have seen consistently when I look at the case studies and talk to the leaders who are navigating this well.

That mindset has three elements.

First, they treat AI as a strategic question, not a technology question. This means the conversation about AI happens in the leadership team and the boardroom, not just in the IT department. It means the decisions about where to use AI are driven by business priorities, not by what is technically possible. And it means the people leading these efforts understand the business deeply, not just the technology.

Second, they think about capability-building, not just tool deployment. There is a big difference between installing an AI tool and developing organizational capability with AI. Installing a tool is fast and relatively easy. Capability means your team knows how to use the tools well, how to evaluate the outputs critically, how to integrate AI into workflows in ways that actually change results. That takes investment and time.

Third, they are honest about uncertainty and build in ways to learn. They do not pretend to know exactly what the ROI on an AI investment will be before they make it. They start with a hypothesis, run an experiment, measure the result, and adjust. They have a tolerance for being wrong and a commitment to learning from it.

These three things, treating AI as a strategic question, building capability not just deploying tools, and learning as you go, turn out to be more important than any specific technology choice. They are the foundation on which everything else is built.

Now, with that foundation in place, let's look at the real world. Let's look at companies that have already been on this journey, made decisions, and lived with the consequences.

Because the best classroom for the AI era is the market itself.

Chapter 2: The Companies That Got It Right, and the Ones That Didn't

Stories are how we learn.

Not just as children. All of us, regardless of age or sophistication, learn better from narratives than from abstractions. When you see a principle in action, when you can picture the room where the decision was made, the team that argued about it, and the results that followed, it sticks in a way that bullet points and frameworks simply do not.

So this chapter is full of stories. Real companies, real decisions, real outcomes. Some of them you will know well. Others may be new to you. All of them have something important to teach about how AI adoption actually plays out in the world, as opposed to how it looks in a pitch deck.

I have organized these stories around four themes: the companies that moved fast and built real advantage, the companies that got tripped up despite good intentions, the companies that figured out a more measured approach and won anyway, and the companies that got so far behind they are still struggling to catch up.

Let's start with the movers.

Amazon: When the Machine Becomes the Business

You cannot talk about AI in business without talking about Amazon. Not because Amazon was the first to use AI, but because Amazon is perhaps the clearest example of a company that did not just use AI as a tool. It built AI into the fabric of what it is.

Start with the recommendation engine. When you go to Amazon and the site suggests "customers who bought this also bought that," you are looking at one of the most commercially valuable AI systems ever built. Amazon's recommendation engine drives an estimated 35 percent of its total revenue. Not 35 percent of some narrow category. Thirty-five percent of the entire revenue of one of the largest companies in the world.

But what is really interesting about Amazon's AI story is not any single application. It is the philosophy behind it. Jeff Bezos was explicit about this in Amazon's early shareholder letters: the company was building systems that would improve automatically as more people used them. The more customers bought things, the better the recommendations got. The better the recommendations got, the more customers bought things. This is what technologists call a flywheel, and AI made Amazon's flywheel spin faster than any competitor could match.

This did not happen overnight. Amazon was investing heavily in what would become AWS, its cloud computing business, through the mid-2000s. It was experimenting with machine learning for logistics optimization years before most retailers had even heard the term. When you order something from Amazon today and it arrives in two days or less, that is the result of AI-driven inventory positioning that ensures popular items are in warehouses close to the people most likely to buy them. Amazon moved inventory toward you before you even ordered it. That is a fundamentally different way of running a supply chain, and it is only possible with AI.

The lesson from Amazon is not "invest billions in AI." Most businesses reading this book cannot and should not try to replicate what Amazon did. The lesson is about integration. AI was not a project at Amazon. It was not something the IT team ran. It was woven into the core business model, the strategy, and the culture. When you are thinking about AI in your own organization, the question to ask is not just "where can AI help us?" but "where does AI change the fundamental logic of how we compete?"

Netflix: Predicting What You Want Before You Know You Want It

Netflix provides another sharp example, and a slightly different angle on the same theme.

Netflix's recommendation system is remarkably well-known, partly because the company has been public about it. They famously ran a competition in the late 2000s, offering a one-million-dollar prize to any team that could improve their recommendation algorithm by ten percent. The competition attracted teams from around the world and generated significant advances in the field.

But the most interesting AI story at Netflix is not the recommendation system. It is the content strategy.

When Netflix decided to move into original content production, it was making a bet that many in Hollywood thought was absurd. Netflix was a tech company. What did it know about making television? What it knew, it turned out, was its data. Netflix had years of watching behavior from millions of subscribers. It knew which shows people watched all the way through. It knew which actors generated the most engagement. It knew which genres were underserved relative to demand. It knew, at a granular level, what its audience actually wanted, as opposed to what the entertainment industry assumed they wanted.

House of Cards, Netflix's first major original production, was greenlit not on the basis of a pilot or the traditional TV development process. It was greenlit in large part because Netflix's data showed that the specific combination of elements in that show, the director, the lead actor, the political drama genre, had an unusually high probability of success with their subscriber base. They committed to two full seasons before a single episode had aired.

The show was a hit. And Netflix went on to become one of the dominant forces in global entertainment, not just distribution.

The lesson here is about using data to reduce uncertainty in decisions that are inherently risky. Content production is a notoriously uncertain business. Most shows fail. Most movies underperform. Netflix did not eliminate that uncertainty. It reduced it enough to make better bets than its competitors. That is a very important distinction. AI rarely gives you certainty. It shifts the odds.

Moderna: The Pandemic Proof of Concept

Not all the great AI stories come from tech companies. Moderna's story is one of the most dramatic examples of AI enabling something genuinely impossible by previous methods, and it happened in one of the most consequential contexts imaginable.

When COVID-19 emerged in early 2020, the world needed a vaccine faster than any vaccine had ever been developed before. The previous record for vaccine development was four years. The historical average was more than a decade.

Moderna produced a vaccine in eleven months.

Part of the reason was the emergency regulatory environment and the extraordinary resources mobilized during the pandemic. But a significant part of the reason was that Moderna had spent years building AI capabilities into its drug development process. The company had essentially become a technology company that happened to make drugs, and that orientation paid off spectacularly when speed mattered most.

Moderna's AI systems were able to rapidly analyze the genome sequence of the coronavirus, which was published by Chinese scientists in January 2020, and begin designing potential mRNA sequences. The company was able to design a candidate vaccine in just two days. Two days, for something that previously took years. Then the clinical trials, which still had to follow proper protocols, took most of the remaining time.

Moderna has continued to invest heavily in AI across its drug development pipeline. The potential applications are enormous: AI-assisted drug target identification, patient stratification for clinical trials, manufacturing optimization, and more. If AI can help cut the cost and timeline of drug development even by a fraction of what is theoretically possible, the impact on human health globally would be staggering.

The lesson from Moderna is about what happens when you invest in AI capability over time, not just for an immediate payoff, but as a platform. When the moment came that required extraordinary speed, Moderna was ready because the capability was already there. You cannot build that kind of capability in an emergency. You have to have been building it already.

JPMorgan Chase: Quiet Revolution in Finance

JPMorgan Chase does not get as much press as the tech companies when people talk about AI, but it deserves to. The bank has made some of the most aggressive and systematic AI investments of any non-technology company in the world, and the results are visible in their financial performance.

One of the most cited examples is COiN, which stands for Contract Intelligence. This is a machine learning system that JPMorgan built to review commercial loan agreements. Before COiN, this review process took an estimated 360,000 hours of lawyer time annually. It was slow, expensive, and prone to human error. COiN does the same work in seconds. And because the system learns from each document it reviews, it gets better over time.

But JPMorgan's AI story goes far beyond contract review. The bank has applied AI to fraud detection, with systems that analyze transaction patterns in real time and flag suspicious activity far more accurately than previous rule-based systems. It has applied AI to trading, with algorithms that can execute transactions and manage risk at speeds no human trader can match. It has applied AI to customer service, to compliance monitoring, to credit risk assessment.

The bank's former CEO Jamie Dimon has described AI as one of the most transformative technologies JPMorgan has encountered, comparing it to the impact of the internet and the printing press. That is strong language from someone who has been in finance for decades and seen a lot of technology come and go. He backs up the words with investment: JPMorgan reportedly spends over fifteen billion dollars annually on technology, with AI a growing share of that.

The lesson from JPMorgan is about scope. AI adoption does not have to be focused on one problem or one department. When you treat it as a cross-functional priority, when you have the data infrastructure and the organizational commitment to apply AI broadly, the cumulative impact can be transformational.

The Stumbles: Good Intentions Gone Wrong

Now let's talk about the companies that ran into trouble. Not because they were careless or naive, but because AI adoption is genuinely hard, and even smart organizations with good intentions can get it wrong.

IBM is one of the most instructive examples, because they had enormous advantages. They had technical talent, financial resources, decades of enterprise relationships, and a brand synonymous with business computing. And they had Watson, their AI platform, which they marketed aggressively as the future of business intelligence.

IBM's Watson story is a cautionary tale about the gap between marketing and reality in AI. In the early 2010s, Watson won Jeopardy!, which was a genuinely impressive technical achievement that generated enormous publicity. IBM then moved to commercialize Watson across multiple industries, with healthcare being the highest-profile effort.

The Watson for Oncology product was supposed to help doctors make better cancer treatment decisions. IBM partnered with some of the most prestigious cancer centers in the world. The marketing was extraordinary. The results were not.

Reports emerged from hospitals that had implemented Watson for Oncology that the system was sometimes recommending treatments that oncologists considered unsafe or incorrect. Internally, IBM staff who had worked on the project described the training data as having significant problems. The system had been trained largely on hypothetical cases created by a small number of doctors at a single institution, rather than on the vast real-world patient data that would have been necessary for reliable performance.

IBM wound down most of its Watson Health operations in 2021, selling off the business unit.

What went wrong? Several things, but the most fundamental was a mismatch between what the technology could actually do and what the marketing claimed it could do. This created a dynamic where customers had unrealistic expectations, which led to frustration when the technology performed at its actual level rather than its marketed level. There were also serious data problems, and healthcare turns out to be an extraordinarily difficult domain for AI for reasons that go beyond just data quality: the stakes are life and death, the regulatory environment is complex, and medical decision-making involves nuances that were harder to capture than IBM's teams initially understood.

The lesson is not that IBM failed because they were foolish. The lesson is that AI has limits, and overselling what AI can do, whether to customers, to investors, or to your own organization, creates problems that technical capability alone cannot solve. Honest assessment of what your AI can and cannot do is not just ethically important. It is strategically necessary.

The Amazon Hiring Algorithm: A Cautionary Tale About Bias

While we are on cautionary tales, it would be incomplete not to address one of the most significant failure modes in AI adoption: bias.

In 2018, Reuters reported that Amazon had developed and then quietly scrapped an AI recruiting tool that it had been working on since 2014. The goal had been to create a system that could automatically evaluate resumes and identify the best candidates for technical positions. Given the volume of applications Amazon receives, an automated screening tool would have been enormously valuable.

The problem was that the system was systematically biased against women.

The reason was structural, and in retrospect, predictable. The system had been trained on resumes submitted to Amazon over a ten-year period. Amazon's technical workforce during that period was predominantly male, which is typical of the tech industry. The AI learned to favor patterns in the resumes of people who had been hired, which meant it learned to favor patterns associated with male candidates. It penalized resumes that included words like "women's" (as in "women's chess club") and downgraded graduates of all-women's colleges.

Amazon's engineers tried to correct for this but concluded the problem could not be reliably fixed. The project was abandoned.

This story matters for every organization thinking about AI, because bias is not a niche problem. It is a fundamental challenge in AI systems, and it often emerges in ways that are not obvious until you look carefully. AI systems learn from historical data. Historical data reflects historical human decisions. Historical human decisions, in many organizations and industries, embodied systematic biases. Unless you actively work to identify and correct for this, your AI systems will perpetuate and potentially amplify those biases.

This is both an ethical problem and a legal and reputational one. The organizations that are handling this well are not just checking a compliance box. They are investing in what is called responsible AI: systematic evaluation of AI outputs for bias and other forms of harm, diverse teams that are likely to notice problems that homogeneous teams miss, and governance structures that ensure these questions are asked before deployment, not after.

Sears: A Warning from the Past

Sometimes the best way to understand a risk is to look at a company that faced a version of the same challenge in an earlier era and did not survive it.

Sears is often cited as a cautionary tale about digital disruption, but it is actually a deeper story about the failure to adapt to fundamental shifts in how business works. At its peak, Sears was the Amazon of its day. It had a vast network of stores, a legendary catalog business that brought retail directly to consumers across America, and a brand that was trusted by tens of millions of families. It was the largest retailer in the United States for much of the twentieth century.

What happened? The short answer is that Sears had every advantage and failed to use it. When the internet emerged, Sears actually had an early e-commerce presence. It had customer data that most online retailers would have killed for. It had supply chain infrastructure and brand recognition that new entrants lacked. It had all the ingredients for a successful digital transformation.

But the organization could not make the shift. The internal culture was shaped by decades of operating a physical retail model. The incentive structures rewarded the existing business. Leaders who raised concerns about the competitive threat were often sidelined. The investments in digital were too small and too late, constrained by the desire not to cannibalize the existing business.

By the time Sears recognized the full magnitude of the threat from Amazon and others, it no longer had the financial resources or the organizational energy to mount an effective response. The company that had once been a retail giant filed for bankruptcy in 2018.

The parallel to AI is not perfect. Every analogy has limits. But the structural pattern is strikingly similar to what some large organizations are doing with AI today: acknowledging that it is important, making modest investments to demonstrate engagement, but failing to make the kind of deep organizational commitment that would actually change the trajectory of the business. Protecting the current model at the expense of building the next one.

The question for leaders is blunt: are you investing in AI at a scale that matches the magnitude of the shift, or are you doing just enough to be able to say you are doing something?

The Mid-Market Success Stories

Not every great AI story belongs to a giant corporation. Some of the most instructive examples come from mid-sized businesses that have found focused, practical ways to deploy AI and generated real competitive advantage from it.

Consider Stitch Fix, the online personal styling service. Stitch Fix is not a household name in the way that Amazon or Netflix is, but it has built a genuinely differentiated business using AI in a clever way.

The challenge in personal styling is that human judgment is irreplaceable for truly personalized service, but human stylists alone cannot scale. Stitch Fix built a system where AI and human stylists work in genuine collaboration. The AI processes each client's data: their style profile, their past orders, what they kept and what they returned, their feedback, their social media where permitted. It narrows an inventory of millions of items down to a manageable set of candidates for each client. Then a human stylist reviews those candidates, applies their own judgment, and makes the final selection.

Neither the AI nor the human could do this job as well alone. The AI cannot replace the creative human judgment that makes a truly great styling recommendation. The human stylist could not efficiently process the volume of data or the breadth of inventory without AI assistance. Together, they create a service that scales without losing the personal touch that justifies the premium.

This human-AI collaboration model is one of the most important templates for businesses in the near term. Not replacement, but augmentation. Using AI to handle the parts of a task that are amenable to data processing and pattern recognition, while keeping humans focused on the parts that genuinely require human judgment, creativity, and connection.

Another mid-market example worth noting is Duolingo. The language-learning app has used AI not just to improve the learning experience but to dramatically lower the cost of producing content. Creating language learning exercises is labor-intensive: you need linguists, educators, and content creators for every language pair you want to support. Duolingo has used AI to automate significant portions of this content generation process, allowing it to offer more languages and more content per language at a fraction of the previous cost.

This has a direct impact on their competitive position. A competitor trying to match Duolingo's content depth without AI-assisted production would face a cost disadvantage that compounds over time.

Patterns from the Field

Looking across all of these stories, a few patterns emerge that are worth naming explicitly.

The companies that succeed tend to start with a clear business problem rather than a technology solution. They do not ask "how do we use AI?" They ask "what is the most important problem in our business that better information or greater automation could help solve?" Then they figure out whether AI is the right tool for that problem. This sounds obvious but it is not how most organizations approach it. Most organizations hear about an AI tool that seems impressive and then ask where they can apply it. That is backwards.

The companies that succeed also tend to invest in the full system, not just the AI model itself. The AI model is one piece. You also need clean, accessible data. You need workflows that are redesigned around what AI can do. You need humans who are trained to work with AI effectively. You need governance structures to catch mistakes and correct bias. Investing in the model while neglecting the surrounding system is one of the most common failure modes.

The companies that stumble often do so because of expectation mismanagement. Either they oversell the capability to stakeholders and face backlash when reality does not match the promise, or they are so cautious about setting expectations that they fail to generate the organizational energy needed to make real change. Calibrated honesty, realistic about both the potential and the limitations, turns out to be a genuine competitive advantage.

And the companies that fall behind tend to share one trait: they treat AI as someone else's job. The IT department's job. The data science team's job. When AI is siloed in a technical function and disconnected from business strategy, the investments rarely translate into meaningful outcomes. The organizations that win are the ones where business leaders are engaged, curious, and willing to put their own priorities on the table when figuring out where AI should be applied.

What You Take from These Stories

I want you to sit with these stories for a moment before we move into the practical framework. Because the goal of this chapter was not just to give you information. It was to give you feel.

When you have seen Amazon's flywheel, you have a different kind of understanding of what it means to embed AI in your business model. When you have seen IBM's Watson stumble, you are more likely to be honest about your own AI's limitations. When you have seen what happened to Sears, you feel the urgency of acting seriously rather than performatively.

Business strategy is ultimately about making decisions under uncertainty with incomplete information. The case studies in this chapter do not eliminate that uncertainty. Nothing can. But they give you patterns to recognize, mistakes to avoid, and examples to aspire to.

The next chapter is where we get specific. Where we take all of this and turn it into something you can actually use.

Because the end goal of this book is not that you come away impressed with what other companies have done. It is that you come away with a clear picture of what you are going to do.

Let's build your playbook.

Chapter 3: Your Playbook for the AI Age

Here is where we get practical.

The first two chapters were about understanding the landscape: what AI actually is and why it matters, and what the experience of real companies teaches us about what works and what does not. This chapter is about you. Your organization. Your decisions. Your next steps.

I want to be upfront about something. There is no single AI strategy that is right for every business. A fifty-person marketing agency and a ten-thousand-person manufacturing company face genuinely different challenges, have genuinely different data assets, and need genuinely different approaches. What I can give you is a framework that is robust enough to be useful across those differences, combined with enough specificity that it does not feel like empty advice.

The framework has five parts. Think of them not as sequential steps but as interconnected dimensions of a coherent strategy. You will likely need to work on all five simultaneously, though you may emphasize different ones depending on where you are starting from.

Those five dimensions are: Diagnosis, Data, Deployment, Development, and Direction. Let's go through each one.

Dimension One: Diagnosis

Before you can make good decisions about AI, you need an honest picture of where you are. That sounds obvious, but in my experience, most organizations skip this step or do it too quickly. They hear that AI is important, they pick a use case that sounds good, they run a pilot, and they hope for the best. Without a clear diagnosis, you are essentially navigating without a map.

A good AI diagnosis answers four questions.

First: where are the highest-leverage problems in your business? This is a strategic question, not a technology question. Think about where cost, quality, speed, or customer experience most constrains your performance. Where do your best people spend time on work that feels routine? Where does the business lose money or leave value on the table because of slow decisions, inconsistent execution, or information that is not being used? These are your candidates for AI intervention.

Second: what are your actual AI capabilities today? This means taking stock honestly. Do you have people who understand AI, not just in theory but in practice? Do you have anyone who has actually built and deployed an AI application? What tools are your teams already using? You may be surprised to find that AI is already present in your organization, because many software products now include AI features that teams are quietly using. Understanding your baseline prevents you from reinventing wheels and helps you identify the gaps that actually need to be filled.

Third: what does your data situation look like? We talked about this in the first chapter, but now we are going to be more specific. For each of the high-leverage problems you identified, ask: what data exists that is relevant to this problem? Is it clean and accessible, or is it scattered across systems and difficult to use? How much of it is there? Is it recent? Is there more being generated continuously, or is it a static dataset? The answers to these questions determine which problems are actually viable candidates for AI solutions right now versus which ones require investment in data infrastructure first.

Fourth: what is the competitive landscape? This is not about panic. It is about calibration. Understanding what your competitors are doing with AI, or not doing, helps you prioritize. If your closest competitor has already deployed AI in a way that gives them a cost or customer experience advantage, that is a different situation from being in an industry where nobody has moved yet. In the first case, you need to move fast in that area. In the second case, you have more latitude to be deliberate.

Doing this diagnosis well typically requires getting the right people in the room. You need business leaders who understand the strategy and the problems. You need people who understand data and technology. And you need frontline workers who understand how the work actually gets done, because that ground-level knowledge is where the most important insights about process inefficiencies and improvement opportunities tend to live.

One practical tool for this is what some consultants call an AI opportunity matrix. On one axis you put the potential business impact of solving a given problem with AI. On the other axis you put the feasibility, meaning how good is the data, how clear is the use case, how technically tractable is the problem. Problems that are high impact and high feasibility are your priority targets. Problems that are high impact but low feasibility are your medium-term investments: you need to build toward them. Problems that are low impact regardless of feasibility are things you can deprioritize.

The diagnosis phase typically takes a few weeks of focused effort. Do not rush it. The quality of your diagnosis will shape the quality of every subsequent decision. A good diagnosis does not just tell you where to start. It tells you why, and that "why" is what keeps your team motivated and aligned when the inevitable difficulties arise.

Dimension Two: Data

We already established that data is your moat. Now let's talk about what to actually do with that insight.

The first thing to understand is that data quality matters far more than data quantity. This runs against the popular intuition that AI needs massive amounts of data to work. For the large foundation models, that is true. But when you are applying AI to a specific business problem using your proprietary data, having a smaller amount of high-quality, well-structured data is almost always better than having a massive amount of messy, inconsistent data.

High-quality data has three properties. It is accurate: the information reflects reality rather than errors, and outdated or incorrect entries have been cleaned up. It is consistent: the same information is represented the same way across different records and systems. And it is accessible: the data can be retrieved and used efficiently, rather than being locked in legacy systems or requiring manual extraction.

Getting your data to this standard is not glamorous work. It is the kind of work that business leaders often want to skip because it feels like infrastructure rather than strategy. That is a mistake. Organizations that invest in data quality tend to find that it pays dividends far beyond AI. Better data leads to better reporting, better decision-making, and better customer experiences across the board.

For most organizations, there are three categories of data to think about strategically.

Customer data is typically the most valuable for commercial applications. This includes transaction history, service interactions, feedback, behavioral data from digital channels, and any other information about how your customers use your products or services. The business question this data can help answer: how do we better understand what different customer segments need, and how do we serve them more effectively?

Operational data is the information generated by your internal processes. This might include manufacturing quality metrics, supply chain logistics data, employee productivity measures, financial data at a granular level, or anything else that captures how your operations actually function. The business question this data can help answer: where are the inefficiencies, risks, and quality issues in our operations, and how do we address them systematically?

Market and competitive data comes from outside your organization: industry trends, competitor information, economic indicators, news and social media. This is less proprietary than the other categories, since your competitors can access much of the same data. But how you analyze and act on it can still be a source of advantage, especially when combined with your proprietary data.

For each category, the practical steps are: audit what you have, identify the gaps, prioritize the most important gaps based on the problems you identified in the diagnosis phase, and make a plan to close them. This might mean changing how customer data is captured and organized. It might mean implementing new systems to track operational metrics. It might mean integrating currently siloed datasets.

One thing I want to flag specifically: data governance. As AI makes data more valuable, questions about how data is collected, stored, used, and protected become more important. This is both a regulatory question, because privacy laws are evolving and becoming more demanding in many jurisdictions, and an ethical one. Customers are increasingly aware of and concerned about how their data is being used. Organizations that handle customer data responsibly, with genuine transparency and respect for privacy, are building trust assets. Organizations that cut corners are building liabilities.

Data governance does not have to be bureaucratic or slow. The best approach is to build clear principles, assign clear ownership, and make the right behaviors easy and the wrong behaviors hard through process design. But it has to be intentional. "We will deal with that when it becomes a problem" is not a governance strategy.

Dimension Three: Deployment

This is the dimension most people jump to first, which is part of why so many AI initiatives underperform. But with a good diagnosis and a data foundation in place, this is where you start to see real results.

Deployment is about getting AI actually working in your operations in ways that change outcomes. Let me give you a framework for thinking about this, and then we will get specific.

Start small, learn fast, and scale what works. This is not new strategic advice. It is basically the lean startup methodology applied to AI. But it is worth stating explicitly because there is a countervailing pressure in AI deployment, which is the temptation to go big. To announce a major AI transformation initiative with a large budget and a tight deadline, driven by competitive pressure or board excitement or both.

Large, high-pressure transformation programs often fail. They fail because the expectations are too high, the time is too short, and the organizational learning curve is underestimated. The companies that have done this well tend to start with one or two focused use cases, run them as genuine experiments with clear metrics, learn from what happens, and then apply those learnings to the next use case. Over time, this compounding of small wins builds both the capability and the confidence to tackle larger and more complex applications.

What makes a good first use case for AI deployment? Several criteria.

It should have a clear, measurable outcome. You need to be able to tell whether it worked. "We improved customer experience" is not measurable enough. "We reduced average customer service resolution time by thirty percent" is. This is both a practical necessity and a political one: if you want to build support for further AI investment, you need to be able to show concrete results.

It should be meaningful but not mission-critical. Meaningful means it matters enough to the business that success will be noticed and valued. Not mission-critical means that if something goes wrong, it will not cause a catastrophic outcome. This gives you room to learn without taking on excessive risk.

It should have a willing and capable team. Even the best AI application will underperform if the team using it is resistant or unprepared. Look for a team that is curious about AI, has relevant domain knowledge, and has the capacity to invest time in learning and adaptation.

Some of the use cases that tend to work well as starting points: customer service automation, using AI to handle routine inquiries and free up human agents for complex cases. Content generation, using AI to produce first drafts of marketing copy, reports, or internal communications. Document review and analysis, using AI to extract information from contracts, invoices, or other structured documents. Data analysis, using AI to surface insights from datasets that are too large to analyze manually.

For each of these, the deployment process has roughly the same shape. You define the use case and the success metrics. You assemble the data needed to run the application. You select the AI tool or build the system. You run a pilot with a small group, measure the results, identify the problems, and iterate. Then you scale.

The scaling step is where many organizations lose momentum. Pilots succeed and then nothing happens. The reasons vary: organizational inertia, competing priorities, failure to allocate the resources needed to go from pilot to production. This is worth anticipating. Before you launch a pilot, have a clear agreement about what happens if it succeeds. Who makes the scaling decision? What resources will be allocated? What is the timeline? Without this, success in a pilot can paradoxically lead to stagnation.

Let me give you a concrete example of what good deployment looks like in practice.

A regional insurance company, mid-sized, with a claims processing department that was constantly backlogged. Claims adjusters were spending a significant portion of their time on straightforward claims that did not require much judgment: fender benders, minor home damage, claims with clear liability. But every claim, simple or complex, had to go through the same queue.

The company piloted an AI system that could automatically review incoming claims, assess the documentation, categorize the claim by complexity, and process the straightforward ones automatically with minimal human involvement. The system needed to be trained on historical claims data to learn what a straightforward claim looked like versus one that needed adjuster judgment.

The pilot ran for three months with a subset of incoming claims. Results: the AI handled about forty percent of incoming claims fully automatically with an error rate lower than the human baseline for those same claim types. Adjusters were freed up to focus on complex claims, where their judgment genuinely added value, and their job satisfaction actually improved. Customers with simple claims received faster resolution.

The company scaled the system company-wide within six months of the pilot conclusion. The efficiency gains were substantial. But the more important outcome was what the freed-up adjuster capacity was redirected toward: more proactive customer communication, more thorough review of complex claims, and a new service where adjusters proactively checked in with customers after major claims. Customer satisfaction scores went up.

This is what good deployment looks like. Not just cost savings, though those were real. But a fundamental shift in how work is organized, what humans focus on, and what the customer experiences.

Dimension Four: Development

This is the dimension most often neglected, and it is probably the one that determines long-term success more than any other.

Development is about building your people's ability to work with AI effectively. Not just knowing which buttons to press, but understanding deeply enough what these tools can do, what they cannot do, and where their outputs need to be critically examined, to use them well.

Here is a useful way to think about the AI skill gap. There are three levels of AI capability that matter in an organization.

The first level is AI literacy. This is the baseline level that every employee in your organization eventually needs. It means understanding what AI is and is not, being able to use common AI tools for basic tasks, and knowing when AI output should be trusted and when it should be scrutinized. This is the level where you need to invest in broad training, not deep technical training, but enough exposure that people are not intimidated by these tools and are not naively over-trusting of them either.

The second level is AI fluency. This is a deeper competency needed by people who work with AI tools regularly as part of their core job. It means being able to prompt AI systems effectively, evaluate their outputs with real expertise, integrate them into workflows productively, and troubleshoot when things go wrong. This is the level where your marketing team using AI for content generation needs to be. Or your sales team using AI for customer research. Or your finance team using AI for analysis. Training at this level is more intensive and more specific to the particular tools and use cases relevant to each function.

The third level is AI expertise. This is a small number of people who understand AI at a technical level and can build, customize, and maintain AI systems for your organization. Most companies do not need large numbers of people at this level, but you need some. And if you do not have them internally, you need a clear strategy for accessing this expertise: through hiring, through partnerships with technology vendors, or through relationships with specialized consulting firms.

Building these capabilities takes time and sustained investment. The organizations that do this well treat AI skill development as an ongoing priority, not a one-time training exercise. They create communities of practice where employees can share what they are learning. They recognize and reward people who develop strong AI capabilities and share them with colleagues. They integrate AI into the way work is evaluated and performed, not as an optional extra but as a core expectation.

There is also a leadership dimension to development that deserves its own attention. Leaders who do not understand AI, even at the literacy level, will struggle to make good strategic decisions about it. They will be overly dependent on technical specialists who may not have the business context to give them good advice. They will ask the wrong questions and miss the important ones. They will be unable to hold their teams accountable for outcomes because they cannot evaluate what good looks like.

Leadership AI literacy does not mean coding. It means having a clear enough mental model of how AI works, what it can and cannot do, and what meaningful adoption looks like, to be able to exercise genuine judgment about strategic decisions. This is increasingly a core leadership competency, no different from financial literacy or strategic planning ability.

If you are a leader reading this book, I want to be direct: you need to develop this competency yourself, not just invest in it for your team. The best way to do this is not to read articles about AI. It is to use AI tools yourself, regularly, for real work. Ask ChatGPT to help you think through a problem. Use an AI writing assistant to draft a document. Explore how AI might help with something you find genuinely difficult. Get your hands on the tools.

The understanding you develop from direct experience with these tools is qualitatively different from anything you can get from a briefing or a report. You will develop intuition about what the tools do well and where they fall short. You will start to see possibilities you would not have seen otherwise. And you will be able to have much better conversations with the technical people in your organization.

Dimension Five: Direction

The final dimension is about governance and culture: how you set the direction for AI in your organization, how you make decisions, how you manage the risks, and how you build a culture that can actually navigate this era.

Let's start with governance. As AI becomes more embedded in how your organization operates, you need clear structures for making decisions about it. Who can authorize an AI deployment? What review process does a new AI application go through before it is deployed at scale? Who is responsible for monitoring AI systems once they are running? Who handles it when an AI system produces a wrong or harmful output?

These are not academic questions. Companies that have stumbled badly with AI often did so because they lacked clear governance. A team deployed an AI system without adequate review. Nobody was monitoring it closely enough to catch a systematic error. When something went wrong, it was not clear who was responsible for fixing it or how.

Good AI governance does not have to be bureaucratic. The goal is not to slow everything down with committees and approvals. The goal is to ensure that the right questions are asked before deployment, that someone is paying attention once a system is running, and that there is a clear path for resolving problems when they arise. A simple governance framework that gets used is far better than a comprehensive one that sits in a document somewhere.

Some key elements of a functional AI governance structure: a clear policy on which AI applications require executive review versus which can be deployed by individual teams. A checklist of questions that any AI application should be able to answer before going live (what data is it using, what are the possible failure modes, how will outputs be monitored, what is the process for human oversight of consequential decisions). A designated person or small team responsible for tracking AI developments across the organization and flagging risks.

Now let's talk about culture, which is ultimately the most important dimension of direction.

The organizations that navigate the AI era successfully are not just the ones that invest in AI tools and training. They are the ones that cultivate a specific kind of culture. A culture that is curious rather than fearful about new capabilities. A culture that is honest about what is working and what is not, rather than performing confidence. A culture that values learning and adaptation over the mere execution of plans made in advance. And a culture that treats the human element of work with genuine respect, not as an obstacle to automation but as the irreplaceable core of what the organization actually is.

On that last point, let me be specific. One of the biggest cultural risks in the AI era is what might be called dehumanization by efficiency. Organizations that become so focused on what AI can automate that they lose sight of what makes work meaningful, what makes customer relationships valuable, and what makes their organization a place where talented people want to be.

Efficiency is genuinely important. Waste is genuinely bad. But there is a version of AI-driven efficiency that cuts costs in the short term and destroys the culture and customer relationships that were actually the source of the organization's competitive advantage. This is worth being very deliberate about.

The question to ask with every AI deployment is not just "does this make us more efficient?" but "does this make us better?" Better at serving customers. Better at creating value. Better as a place to work. Sometimes the answer to those questions is the same: an AI that handles routine customer inquiries frees up human agents to have more meaningful conversations, which is better for customers and more satisfying for employees. But not always. Sometimes the efficiency gain is real and the human cost is real too. Being honest about that trade-off, and making conscious choices about it, is what distinguishes thoughtful leadership from mere optimization.

Putting It All Together: Your First 90 Days

I want to close this chapter, and this book, by giving you something concrete. Not a five-year plan, because five-year AI plans require a stability in the technology landscape that does not currently exist. But a ninety-day starting point. Ninety days of focused work that will meaningfully advance your organization's AI readiness regardless of where you are starting from.

The first thirty days are about diagnosis. Organize a small cross-functional team including business leaders, technical people, and frontline staff. Work through the four diagnosis questions I outlined earlier: where are the high-leverage problems, what are your current AI capabilities, how is your data, and what is the competitive landscape. By the end of thirty days, you should have a clear priority list of two or three problems that are the best candidates for AI intervention, and an honest assessment of what stands between you and solving them.

Do not try to do this alone or only at the senior level. The insights about where AI could make a real difference often come from the people closest to the work. A customer service rep who has been answering the same fifty questions for three years knows exactly which queries are routine and could be handled by AI. A financial analyst who spends half her time manually compiling data from different systems knows exactly where better data integration would save enormous effort. Go talk to those people. Listen carefully.

The second thirty days are about quick wins. Take the top priority problem from your diagnosis and launch a rapid experiment. The goal is not to build a perfect system. The goal is to learn. Use an existing AI tool where possible, the major AI platforms now have capabilities that would have required significant custom development just two years ago. Set clear metrics. Run the experiment for a defined period. At the end of thirty days, you should have real data about whether this approach is working and what needs to change.

Simultaneously, begin the first phase of development. Host a session for your leadership team on AI literacy: what it is, what the case studies show, and what your own organization's AI strategy will look like. Make sure every leader leaves with a hands-on introduction to at least one major AI tool. The goal is not to make them experts. It is to make the conversation real.

The third thirty days are about scaling and planning. Take the learnings from your quick win experiment and make a decision: scale it, modify it, or drop it and move to the next priority. Whichever direction you go, the decision should be based on evidence. Then use what you learned from the experiment to inform a broader plan. What resources do you need to scale AI adoption more broadly? What data investments are needed? What training? What governance structures?

By day ninety, you will not have transformed your organization. Anyone who promises that kind of result in ninety days is selling something. But you will have done three things that matter. You will have developed real understanding of your AI landscape, not based on theory but on direct experience. You will have created organizational momentum around AI, with a team that has seen what works and what does not. And you will have a credible plan for the next phase, grounded in evidence rather than aspiration.

That is the foundation everything else is built on.

A Final Thought on What This Era Asks of Us

I want to close with something that is a bit less practical and a bit more personal.

We are living through a moment that genuinely changes what it means to do knowledge work. The tools available to us now can extend human capability in extraordinary ways. They can make individual people more productive. They can give small teams the leverage of much larger ones. They can make sophisticated analysis accessible to people who previously could not afford the expertise. In the right hands, used thoughtfully, AI is genuinely democratizing, making certain kinds of capability available to a much broader range of people and organizations than before.

But tools do not have values. They amplify whatever human intentions they are pointed at. AI used thoughtfully by leaders with good values will create value for employees, customers, and communities. AI used carelessly or with purely extractive intent will create harm and resentment and, ultimately, failure.

The organizations that will thrive in this era are not necessarily the ones with the most AI or the biggest budgets. They are the ones led by people who understand that technology is always in service of something. The question is what.

What are you in service of? What is your organization for? What would success look like not just next quarter but five years from now, when the technology landscape has shifted again and what matters is the reputation you built, the talent you developed, and the customer relationships you earned?

Those questions are as old as business itself. AI does not change them. It just makes the answers more important to have clearly in mind, because the leverage you now have, the ability to do more, faster, at greater scale, means that those values, if they are good, will do more good. And if they are not, they will do more harm.

This new era is genuinely exciting. Not in a breathless, hype-filled way, but in the way that genuinely significant moments in history are exciting: with an edge of uncertainty, a weight of responsibility, and an awareness that the choices made right now will matter.

You have been given an extraordinary set of tools. You have the context to understand what they can do and what they cannot. You have the case studies to learn from what others have done. And you have a framework, however rough and ready, to guide your own decisions.

The rest is up to you.

Go build something worth building.

This book was written for leaders, operators, entrepreneurs, and curious minds who believe that understanding leads to better decisions. The AI story is still being written. The most interesting chapters are ahead.

Book Cover