[Warning! Spoilers Alert!]

In it, Dr. Louise Banks, a professor of linguistics, tries to work out a concept with her daughter, largely influenced by Banks’s physicist ex-husband. Her daughter Hannah asks her busy mother one night what the term is for this definition she has in her head:

HANNAH: What’s the term for that thing, like a technical term, where we make like a deal, and we both get something out of it?

LOUISE: A compromise?

H: No.

L: You remember what it sounds like?

H: Like it’s a competition but both sides end up happy.

L: Like a win-win?

H: More science-y than that.

L: A non-zero-sum game!

I loved that scene. If you watch the movie you'll see this is nearing the pivotal point where we as the audience start feeling the pull of time and space unravel in this beautiful, fractal way. The whole film is about what happens when you can see all of time at once, and what you choose to do with that knowledge. And embedded inside that enormous question is this small, quiet one: are you playing a game where someone has to lose, or are you playing one where everyone can win?

I don’t think we learn about this concept enough. What are zero-sum games and what are non-zero-sum games?

Through the years I’ve found that this is an important mental model and heuristic to think about when investing one’s time, money, and energy. It applies to nearly every aspect of business strategy and product strategy. Knowing which game you’re playing is the act of becoming metacognitive, and I believe this is the highest form of accumulating knowledge and honing one’s craft.

As folks interested in both product and investing, being metacognitive is the highest form of these crafts.

This topic is extensive and there are many amazing articles I've read over the years about applying non-zero-sum mental models to support an investment thesis. But for today I'd like to highlight some areas that might tease out sum things for sum games. (Sorry, I had to.)

To Eat Or Not To Eat

Eating effectively is a zero-sum game.

Everything you put into your mouth is one less other thing you cannot put in. Though I know some of us, at times, put the gamut of things in.

But here’s what I think a lot of people miss about this one. It’s not enough to avoid the foods that are “bad” for us. You also have to replace them with foods that are “better” for us. Every meal has an opportunity cost. And it’s not just calories in, calories out, it’s micronutrients, energy levels, how you feel two hours later, whether you can focus after lunch or need a nap.

This is the purest form of zero-sum thinking and honestly it’s where I think most people would benefit from being more metacognitive about it. You have a fixed number of meals in a day. Maybe three, maybe four, maybe two if you’re intermittent fasting. Every single one of those is a slot. And what you put in that slot means something else didn’t go there.

Work It Out

Exercising or improving one’s fitness is a non-zero-sum game. Period.

When you exercise, the returns spill over into every other game you’re playing. Physical fitness improves mental clarity. It improves sleep quality. It improves emotional regulation and even decision-making under pressure.

You put in an hour at the gym and you get back better performance in your work, your relationships, your focus, your mood. The returns compound across domains. That’s the definition of a non-zero-sum game: the total value created exceeds the value invested, and the surplus shows up in places that isn’t always obvious at first.

Skin in the Game

The concept of skin in the game, which Nassim Taleb popularized but which has been practiced by great investors for decades, is fundamentally about aligning the game so that it becomes non-zero-sum between the manager and the shareholders.

Think about the Buffett partnerships. Buffett didn’t charge a management fee on assets. He charged a performance fee above a hurdle rate. That meant if the partnership didn’t do well, Buffett didn’t get paid.

Now compare that to the typical hedge fund or mutual fund structure where the manager collects a management fee regardless of performance. That’s a zero-sum game between the manager and the investors, because the manager’s compensation comes out of the investors’ capital whether or not the fund makes money. The manager’s incentive is to gather assets, not to generate returns. And once you identify that game correctly, you start making very different decisions about where to put your capital.

This most aptly applies to Founder-CEOs, executives of manager companies, and investment vehicles where the operators don’t seek rent on minority shareholders. When the company does well, you do well. When the company doesn’t, you feel it too. These are also likely CEOs who have a large part of their own personal net worth in the company as well as not taking a huge (or any) salary for their involvement in running the business.

Contrast that with companies where the CEO has minimal ownership and a massive salary, and you start to see why the capital allocation looks different. They’re playing a different game. Not necessarily a bad game. But a different one.

Reaching for the Stars

I haven’t come across SpaceX or Blue Origin explicitly talking about zero-sum and non-zero-sum framing, but maybe Musk and Bezos have somewhere that I’m not aware of. I’ve been thinking about this one a lot though.

There’s a lot of negative talk about how we’re burning up millions of pounds of fuel and dollars to build what is effectively space tourism or more modern ways of launching into orbit or going to the Moon or Mars. And on the surface, it looks zero-sum or even negative-sum. Resources in, not much back (yet).

But I think in the minds of Musk and Bezos, this is fundamentally non-zero-sum. And the historical evidence supports this.

Advances made during the first space race of the mid-20th century led to many cottage industries and pushed forward advances in countless areas of applied science and technology. Memory foam, scratch-resistant lenses, water purification systems, satellite communication, GPS. None of these were the point of going to space, but they were the spillovers. The non-zero-sum surplus that showed up in domains nobody was targeting.

This modern space race feels the same to me. Some advance in materials, process, or propulsion technology makes its way back to improve our terrestrial lives years or decades from now. The game looks zero-sum at the rocket fuel level (every pound burned is gone) but non-zero-sum at the technology spillover level. The metacognitive skill is identifying which layer of the game you’re looking at.

And beyond the physical, there is also the inspiration, hope, and creativity that come from seeing Artemis II fly around the Moon and back. This is also massively positive-sum for inspiring the next generation to, quite literally and metaphorically, reach for the stars.

One Battle After Another

While we don’t like to admit it, like any race of the space variety, war itself potentially pushes advancements in certain areas that we might otherwise have not explored.

Of course no truly rational person wants world conflict, of which we see too many as I publish this today. But many lessons learned prepping for or on the battlefield have accelerated the cycles of experimentation needed to understand how to correct, fix, and make better products and services.

The internet itself started as ARPANET1, a military project. Medicine (Penicillin) mass production2. Duct tape3, for that matter. The list of technologies that were accelerated by wartime necessity is long and uncomfortable.

While war itself is always negative-sum (too many negatives to count at times), some of the technology that comes out of it may very well be positive-sum over the medium and long term, should we choose to take those lessons learned wisely. This is one of the most uncomfortable examples of a nested game: the top-level game is negative-sum by any moral or humanitarian measure, but nested inside it are pockets of non-zero-sum outcomes in technology, medicine, and communications.

I don’t say this to glorify conflict. I say it because I think the metacognitive practice of identifying which game you’re looking at, at which layer, is the skill. And sometimes the layers contradict each other. Sometimes the game is negative-sum at one level and positive-sum at another, and holding both of those truths at the same time is part of the work.

If there’s one thing I want you to sit with from this piece, it’s the idea that most of us are playing games without knowing the game board, the rules or who are our “opponents”, to name a few. The metacognitive act of stepping back and asking “what kind of game is this?” starts the process of deepening what we are actually trying to do and how likely are we to achieve our stated goals. Here’s some questions to tease that out:

1. Are the operators and managers you’re invested in playing the same game as you?

2. When you look at something that appears zero-sum on the surface, have you checked whether there’s a non-zero-sum game nested inside it? Or vice versa?

3. Can you name the game you’re playing right now in your career, your portfolio, and/or your health?

Five decades.

And if they have their way, what I’m gathering is that Apple will continue to be even more relevant — entrenched even — in the next 50 (but certainly the next 5).

For those who’ve been subscribed for a while (and welcome to the new readers and subscribers), you know that here on Product & Investing Nerd, or what I lovingly call PIN, we focus almost 100% of our time on enterprise software products. Consumer products, be it software or hardware, aren’t really in my circle of competence. But I’ll make an exception today because I want to synthesize the takes of my three favourite analysts in the product and investing space on how Apple is poised to remain relevant and maybe even “win” the AI race. Plus a bit of my own on top.

Investing Take: Expensive, Moat-y Business

It's worth sitting with just how entrenched Apple is today as a baseline.

As of January 2026, there are around 2.5 billion Apple devices globally. I know some households (like mine) have multiple ones, but even accounting for that, it’s still a staggering chunk of the planet carrying Apple in their pockets.

And since the introduction of the iPhone in 2007, Apple has only deepened their moat year after year:

• 2016: Wearables (AirPods, Apple Watch)

• 2019: Bundled Services (TV+, Arcade, Fitness+, News)

• 2020: In-house Silicon (M-Series chips)

• 2024: Apple Intelligence (Siri reborn, on-device inference)

• 2024: New Form Factors (Vision Pro, and a foldable phone reportedly on the way)

That’s the kind of cadence Buffett was talking about. Widening the moat every year, not every decade.

Warren Buffett on Apple

Warren recently did an interview on CNBC where he said something I found interesting. Remember, Buffett bought a huge stake in Apple through Berkshire ($BRK) when it was trading around 15x PE. He held and even marginally added through the years, but famously trimmed between Q1-2024 and Q3-2025 around the $180s. Then he told Becky Quick something I think a lot of long-term Apple holders will relate to.

He said he “sold too soon.”

The current market would suggest he sold far too early. But whatever, you can only see these things when you connect the dots looking backwards, as Steve Jobs himself would say.

The point isn’t that Buffett made a mistake. The point is that even Buffett can misjudge just how entrenched Apple is with its customers and the degree to which it keeps widening that moat every quarter by creating products and services that people and businesses can’t live without.

Which brings us to why I continue to think Apple will do well in the age of AI.

Product Take: Sticky, Agentic Even

Of course, Apple is more than one product. But there's probably only one product that still meaningfully moves the needle: the iPhone.

Nate Jones of Nate’s Substack

I love ’s work. Him and I write in a similar space about how AI and products evolve and diffuse into our organizations and personal lives. Here's how Nate captures the current moment:

Everyone thinks Apple lost the AI race. Google shipped agentic features on Android months ago. OpenAI has 900 million weekly users. Anthropic’s agents are writing compilers. And Apple, the company with more than 1.5 billion iPhones in the pockets of the highest-spending consumers on Earth, is still shipping notification summaries that hallucinate news headlines.

Apple isn’t behind. Apple is playing a different game. And if the evidence is right, it’s the game that matters more.1

I think most AI pundits are missing this piece. While Anthropic and OpenAI work very hard to diffuse artificial intelligence into our lives, look at where they’ve been fishing.

In Q1-2026, Anthropic and OpenAI’s focus has overwhelmingly been on an enterprise strategy, rolling out MCPs, agents, and plugins aimed at various verticals. Look no further than a recent job posting from Anthropic as a prime example of what they’re looking for and what their unspoken strategy is.

Subscribe now

Synthesizing it all, it raids like the requirements document for what Workday and other HCMs/Human Capital Management companies does, pretty much. I suppose this is more dogfooding using their own compute, but there's no reason why this couldn't eventually be productised in whole or in part and sold.

The reality is that enterprises, as is our focus here on PIN, are ripe for this kind of AI diffusion largely because they’re closed-loop systems and mostly very fragmented. So there’s a lot of low-hanging fruit.

Even here, Apple could do a lot. But as has been the case in the past, Apple was never in the business of selling shovels in the computer race of the 80s, 90s, and early 2000s. They were selling the gold itself. And I think that continues now too. We’ll come back to their product a little later.

Ben Thompson of Stratechery

When the ChatGPT moment happened, it was the best thing since sliced bread. For a moment there, even the best tech pundits forgot about the economics of tech. Most people saw this incredible technology and wondered: where is Apple in all of this?

And for a while there, it seemed like even Google was losing that race. Google looked as though it was behind ChatGPT, and Microsoft made a retrospectively amazing bet in OpenAI.

Remember all the benchmarks? Remember DeepSeek? Those feel like distant memories now.

I think our world in accelerating into the Singularity that it’s hard to see the forest for the trees. I don’t blame you or us.

Our world is accelerating into the singularity so fast that it’s hard to see the forest for the trees. I don’t blame you or us. And in that era, roughly from 2021 to 2025, Apple felt like they were left in the dust, particularly when it came to Layer 3 of Jensen’s 5-layer cake: data centers.

Guess which Mag7 doesn’t below here on this graph?

In Ben Thompson’s latest piece, Apple’s 50 Years of Integration2, he details exactly that. While it seems Apple “missed” the AI moment, it’s starting to look like Apple didn’t miss it at all. As Nate Jones says above and as I’ve written about before, choosing which game you want to play is often more important than simply playing the cards you’ve been dealt in the past. Whether Apple was doing this on purpose or not, it seems like they might eventually “win” in the end.

Thompson puts it as directly as you can:

“Apple has survived 50 years by being the only company integrating hardware and software; if the company loses because of AI it will be because the point of integration changes.”

Subscribe now

Let me lay out what I think are the most important elements of Apple's moat right now:

1. The iPhone as gatekeeper. Apple’s moat is the complete synthesis of hardware and software. And coincidentally, with AI, the top couple of layers of Jensen’s cake all require a cohesive marriage between the two. You can’t deliver a great AI experience on a device if the chip, the OS, the memory architecture, and the models aren’t talking to each other natively. Google can ship agentic features on Android months before Apple, but Android runs on hundreds of different hardware configurations. Apple runs on one.

2. AI as a commodity (at least for consumers). There have been frontier AI labs betting the opposite, but for consumer use, AI is becoming at minimum interchangeable. I find myself using Perplexity for some specific domains (like when I want a good read on 10-Ks and 10-Qs), but I’ll go to Gemini for reliably making slick novel graphics (for this blog, for example). But for everyday questions, most of the time it’s the one with the most context on me that I spin up. That’s also why both Claude and Gemini now have pre-written prompts that help you “pull” memory and history from one chatbot to another, effectively migrating context from one closed garden to another. It’s a signal that general-use, perhaps more personal-use chatbots are already commoditised in most cases. When the model becomes commoditised, the point of differentiation moves to distribution and integration. And nobody integrates better than Apple.

   

3. Extension of our forms. While Apple has dabbled in the Vision Pro and whatever comes next, the phone is still a form factor that fits the way we see, touch, feel, hear, and interact with the world. Short of whatever Neuralink cooks up in the future, the surfaces through which we access our inference devices will continue to be something we can see and touch, readily and easily.

4. Winning on the Edge. And here’s the one I think is the most underappreciated. It’s because of the integration moat that Apple has the ability to not only win without needing to build out its own mega data centers, but to win on the edge. Edge computing on Apple devices is possible because of decades of investment in their own in-house M-series silicon, but also because as more powerful chips and more memory end up on device, be it iPhone or Mac, those devices become the private, fast, user-controlled gardens where inference happens.

The Destination >> Journey

I want to sit with that edge computing idea for a minute because I think it's the most under-appreciated piece of the whole story.

Read more

By response I mean, I knew from my early grounding in Human-Centred Design that if you just understand the market, understand the users, and uncover what they need and want, you could build the tool they needed. So I joined an incubator that followed that philosophy, bootstrapped a SaaS product from early founding customers, used those funds to build the first MVP and as they say the rest is history.

In both my schooling and in my own personal experience I’ve found this golden rule to be true: that the focus on what your users and customers need and want will get you very very very far in business.

Don’t just take it from me.

The People Who Took It Seriously

You see many folks who have truly lived this, not just lip service, and have taken this simple idea and taken it very very seriously to achieve unmatched success.

Jeff Bezos famously turned “customer obsession” into Amazon’s first and most important leadership principle. In his very first shareholder letter in 1997, the word “customer” appeared 25 times in roughly 1,500 words. His framing was simple: “The most important single thing is to focus obsessively on the customer. Our goal is to be earth’s most customer-centric company.” And he meant it. Everything from Prime to one-click ordering to the return policy was designed by working backwards from what the customer wanted, even when it hurt margins in the short term.

Mark Leonard and Constellation Software $CSU.to took this in a different direction but arrived at the same place. The whole CSU acquisition model is built on the idea that when you buy a vertical market software (VMS) company, you’re not just purchasing the code and the employees. You’re purchasing decades of “distilled operating experience” (ML’s words) in that vertical, the accumulated knowledge of what those customers actually need, the problems they’re facing, the workflows they’ve built their businesses around. That’s why CSU keeps these companies autonomous after acquiring them. The people closest to the customer are the ones who understand the customer. Don’t mess with that.

Alex Karp and Palantir learned this lesson on the literal battleground. Political affiliation aside, Palantir’s forward-deployed engineering (FDE) model, where you embed your best technical talent directly with the customer, exists because Karp figured out early that in order to get anything done you have to roll up your sleeves and sit with the people doing the work. You can’t understand their wants and fears from a conference room. You have to be in the field. The whole company is structured around this idea.

And then there’s Qasar Younis at Applied Intuition2, who Marc Andreessen has called the “AI CEO of our time” recently3. Applied Intuition builds simulation infrastructure for autonomous vehicles and Younis has built the company on a deceptively simple tenet: never disappoint the customer. In a B2B environment, your reputation for reliability is your most valuable asset. While he doesn’t believe in too much press for his company (I wonder where I’ve heard that before), he attributes the company’s success to massive customer obsession.

Four very different operators. Four very different industries. All independently arriving at the same conclusion: Customer-focused Obsession.

The Counterpoint

And then there are people who are famously the opposite. Like Steve Jobs.

Jobs said a lot of things, and some of them contradicted each other. He famously said at the 1997 WWDC that Apple ($AAPL) needed to “start with the customer experience and work backwards to the technology.” But he also said, and I think this is more representative of how he actually operated: “It’s really hard to design products by focus groups. A lot of times, people don’t know what they want until you show it to them.”

I like this analogy best. Say you’re driving in a city you’re not familiar with and your map apps aren’t working. You roll down your window and ask some locals for directions. Well, you have to know where you’re going before you can take those people’s direction and advice. That’s what Steve was getting at.

Matt Shlosberg, an 8-time co-founder and now COO of Deep Origin4, recently wrote a piece titled “Why Listening to Your Customers Is the Dumbest Strategy in Business”5. His argument is that too many founders are obsessed with what their customers asked for last week and they call it being customer-centric when really it’s being historically accurate. You’re optimizing for a world that no longer exists.

Both approaches can work. Both can fail. That’s the reality of business and markets. We never know until we do and then it might be too late. And if you don’t hear the signals from your customers, whether it’s the explicit signals they intend on sharing with you or by voting with their silent feet, you could still be wrong.

But here’s the thing that both sides have in common.

They both assume the customer is a person.

It’s pretty self explanatory.

Both Can Be True … and Still be Wrong

The reality is both strategies can work at the Product & Investing level. It all depends. It also can both fail even when used correctly and intensely. That’s the reality of business and the markets. We never know until we do and then it might be too late. And if you don’t hear the signals from your customers be it the explicit signals they intend on sharing with you or by voting with their silent feet, you could still be wrong. There are a zero guarantees, well made a limit of zero.

Best Before or Expiration Date?

I still don’t know what the real difference between best before and expiration date is. Does it depend on the product!? But that’s a topic for another day.

The question I really want to ask is this: how does customer obsession as a rule of thumb break down when agents show up on the scene?

What if your customers are no longer people but “artificial” consumers?

Naval Ravikant has been saying something that I think is directionally correct even if we're early in seeing it play out: that UI is pre-AI, that every interface built around menus and form fields was designed for a world before users could simply describe what they want. That instead of buying a CRM to manage customer relationships, you instruct an AI agent to manage customer relationships. Instead of buying project management software, you instruct an agent to manage your projects.

While you may not agree with it 100% and still be the default workflow of today, I would at least agree it is generally directionally correct.

This is the risk that I think is under-appreciated.

All good things must come to an end. And all good things then it needs revising.

We have some models in the past from which to rely on. In the first industrial revolutions, we had to learn how to ‘interface’ with our machines and learned hard lessons in World War II, notably in how to build better interfaces for cockpit designs6.

What happens when the users no longer need an explicit interface? I agree directionally that per-seat pricing for SaaS products is having a hard time growing and is compressing. There’s the opposing force that customers may want financial regularity over accordion pricing based on outcomes. But the reality beyond pricing is how do we build software anymore if it’s no longer for real, living, breathing users?

Perhaps we can unpack how Agents are changing eCommerce. I think it might help uncover a future model for how designing for Agents (as an intermediary) differs from design for a person.

The reason why I picked this is partly because it’s one of the most well defined areas of any workflow and as such in early 2026, it’s also the one where there are most companies building toward that future. Secondly, the objectives are typically clear. Purchasing decisions are usually active decisions out clear outcomes and is a binary decision making process. It helps compress and ideal for exploration (at least for now).

Customer obsession worked because it solved the problem perfectly for which I would call the: INFORMATION ASYMMETRY PROBLEM.

The closer you are to the customer, the better you understand their needs, the better product you build, the more value you capture. But this whole framework assumed the customer could articulate, explicitly or through their behaviour, what they need. Agents don’t have desires or fears. They have objective functions. The entire practice of “uncovering what they need and want” starts to break when the customer is optimizing programmatically.

And the psychology that made customer obsession so powerful, the loss aversion, the status quo bias, the emotional attachment to brands, the switching cost inertia, all the human irrationality that makes customer relationships sticky and durable; none of that applies to an agent comparing options via API. Agents are closer to rational optimizers. They don’t have loyalty in the human sense.

They don’t have fear. They have benchmarks.

Now, I want to be clear here because agents aren’t truly neutral either. An Amazon agent is still going to prefer Amazon products. A Perplexity shopping agent carries the incentives of whoever built it and whoever is paying for placement. So agents have biases too, they’re just different biases. They’re the biases of their creators, not their users. And that changes the game in ways none of us has completely thought through.

All good things must come to an end. And all good things at some point need revising.

Agentic Commerce: The Proving Ground

Perhaps we can look at how agents are already changing eCommerce. I think it might help uncover a future model for how designing for agents as an intermediary differs from designing for a person.

I picked eCommerce partly because it's one of the most well-defined areas of any workflow and as such in early 2026 it's also the area where the most companies are building toward that agentic future. Secondly, the objectives are typically clear. Purchasing decisions are usually active decisions with clear outcomes, a binary decision-making process that compresses well and is ideal for early agent adoption.

The model makers are moving in. Perplexity, Gemini, ChatGPT and others are building intent-based eCommerce products and services, much in the same way that Google Search monetized intent through ads. The difference is that an agent doesn't browse. It resolves. It doesn't need to be persuaded by a product page, it needs structured data it can evaluate.

The incumbents are building agents into their networks. Amazon has Rufus7 and Walmart has Sparky8. (Why do they all sound like pet names?) These are agents embedded directly into already-dominant eCommerce platforms, and they change the interaction model from a person scrolling through a catalogue to an agent filtering and recommending based on stated criteria.

Headless SaaS is emerging. Traditional software like Photoshop, Slack, and Jira is being rebuilt with agent-first APIs. No UI. Programmatic access. Essentially the same product with a different interface. This is what happens when the assumption that a human is at the interface gets stripped away.

New payment primitives are being built. Companies like Natural9 are building payment system protocols specifically for agents, because the existing payment infrastructure was designed for humans clicking “buy now” and entering credit card numbers. Agents need something different entirely.

Stripe’s Agent Commerce Protocol (ACP) is perhaps the most telling signal. When the company that processes a huge chunk of internet commerce builds dedicated infrastructure for agent-to-agent transactions, that tells you something about where the puck is heading.

SEO is becoming AEO. Search Engine Optimization, the practice of making your content discoverable by humans through search, is evolving into Agent Engine Optimization, the practice of making your products and services discoverable and evaluable by agents. The entire discipline of “how do we get found” is shifting from persuading a person to click to providing an agent with the structured information it needs to make a recommendation.

Perhaps the first and most prominent casualty of agentic-first commerce is the demise of the UX interface as we know it. It won’t be fully done away with as there will always be people who want to do “window shopping.” But on the backend, companies will need to build feeds that are agent-ready, low-latency, and readable by a whole host of agents from different platforms.

Survival of the Fastest: When Agents Eliminate Info Asymmetry

Customer obsession was the right strategy for a world of human customers. It may still be the right instinct, but the object of the obsession has to evolve.

The companies that figure out how to be “agent-obsessed”, understanding what agents need (relatively structured data, low latency, reliability, machine-readable everything) the way Bezos understood what humans need (convenience, speed, trust), those are the ones who’ll have the edge.

The muscle that the best operators built for understanding human customers, that adaptation that made them great, could become the vestigial (organ) if they don’t evolve it. The organ is still there but the organism around it is changing.

The golden rule still holds. But the definition of “customer” is what’s mutating.

1

Human Design for Tech Workers

I Built a SaaS Before I Knew About Human Design. Here's Why I Think It Actually Worked.

Before I ever heard the words “Wait to Respond” and before I knew anything about BG5 or Human Design, I built a bootstrapped SaaS business from zero. It became profitable. It worked beautifully…

Read more

a month ago · 3 likes · Eugene Ting

5

Matt Shlosberg

Why Listening to Your Customers Is the Dumbest Strategy in Business

I have met too many founders and executives, across companies of every size, who are obsessed with what their customers asked for last week. They call it being customer-centric. It’s not. It is being historically accurate. Archeologists love it, except you are often optimizing for the world that no longer exists…

Read more

16 days ago · 1 like · Matt Shlosberg

“What doesn’t kill you makes you stronger.”— Friedrich Nietzsche

“And if at first you don’t succeed, Dust yourself off and try again, You can dust it off and try again.”— Try Again by Aaliyah featuring Timbaland

I think I’ve always been fascinated by perpetual motion or perpetual machines. I think those of us who like efficiency or want to get/create/make something out of nothing (i.e. alchemy) is exhilarating.

The oldest Idea of the perpetual motion machine was invented by an Indian mathematician, Bhāskara II, in around 1150AD.

The wheel consisted of curved or tilted spokes partially filled with mercury. Once in motion, the mercury would flow from one side of the spoke to another, thus forcing the wheel to continue motion, in constant dynamic equilibrium.

We hear stories in the news like that -- in our modern day, examples include the “flywheel class”: success of the mega, unicorn-like companies, Uber , Lyft , Airbnb to name a few.

Read more

A few weeks ago I came across a write-up1 from a founder who did something I don’t see very often. He wanted to build vertical SaaS for the pest control industry. So he went and got a job as a pest control technician.

He’d been doing white-collar consulting work for companies selling into traditional industries and noticed something that a lot of SaaS founders eventually notice: companies have become less likely to offer their time for research calls and ride-alongs. They get too many requests. So he decided to skip the asking and just go do the groundwork to build a new SaaS!

Terry Clarke took a radical approach to understanding the industry by getting a real job as a pest control technician rather than just doing “ride-alongs” or research calls. So he up-levelled by doing the front-line things:

• Got licensed in record time

• Switched to sales

• Identified “Undercover Boss” gaps

The company he joined was a subsidiary of one of the biggest pest control groups in the US, doing billions in revenue through a nationwide portfolio of local brands. The kind of operation that from the outside looks like it should be running on clean, modern tooling.

Big surprise: it wasn’t.

Onboarding required registering for 10+ apps on the company phone. He used two of them. The rest just sat there, part of a stack that had accumulated over time the way barnacles accumulate on a ship’s hull. Nobody chose this architecture. It just happened, one integration at a time, one workaround on top of another workaround, until the whole thing was too tangled to pull apart. (The core software was Salesforce ).

I’ll talk more about the details but here’s the punchline: Clarke ended up abandoning the build out of the new SaaS even though he had decades of experience in the industry, knew the ins and outs and also hands-on experience. In other words both back of house and front of house — the entire vertical in his head — still wasn’t enough for him to build a new SaaS!

This is the thing about VMS that I think gets underestimated constantly and it’s the reason I keep coming back to it as we think about the potential and threats of AI.

Most people assume fundamentally misunderstand and think that software is sticky because the software is good. That somehow the product earns its place by being excellent and that’s why customers stay. And sure, sometimes that’s true. But most of the time in vertical markets the software is sticky because the cost of switching exceeds the cost of living with something mediocre.

Subscribe now

The pest control techs Clarke worked with didn’t have any particular love for Salesforce. They had workarounds for the monitoring systems. The quoting process nearly killed a $24k deal because it required multiple signatures and yet another account to be created by corporate. The sales training was a Zoominfo webinar. The fuel card took five weeks to arrive. None of this is a picture of great software or a well-oiled business process for that matter.

There’s a psychological dimension to this that I think matters. Humans are loss-averse by nature. Daniel Kahneman showed us that the pain of losing something is roughly twice as powerful as the pleasure of gaining something equivalent. This doesn’t just apply to individuals, it applies to organizations too.

The certain pain of switching (we know this will be disruptive, we know there will be a learning curve, we know some things will break) always feels bigger than the uncertain gain of better software (it might be faster, it might save us money, it might make the techs happier, but who knows for sure).

So the decision at the individual level is rational even if it’s suboptimal at the organizational level. The employee who says “I don’t want to rock the boat” isn’t being lazy. They’re being smart about their own position. The pain of a botched software transition lands on the people who pushed for it. The gain of a successful one gets absorbed by the organization.

This is why enterprise software sales cycles are so long and so painful. You’re not just selling a better product. You’re asking an entire organization to overcome its own loss aversion2, and will always be a harder sell.

Here’s what I found most telling about the founder’s story. He went in wanting to build vertical SaaS. He did the hardest version of customer research possible, actually doing the job. He saw every broken workflow, every clunky integration, every frustrated tech.

And his conclusion wasn’t “great, now I know exactly what to build.”

His conclusion was to acquire a small residential operator and build the tooling around it.

Subscribe now

Think about that for a second. The person who understood the problem better than almost any founder possibly could still decided that building from scratch and selling into these organizations wasn’t the play. He’d rather own the operation and build the tooling from the inside out.

That tells you something about how deep the switching cost goes. It’s not just about the software itself. It’s about the fact that even in a seemingly straightforward vertical like pest control, the operations are complex systems. There are regulatory requirements, fleet management, chemical handling, licensing, customer scheduling, route optimization, invoicing, and all of it is wired together through whatever software happened to be there when the company was growing. The system isn’t just the software. It’s the software plus the people plus the processes plus the workarounds plus the muscle memory that built up over years.

That’s why I’ve written about before, that it’s not AI itself that will be disruptive, but the higher-order effects coming out of AI. AI poses a real threat from those that in-housing and building their own custom software.

Even for verticals that look simple from the outside, the reality on the ground is that these are complex adaptive systems and the VMS that sits at the centre of them has become load-bearing in ways that aren’t obvious until you try to remove it.

The industry-insider went looking for an industry to disrupt with better software. What he found instead was an industry where the switching cost is the software, where the organism and its environment have grown around each other so completely that you can’t separate them without doing damage. The software isn’t a tool the company uses. It’s closer to an organ the company grew.

All companies continually ask this question even if it’s not always conscious: What is the bigger pest, the costs and the inertia required to rip it [software, a system] out or start new with the promise of increased returns?

-&

If you liked this article you might like this funny little story I put together about VMS:

2

Cerebrate by thinkabled

🧠 Why Losing $100 Hurts More Than Finding $100 Feels Good

Someone borrowed my favorite book last year. Never gave it back…

Read more

a month ago · 5 likes · Sabith Mohammed

The world worships efficiency. The best investors and operators build for pro-entropy.

While I typically write about the micro (verticals, industries, individual business strategy), I do also follow the macro and the major trends. I’ve read The Fourth Turning many moons ago but many other folks have come out in recent times to talk about their own prognostications of similar things, mainly how we’re shifting from one era to another.

Regardless of exactly how the macro will play out, I think there are a few things that are undeniable.

Perhaps most businesses and industries have been asleep at the wheel because our world is actually full of entropy and cheap energy and heavy industrialization has made us believe that all the just-in-time products and services we get (in most of the world) is a given, not a luxury.

The Altar of Efficiency

Here’s the assumption most people don’t question: efficiency is always good.

Decades of lean manufacturing, just-in-time logistics and optimization culture have drilled this into every MBA, every earnings call, every operations review. And it’s true, up to a point. But there’s a trade-off that gets buried in the optimization math.

Every unit of efficiency you gain reduces the slack in the system. And slack is what absorbs shocks.

Think of it like this: efficiency is borrowing from your future resilience. It’s a loan that comes due when the shock arrives. And from the supply shocks of the 2020s (between the pandemic and wars), every business is now waking up to the fact that the mindset is herding from JUST-IN-TIME to JUST-IN-CASE.

Pair this with the developments in AI and the build out of the five-layer cake that Jensen Huang talks about, we’re basically daisy-chaining all our systems to rely on our grid and needing Six 9’s of reliability and availability to function. Which was where we came from, not where we should be skating to.

Perhaps this is my civil nuclear power plant engineering risk management hat on, but the reality is that (IMHO) we are dangerously over-reliant on our on-demand systems and what I call worshipping at the altar of efficiency, rather than building in resilient systems.

Subscribe now

The Breaking Point is Match Point

Since I do believe most of us, investors, business owners, execs, heads of households even, have underinvested in resiliency, I think it’s best to talk about what the ideal state looks like and not just about how to avoid fragility (though that is a very good place to start).

The ideal is what I’ve found to be Pro-Entropy.

I’ve stolen this term from Antonio Gracias, Founder, CIO and CEO of Valor Equity Partners on Invest Like the Best who famously invested in Tesla and other Silicon Valley notables.

If on one end of the spectrum is Fragile (businesses, households, systems etc.), then the other end is what I think is Pro-Entropy which grows stronger from entropic destruction be it shocks, competition, loss of customers, basically any kind of loss.

Here’s what I propose as the full spectrum:

1. Fragile: breaks under stress

2. Robust: withstands stress

3. Resilient: recovers after stress

4. Anti-Fragile: gains from stress (made famous by Nassim Taleb)

5. Pro-Entropic: creates new order from stress

Power Laws Hold Firm

Pro-Entropy comes in many forms and if you look carefully enough, you see the best Owners, Investors, Heads of whatever are extreme preppers not because of the eventual supply shocks per se (that will of course come) but because it’s the sane thing to do to handle the downside.

More importantly, it’s one of the main “big ideas” that we borrow from Engineering: to engineer the load of a bridge to be X times what it can handle because you want to avoid all bad outcomes at all costs.

We talked about The Blind Men and the Elephant allegory before and while sometimes it is limiting, it is useful to understand a concept that is hard to capture because there are so few examples of it in real life. The Power Law is in effect here: most companies are far below and to the left. So it’s not just on a linear spectrum, it’s more on the exponential in terms of the number of businesses that are able to be aware of and actually have the economics to produce a truly pro-entropic organization.

In other words: they are few and far between.

All Models Are Just … A Proxy

Here’s another assumption worth stripping away: we think we can predict and prepare for specific risks.

Most risk management is built on the idea that you identify threats, assign probabilities and mitigate accordingly. But the most damaging events, by definition, are the ones you didn’t predict. Buffett’s “Big One” (you’ll see his quote below) may come from a “total surprise.” The NZS Capital framing is even sharper: we model the world in bell curves when the world actually runs on power laws.

The first principles reframe here is this: preparing for specific risks is fragile thinking. Preparing for any risk, building the capacity to absorb and reorganize regardless of where the shock comes from, is pro-entropic thinking. That’s the engineering load-factor principle. You don’t design a bridge for the exact weight it needs to carry. You design it for multiples of that weight because you want to survive the thing you didn’t model.

How the Best Describe Pro-Entropic Businesses

These folks say it better than I do, but I want to share the most pertinent quotes that try to get at this idea. They span leaders of Vertical Industries from Insurance, Automobiles, Software, Retail, Money Management, Aerospace, so it’s not just possible in one industry.

“[Pro-entropy] It’s a word we use here internally in our investment framework, and we think about - there are lots who use the word ‘resilient’. And to us, resilient things, resilient companies are things that recover quickly. So when you talk to neuroscientists about the word resilient, they define it as you come out of homeostasis, something happens to you, the adrenaline goes up, cortisol, whatever, and then you recover quickly, you go back to homeostasis and make a good decision. If you don’t recover quickly, then you can’t make a good decision.

Pro-entropic, as you think about a company, if a company’s resilient, it means that it recovers quickly when something happens. There’s a crisis, management’s good, they figure out a pivot, they figure out what to do. For us, pro-entropic, it really is a company that is good at chaos.”

Antonio Gracias, Founder, CIO and CEO of Valor Equity Partners on Invest Like the Best

Subscribe now

Read more

We’ve heard about all sorts of Curves.

Learning curves.

S-curves.

Maturity curves (or is it cycles?)

A recent acquisition of a rather insignificant new acquisition (eNoah1) by Constellation Software (CSU) jogged my pattern recognition of working backwards on why this works.

Having been in this industry for many years, eNoah’s Business Process Outsourcing (BPO) services is nothing new. In fact, it’s actually something really really old. Today the acronyms we all must use start with an “A” and you know which ones I’m talking about in the tech industry these days.

In fact, it’s the only word acceptable to use in the increasing majority of earnings calls of public companies. And for good measure, the management teams have to use it at least a dozen times in order to get their mantra across.

Before we fast forward to today, let’s hop on a quick time machine to look at how a company came to dominate in a region against a behemoth. We’ll then come back to CSU, I promise, it’s worth it.

Around the Curve of the Darien Gap

At the beginning of the “curve” of eCommerce, every smart betting person would’ve assumed Amazon would dominate the continent where its eponymous name is actually located. I’ll give you a hint, it’s the opposite of North America. But turns out that was not meant to be.

MercadoLibre dominates eCommerce today in LATAM with about ~30% of the gross merchandise value and over ~55% in terms of the Ads business (which is Amazon’s fastest growing segment in the US). Amazon still operates in the region but it’s not a brand that dominates in the same way that they do in the US. And it’s not for a lack of trying.

Subscribe now

Here’s the quick timeline of their fight:

Infrastructure Build-Out Phase (2013-2017)

• MELI: Pivots to a fulfillment model with a new logistics arm while launching a credit division for the unbanked.

• AMZN: Enters the region passively with digital goods before launching a limited marketplace for high-income credit card users.

Logistics Arms Race (2019-2021)

• MELI: Opens regional fulfillment centres and launches a private cargo airline to ensure 48-hour delivery.

• AMZN: Launches a subscription membership for video and shipping while prioritizing cloud infrastructure over retail density.

Ecosystem Moat Expansion (2023-Today)

• MELI: Bundles third-party streaming into its loyalty program and secures dominance in the digital retail ad market; expands its credit portfolio to lock in shoppers through its own financial ecosystem.

• AMZN: Scales regional sales events with deep discounts; optimizes for global retail margins but remains a distant second in total regional market share.

At each turn, you can see the fingerprints of Amazon's own playbook successfully orchestrated in their own dominant markets particularly in the US, like bundling of services and advertising.

Subscribe now

The point is that, taken at face value, geographic moats are still a thing. And this is something people tend to forget in a world where software and the internet are supposed to flatten everything. The moat here isn't really the mountains or the oceans though, it's the contextual knowledge that accumulates when you're embedded in a region. The localization needs, the regulatory differences, the cultural nuances of how people buy and sell, these are what create the real barrier. Just look at any maps, you're likely to see the contours of mountains and the curves of the waterways be the places where national and regional boundaries have historically been drawn for a reason. And while the physical barriers matter less in a digital world, the contextual differences they created over centuries still very much do.

Similarly, in this competition between MELI and AMZN, we see (albeit from a very small sample) that two things hold:

1. As investor Mohnish Pabrai will say, being a shameless cloner2 is okay in investing and business (unless it's actual IP). Why start from scratch when you can learn from other people's failures and successes?

2. Watch for the lost-in-translation phenomena. You need to know what to copy and how to learn what works and doesn't work in your region and localization efforts. Having boots on the ground here helps and not just hired proxies (sorry about this phrasing, but it is most apt).

With that, let’s pivot back to eNoah and Constellation Software.

Boring Locally, Works Maximally

Business Process Outsourcing (eNoah's bread and butter) is literally the opposite of AI, Agentic AI and Agents. In a world where companies are in-housing because linear algebra3 done in the trillions of calculations per second can replace many tasks and functions in any company, in any industry, why would outsourcing be something businesses do anymore? That's a question that we've written about and more for another day.

But here’s what I think most people miss about this acquisition. The assumption is that BPO is dead because AI replaces outsourcing. And in mature economies where labour is expensive, that’s increasingly true. But BPO doesn’t exist because companies are too lazy to build software. BPO exists because of where a given economy sits on the labour-versus-technology trade-off. In economies where labour is still relatively cheap compared to software adoption, BPO is the rational choice. It’s not a failure of innovation, it’s just where the isoquant curve says the economics make sense.

We’re here to talk about why CSU does what CSU does which is buy boring companies, doing age old things, in parts of the world where we might not have even heard of before (Chennai, okay I have heard of it before).

I can see how CSU takes advantage of the geographic “curves” above like in the case of MELI vs AMZN and other curves to keep growing. Remember being a “shameless cloner” is okay if you are honest about admitting to yourself that the world is not awash with Power-Lawed companies readily available for sale all the time and that you just have to make do with good-enough businesses to buy and hold.

Here’s what most people miss in why CSU executes the way they do (even though it’s not a grand plan per se) and it all comes back to the many CURVES:

1. Take advantage of the LEARNING CURVE. Buying a BPO company in South Africa in 2018 helps you develop an understanding of how to recognize it in other geographies. And this is a form of intelligence that people underestimate. CSU isn't copying a business model when they buy eNoah. They're copying a recognition pattern. The learning curve isn't really about the business itself, it's about building the acquisition muscle to know what a good BPO-to-VMS conversion opportunity looks like no matter where in the world it shows up. Now, there's a survivorship bias here that's worth naming. We're looking at the pattern from the wins. CSU has made hundreds of acquisitions and the learning curve theory looks clean when you trace it through the ones that worked. Not every geographic pattern transfer lands perfectly and some of those recognition patterns probably misfired in ways we don't see in the numbers. But the compounding effect of doing it over and over again for decades is real even if any individual acquisition can go sideways.

2. Getting ahead of the MATURITY CURVE (or S-Curves). For example, South Africa (e.g. CSU's Adapt IT subsidiary) as an industry for BPO type companies might be further along the maturity curve than India in certain verticals. Mark Leonard said many years ago that the best places to buy VMS companies and a ripe hunting ground are the mature economies. I'm working from memory here, but he said it's two-fold:

   1. First, mature economies are at the point on the curve where companies can afford these software services. What I'm talking about here is the ISOQUANT CURVE.

Read more

Constellation Software recently disclosed their ~13% equity stake in Sabre $SABR. We covered it when the news came out and also did a VMS Deep Dive into Sabre’s products, including its strong network effects.

If you take Constellation’s new PEMS strategy, what they call Permanent Equity Minority Shareholders, it really amounts to being an Activist without the bad connotation. So how will CSU hold up as an ‘Activist Investor’?

First of all, it’s not their first rodeo.

In early 2025 and through their subsidiary Topicus/TSS which primarily operates out of Europe, they acquired a ~25% equity stake in Asseco Poland.

Asseco is massive, a publicly traded Polish IT and software conglomerate. Perhaps the only similarity here is that both Sabre and Asseco are too large to acquire entirely outright, but like Berkshire/Buffett/Munger, they’re happy to be part owners of an excellent business at a reasonable price than a full owner of a mediocre one.

Let’s stop here for a sec.

It’ll be good to explore first what being a significant equity owner of a large publicly traded corporation typically looks like. Because the word “activist” carries a lot of baggage and most of that baggage comes from a very specific kind of owner. But activism itself is just governance. An outside owner applying pressure to a business is neither good nor bad on its own. The question is what kind of pressure and over what timeline and that’s where things diverge dramatically.

Parental Duties

Quite literally and figuratively, being an owner, even a minority one, is like being a parent. They are the parent company in many ways (which is exactly why it’s called that) and that comes with many parental duties, responsibilities and also privileges.

Part of that is being elected to the Board where you can literally take all those roles much more seriously.

On one end of the spectrum we have those new parents who behave like “helicopter parents”.

Helicopter Parenting is often fuelled by anxiety, such as fear for the child’s safety, pressure for them to excel, or the need to compete with other families. Typical behaviours and signs of a helicopter parent:

• Direct Intervention

• Preventing Failure

• Over-supervision

• Excessive Decision-Making

• Reduced Autonomy/Agency

• Lack of Coping Skills

In comparison and on the other side of the spectrum, we have what I term a Wealthy In-Law. It’s not quite your blood relative, but you are tied together and want to do well together. It’s a strategic partnership in the truest sense while providing joint cooperation and present from all the over-reliance and overstepping. There’s still some strings attached and it’s focused on expanding the “family” but not with unconditional love per se. That’s the best I could come up with for now.

While being a Helicopter Parent or the Wealthy In-Law both involves taking significant minority stakes to influence a public company, their timelines, tactics, and ultimate goals are nearly opposites.

Subscribe now

And here’s the thing that’s interesting when you strip away the labels and just look at what’s actually happening. The difference between a Helicopter Parent and a Wealthy In-Law isn’t really about strategy or governance philosophy. It’s about anxiety. The Helicopter Parent is managing their own fear. The Wealthy In-Law is managing the relationship.

There’s actually research on this in developmental psychology. Helicopter parenting consistently produces worse outcomes for children: reduced coping skills, lower autonomy, higher anxiety in the kids themselves. The instinct to intervene and control isn’t really about improving the child’s outcomes, it’s about soothing the parent’s discomfort with uncertainty. And the investing parallel is almost too clean. When an activist is driven by the anxiety of a price gap that needs to close within their fund cycle, every decision filters through that urgency. The company becomes a vehicle for resolving the investor’s timeline problem, not a business being run on its own terms.

Perhaps it’s first best to reference what is the prototypical example of the bad governance practices of Helicopter Parenting, the one which is most prominent in my mind: Bill Ackman’s foray into Valeant Pharmaceuticals1 (which is coincidentally another Canadian-HQed roll-up like CSU).

After the disastrous Valeant misadventure Ackman said these are one of the key lessons he’s learned:

“That managerial competence in deploying capital doesn’t translate to value-added for a business” and that “While I and the rest of the Pershing Square team have suffered significant losses from this failed investment as we are collectively the largest investors in the funds, it is much more painful to lose our shareholders’ money, and for this I deeply and profoundly apologize.”

And if we dissect why it’s like this we see how Helicopter Parents versus Wealthy In-Laws differ on these 4 dimensions:

Perceived vs. Desired Catalyst. This speaks to Mr. Ackman’s lessons learned post-Valeant. It says what the circle of competence is and just because one knows how to allocate capital as an investor, does not make someone know how to turn a company around. Knowing how to price a business and knowing how to operate one are fundamentally different skills. The Helicopter Parent assumes one translates to the other. The Wealthy In-Law knows it doesn’t and that’s why CSU sends an operator to the Sabre’s board, not a portfolio manager. The credibility is earned through operating, not through owning.

Time Horizon. For most activist’s minds this equates simply to price arbitrage. If they woke up and magically the stock was at the price they had projected the value to be, they would likely sell their stake. It’s a thought experiment, an Occam’s Razor for how the stake isn’t a bet on the company, but a bet on the value of the company within a certain amount of time. Not making a judgment on the ethics of it, just exposing the fact that it’s price-first, not business-first. The Wealthy In-Law is business-first. The price takes care of itself over decades if the business compounds. That’s the difference between holding for a fund cycle and holding permanently.

Relationship with Management. This ties closely with the prior one but Helicopter types are more adversarial and overbearing overall. This is in comparison to the other side of the spectrum of those that prefer collaboration, cooperation and win-win. And it makes sense when you think about the anxiety framing. If you’re anxious about your timeline, every conversation with management feels like a negotiation. If you’re not, it feels like a partnership.

Subscribe now

Public vs. Private Tactics. The tactics used in the past by Ackman and Pershing Square Capital have historically been very front-facing. Typically they take the form of public presentations, admonishments on webinars and on appearances in financial news networks to get the word out. If you haven’t seen the HBO show Industry, you should consider it. There is a scene in Season 4 right on the money about this very playbook. In contrast, private tactics like the ones preferred by Constellation with Asseco and Sabre have been more cordial and agreements are more on the quiet side (unless mandated by regulations for disclosure). The Helicopter Parent needs the public pressure because the timeline demands it. The Wealthy In-Law doesn’t need public pressure because the timeline doesn’t.

As we’ll see in the next section, David McKay (CEO of Vela Operating Group at Constellation) and now newly appointed Board member of Sabre, highlights exactly how CSU actively chooses to be the Wealthy In-Law.

Highlights from the New Kid on the Board

In the interview2, it’s what McKay does not say that says it all. In classic Constellation Software and Mark Leonard fashion, he’s very short on tactics but very, very long on strategy and thinking. It seemed the person who interviewed him did not do his homework as you’ll probably see in some of his questions.

1. No changes to personnel. This alone tells you a lot. A Helicopter Parent’s first move is almost always personnel changes because that’s the fastest lever to pull. McKay doesn’t touch it.

2. No changes to CSU’s portfolio in the travel vertical in particular. No news here, but worth noting the discipline.

3. CSU is taking this as a partnership, not about control while he’s on the board. As we wrote about, CSU is embarking on a Berkshire Hathaway $BRK.B-like holding company model whereby they have full ownership of businesses and also minority stakes in their equity portfolio (Sabre being one of their first). I’ll be writing more about this later in the week.

4. Notably McKay will also be on Sabre’s “Technology Committee”. An operator on the tech committee. Not the finance committee. Not the compensation committee. The technology committee. That’s where an owner-operator adds value.

5. When asked about how they would “measure” if this venture is success or not, McKay’s answer was: “We would measure as it relates to all shareholders and over a very long period of time.”

That last one is the whole piece in one sentence. “All shareholders” and “a very long period of time.” No price target. No timeline for returns. No catalyst they’re engineering. Is this business getting better for everyone who owns it, measured in years and decades?

So the real thing here is about stewardship. The word “activist” has been so thoroughly captured by the Helicopter Parents that we’ve forgotten what it actually means. To be active in the governance of a business you partially own. To show up, contribute, and think long-term. CSU isn’t redefining activism. They’re just doing the version of it that actually works.

-&

Subscribe now

New build civilian nuclear power generation was going the way of the Dodo bird in the early 2000s when I first joined the industry. Fast forward to today and with the ever-hungry demands of data centres for AI, nuclear is now suddenly the only reliable name in town.

Trust me, I’ve been there when it was very much on the outs. When compared to the early 2010s (as you see below), the planned projects for nuclear is an incredible build out like we haven’t seen before both in the capacity under construction and also planned/proposed. So this is a growing industry1 for many years and as I’ve discussed later, these are very very long lifecycles in the Nuclear Vertical.

Like I’ve said, I’ve worked close to 15 years in this industry first as an Engineer, designing the control rooms for the operators and also various systems in what they call the Balance of Plant including special safety systems if shit hits the fan.

Working in the nuclear industry as an engineer is supreme risk management: there is a back-up of the back-up and there is a back-up of that back-up. You learn very early on how to anticipate and mitigate against risks of all sorts. Might be why I also love investing or at least have a natural affinity for it.

This is a great segue into the nuclear’s background first before we dive into the Vertical Market Software that services this industry. And it’s the reason for why there is so much risk engineering and processes involved that developed over the last few decades.

Learning The Hard Way

The nuclear industry has a grim relationship with progress. Almost every major leap in safety standards and regulatory rigour came after something went terribly wrong. Three accidents in particular shaped the regulatory environment that exists today and they’re essential context for understanding why software in this vertical operates the way it does.

Three Mile Island (1979)

March 28, 1979. A cooling system valve gets stuck open in Unit 2 of the Three Mile Island plant in Pennsylvania and the operators don’t have the instrumentation to see what’s actually happening inside the reactor. What follows is a partial core meltdown, the worst nuclear accident in American history at that time and the moment that changed how the entire U.S. regulated nuclear power.

Nobody died. But the regulatory fallout was enormous. The US Nuclear Regulatory Commission (NRC) expanded its on-site presence to two full-time resident inspectors at every operating plant. They established a 24/7 operations centre. And they mandated immediate notification for any serious event, no more waiting around. The industry itself created INPO (Institute of Nuclear Power Operations) as a self-regulation body because the trust between the public, the operators and the government had been shattered. New reactor licensing effectively froze for decades.

Subscribe now

Chernobyl (1986)

April 26, 1986. Soviet engineers at Unit 4 of the Chernobyl plant are running a safety test and they push the RBMK reactor into a configuration it was never designed to handle. The reactor had a fundamental design flaw, a positive void coefficient, which meant that if you lost cooling the chain reaction actually accelerated instead of slowing down. A power surge destroyed the core, blew the roof off the building and released radioactive material across much of Europe.

Chernobyl changed everything at the international level. The IAEA developed two new conventions for early notification and emergency assistance because there was no framework for one country’s nuclear disaster affecting its neighbours. The remaining RBMK reactors worldwide got significant redesigns, control rods, neutron absorbers, fuel enrichment, all of it. Several countries imposed outright moratoriums on new nuclear construction. Italy shut its program down entirely. Germany started its long phase-out.

For the context of this deep dive, Chernobyl is the event that made nuclear safety a global regulatory conversation, not just a domestic one. Before Chernobyl, each country more or less set its own rules. After Chernobyl, there was an expectation of international transparency and cross-border accountability that didn’t exist before. That layer of international regulatory complexity is part of what makes this vertical so deeply specialized.

Fukushima (2011)

March 11, 2011. A 9.0-magnitude earthquake off the coast of Japan followed by tsunami waves exceeding 10 metres knocks out the backup generators at Fukushima Daiichi. Three reactors lose cooling and sustain severe core damage. Hydrogen explosions blow apart the reactor buildings. The IAEA rates it a Level 7 event, the same severity as Chernobyl.

Japan restructured its entire nuclear regulatory framework, dissolving the old body and creating an independent Nuclear Regulation Authority with significantly expanded authority. The U.S. NRC mandated new requirements for maintaining key safety functions during total loss of electrical power, better spent fuel pool monitoring and improved pressure venting systems. Globally, the industry shifted to planning for multi-reactor accident scenarios and natural disaster combinations that had previously been treated as too improbable to engineer for.

Subscribe now

Fukushima is critical context for VMS in nuclear because it added another layer of regulatory density on top of what was already the most regulated industry in the world. Every piece of software that touches safety assessments, risk modelling, asset management or compliance documentation in this vertical exists because of what these three accidents taught the industry. The regulatory bar doesn’t just go up, it compounds.

Nuclear 1.5 Era

Nuclear 1.0 was really Oppenheimer, so if you don’t know you just go watch that brilliant Nolan film.

What came as a result of all the research labs around the world was that the only people who could, 1) reasonably afford to fund a civilian nuclear power program and 2) be sanctioned in order to do so were the nation-states.

So between WWII and well into the Cold War, only a handful of countries would have the might and bandwidth to develop civilian nuclear programs. The rest of the countries who weren’t on the bleeding edge would be the ones importing the technology.

The makers and the IP owners of this era were:

1. USA

2. Russia

3. Canada

4. Japan

5. France

6. Great Britain

Of course all the maker countries, dog-fooded (dog-fed?) their own nuclear tech to generate electricity.

And the primary buyers, typically other nation-states, were in this Nuclear 1.5 era:

1. China

2. India

3. South Korea

4. Romania

5. Argentina

6. There’s many more here, but these are the ones I know.

Of the countries that were makers, many of them would settle into this model: economies of scale. Each successive generation of plants in this era were incrementally larger, higher in total megawatt output because it’s freak’in hard to get a nuclear power plant designed, approved, built and commissioned. And it’s because of the aforementioned accidents that have happened almost every decade since we’ve had nuclear powering our homes and buildings.

Subscribe now

Nuclear 2.0+

This is more or less the era we are in now. Roughly post 2018 I would say and definitely post the ChatGPT moment.

While I was still in the industry, there was already in development (though very quiet and very small) of SMRs or small modular reactors. There were a few quietly developing in the background with whatever seed money they had and frankly they were never really going anywhere since no new nuclear development was gonna happen. It also speaks to why in Nuclear era 1.5+ that it was always incremental changes to the same underlying plant design (even though in most cases much of the guts was swapped out and improved immensely) because the regulatory hurdles were immense to jump over.

And so while SMR was a very early-stage development thing, NuScale being one of them, the bar was even higher since it was a novel design that regulators had to deal with. (BTW, I have experience dealing with regulators and let me tell you it’s a process, think slow, long and deep.)

But data centres and their insatiable hunger for electrons to power AI has made that all but go away. What used to be fear from 3-Mile Island is now replaced by hope of powering likely every workflow across every industry. And that’s where we are today. Without getting into the specifics of the nuclear tech too much, really the story here is the twists and turns of how the nuclear industry evolved from the development of weapons of mass destruction to civilian use today. We’ll talk more about the value chain in the next section but just know the main ‘actors’ in this Vertical and their duties and responsibilities:

1. Regulator (domestic and international, e.g. IAEA)

2. Plant Operator (gov’t or non-governmental or hybrid)

3. Uranium (or thorium or equivalent) Miner

4. Fuel Bundle Supply Chain

5. Vendor Suppliers

6. Nuclear Power Plant Design/IP Owner

7. Engineering and/or Construction specialists

8. The Public

What It All Comes Down To

A nuclear power plant is really an entire highly-specialized supply chain. Given how large it is, you typically work with 3rd party vendors that are highly specialized and niche. In other words, you either work in nuclear or you do something horizontal. Which means it’s very much a hyper-Vertical if we ever were to review one.

Let’s take a small example of why that is the case.

Given situations like Fukushima but even well before that six-sigma event, earthquakes or acts of god need to be part of engineering requirements for any plant. As an Engineer, every piece of equipment, especially anything that either contributes directly to or even indirectly to the functioning of the plant under very adverse conditions needs to be designed for the eventualities including an earthquake. So all the equipment that we source needs to be seismically qualified2.

This goes for things you wouldn’t even consider like my having to write this requirement into the HVAC system we had to source from a vendor. Imagine that the 3rd party air conditioning unit we need has to be shaken on a table, violently, and all the screws still need to be in place and the thing needs to be functioning still. You cannot just go to your closest Home Depot to source this. Besides these are massive HVAC units. Did I mention we need many of them because ones needed to be the back up of the back up of the back up!?

So that’s all to say that NUCLEAR as an industry vertical is extremely niche.

Subscribe now

Everyone who works in this vertical, whether designing custom controls or sourcing from 3rd parties needs a high degree of understanding of the regulatory and very stringent requirements in order to piece this very complex system together.

And this mix of complexity and regulatory high-bar is why Vertical Market Software (should there be any) would do well to service this space given the height and depth of its economic moat. Not to mention, if the goal of VMS is to have a strong economic moat around being “mission-critical” I cannot think of a better Vertical to be in since the cost of errors in this space is not just fatal, but catastrophic.

But it also is why, unlike other industries, the core systems (both mechanical, electrical, chemical) parts of the plant will be retained by the electricity operator of the plant. Unlike other Verticals where a 3rd party VMS provider could be placed at the “core”, such as the Core Banking Systems that many regional banks may rely on, the core parts of a nuclear power plant cannot be outsourced. Instead, this vertical can be broken up into sub-niches that still service mission-critically.

We’ll explore this next.

But before we get there, something to keep in mind at a high level.

Nuclear power generation is like other power plants. Effectively the energy source powers some turbines and in turn those spokes generate electricity. It’s not different in that sense than a gas-turbine electricity generation. The only difference here is that the core parts of the nuclear power operations and of course the fuel itself is fission materials like uranium or more recently thorium. That’s all to say that there are many parts that are commonly used and then the core part of it that is highly specific to nuclear needs.

Vertical Nuclear: Value Chain

Here’s the value/life-cycle chain of nuclear power plants:

It’s a lot.

The parts of it that I’ve mentioned before that are typically in-housed entirely are the Operations, Fuel & Waste Management and Radiation Protection. Which means there are many parts of the lifecycle of a nuclear power plant that require outside parties.

Again VMS here isn’t exactly the entire vertical like other industries since the core systems are typically developed custom in-house with some specialty outside vendors. The key here is the custom designed aspects that make it fit-for-purpose. So a conventional 3rd party VMS wouldn’t be able to develop once and apply it uniformly. Plus the regulatory hurdle is very high. The strategy here for most software vendors will be to take parts of the value chain as their own sub-Vertical. Those sub-Verticals are highly specialized, less directly regulatory intensive and have enough of the common business/workflow that warrants developing software that a business can sell into.

Let’s explore where along the sub-Verticals along the value-chain developing a VMS makes sense:

• Reactor Design (Preliminary and Detailed)

• Scaled Operations (Fuel & Waste Management)

• Quality and Qualifications (Testing, Re-qualification, Maintenance, Cybersecurity)

• Radiation Protection and Public Safety (Safety Assessments/Audits, hardware/software systems to measure radiation levels at or around the plant and access controls)

Let’s go one by one now.

Reactor Design (Preliminary and Detailed)

This is the part of the value chain where the reactor itself gets modelled, simulated and stress-tested before anyone pours concrete. The software here is specialized to the point where most people outside the industry wouldn’t even know it exists. You’re talking about thermal-hydraulic codes, pressure vessel design calculators, digital twins of control rooms and simulation platforms that model how neutrons behave under extreme conditions.

Read more

I’ll be the first to admit I can’t name more than ten of CSU’s business units off the top of my head. But I know this,…

Read more

I want to try something different today. I want to make the case that we can actually predict, with reasonable confidence, what this transition looks like. And the reason I believe that is because three completely independent areas of science and engineering already have frameworks for exactly this kind of problem. They’ve been studying human-machine systems, organizational shape, and competitive adaptation for decades. The answers are sitting right there if you know where to look. As Charlie Munger said you need to build a latticework of big ideas and that the value of applying those big ideas sit at the intersection of the major theories, just like below’s graphic:

Here are the three we’ll focus on today:

1. Function Allocation from Human Factors Engineering, on how individual knowledge work will shift.

2. Aerodynamics from Mechanical Engineering, on how AI changes the velocity your competitors can move at.

3. Convergent Evolution from Evolutionary Biology, on how the shape of organizations will change and likely take the same shape across industries.

Let me walk through each one.

Read more

Subscribe now

One power move that has pervaded Charlie Munger’s latticework of mental models is the concept of inversion.

Munger’s thing was always: don’t ask how to succeed, ask how to guarantee failure, and then don’t do that. Flip the problem around. Work it backwards.

And it hit me that if you really want to understand why Constellation Software works, the place to start is what they refuse to touch.

Because CSU has done over a thousand acquisitions since 1995. Mark Leonard and his operating groups have looked at tens of thousands of software companies. They have more deal flow than probably any single technology acquirer on the planet. And yet there are entire massive software verticals where they have zero presence. Not “light exposure” or “one small toehold.” Zero.

I think if you invert it, if you study the shape of the hole, you end up understanding the shape of the machine better than if you just catalogued what’s inside it.

The Anti-Portfolio

I’ve identified four major software verticals where CSU has had every opportunity to participate and has chosen not to. Not once. These aren’t obscure categories. They’re massive, well-known, and in some cases wildly profitable for the companies that dominate them. But they share structural characteristics that make them fundamentally incompatible with the way CSU makes money.

1. Electronic Design Automation

Read more

Read more

Damian McKay, who runs Vela Software Group for CSU, got appointed to Sabre’s board. CSU holds roughly 13% of Sabre.

Sabre is a publicly traded company doing close to $3B in revenue but sitting on $4.5B in debt! Perhaps CSU can help lend a hand.

So what does Constellation see in Sabre in its product and as an investment?

And what does it tell us about where they think travel technology vertical is as AI rises?

What is Sabre’s Primary Product

I think most people outside the travel industry don’t really know what Sabre does and the people inside the industry take it so for granted that they’ve stopped thinking about what it is.

SabreMosaic Travel Marketplace (their product name) is a Global Distribution System, a GDS. When you book a flight through a travel agent or a corporate travel portal that transaction very likely flows through Sabre’s network.

The GDS connects airlines and hotels and car rental companies to the travel agencies and booking tools that sell their stuff. Real-time availability, pricing, booking, all of it aggregated across hundreds of suppliers. Sabre alone connects to over 400,000 travel agencies in over 150 countries.

Fascinating origin story on SABRE

In early 1950s the president of American Airlines met an IBM salesman on a flight and they got talking about automating reservations. That turned into SABRE (Semi-Automated Business Research Environment) which went live later that decade.

It’s an acronym!

By ‘76 they were putting terminals in travel agency offices. By the late 80s it was one of the biggest real-time data processing systems on the planet. American eventually spun it off, TPG and Silver Lake took it private for a while, and it’s been public on the NASDAQ since 2014.

Here’s the thing about a GDS that makes it valuable. It sits between supply and demand and makes itself essential.

On one side: Agencies need it to see inventory from hundreds of suppliers in one place.

The other side: Suppliers need it to reach hundreds of thousands of sellers globally.

But Wait, There’s More: Let’s Count All The Moats

Being Number Two Is Better Than Three I Guess

Three players are dominate in this market and Sabre is one of them. And if you look at what’s hiding in plain sight is what CSU minimum criteria for their subs:

Good Businesses

We are pleased to hear from principals or representatives of Good Businesses that meet the following criteria:

• Number 1 or Number 2 market-share holder in a niche vertical market

• Revenues of at least $5-million

• Hundreds or thousands (not dozens) of customers

• Unimposing competitors

• An offering price that has been determined

A good business, even if they cannot buy it outright.

As we’ve mentioned Amadeus is the other one and is the biggest at roughly over a third of global air distribution, headquartered in Spain, most diversified.

Sabre is ~35% share and strongest in North America.

Travelport is smallest at around 20%, competing more on price and trying to be the developer-friendly option.

Competition is for Losers (Sabre’s Strongest Moat)

But with these 3, even the regulators have consider them not just dominant but near-monopolies or oligopoly.

Take for example that the EU had numerous probes into Amadeus but however closed without fines.

And since 2011, Sabre has face various antitrust scrutiny with airlines because Sabre’s GDS contract rules impeded travel agents and others from using less expense alternatives for booking seats, and imposed an unduly restrictive distribution agreement.

Read more

I’ll start by sharing a personal anecdote, but just wait for it and how …

Read more

Intercom went from a breakout SaaS company, $1M to $50M ARR in three years, to betting their entire 1,200-person company on AI within weeks of ChatGPT’s release. They called it a “one-way door for humanity” and an “iPhone moment.” They ripped up their strategy. They let go of a third of their staff. They banned their own design principles. They built an AI agentic product and business-line called Fin that now handles over a million queries a week.

I went through something similar. Smaller scale, different industry, same pain. I was the Lead PM on the AI product team at a biotech SaaS company that had built a platform cataloguing scientific literature, both public and proprietary, so that scientists doing drug discovery could validate hypotheses, generate new ones and understand risks. When ChatGPT landed, the pressure was immediate. Users wanted us to be more useful than ChatGPT. And leadership saw it as existential.

What followed was one of the hardest professional experiences of my life. And I suspect a lot of people reading this are either going through it right now or about to.

Let’s break it down into a key aspects:

1. The ‘Everything, Everywhere All At Once’ Moment

2. Making a Killing, but first Killing a Making

3. The Compounding Reliability Problem

4. If Selling the Demo, Then what is the MVP?

5. Resistance is Futile (but oh so real)

6. Running Up That Hill (of two companies)

1. The ‘Everything, Everywhere All At Once’ Moment

Intercom’s version: Leadership framed the transition as non-negotiable. The CPO told the company explicitly: adopt AI and we become category leaders at $100M ARR. Fail to adopt and we face irrelevance, down rounds, and our best people leaving to join real AI companies. There was no middle ground. No “dipping your toes in.” Employees who suggested delaying AI initiatives to the next quarter were told to leave.

What I saw: The push came from our CEO too. But here’s where our experience diverged from Intercom’s in a critical way. Our CEO had the right instincts on direction but didn’t fully grasp the technical and strategic limits of what we were building. He didn’t understand the difference between a deterministic product and a probabilistic one. And that gap between vision and understanding created real problems downstream for those of us tasked with building the thing.

Subscribe now

The old product was deterministic. You searched the catalogue, you got results, the results were accurate and verifiable. The new AI product was probabilistic. It synthesized across literature using RAG and data chunking. Sometimes it was brilliant. Sometimes it hallucinated. And the question of how to make it accurate, precise, truthful and observable was a fundamentally different engineering challenge than anything we’d done before. Going from zero to one on that was brutal.

Lessons Learned: The “this is life or death” framing is necessary. But it’s not sufficient. Leadership has to understand what kind of death they’re avoiding and what kind of life they’re building toward. Intercom’s leadership seemed to grasp both. Ours got the direction right but underestimated the complexity of execution.

2. Making a Killing, But First Killing a Making

Read more

A16z partners Alex Rampell and David Haber recently sat down to talk about moats in the age of AI1. And the more I listened the more I kept thinking, this is just Mark Leonard’s playbook. Almost word for word. The same ideas, the same instincts, the same conclusions, just dressed in different language and arrived at from completely different starting points.

Now I’ll be the first to say this could be confirmation bias on my part. I spend a lot of time thinking about $CSU.TO and VMS economics so maybe I’m just pattern-matching everything back to Constellation. But I don’t think so. I think the reason they converge is because the underlying truths about software defensibility haven’t actually changed that much, even in the age of AI. And when two very different observers, one deploying venture capital in San Francisco and one acquiring vertical market software companies across 150+ niches, keep pointing at the same things, that’s signal.

Here’s what I mean.

Goldilocks

Rampell introduces this concept he calls the “Goldilocks zone of irrelevance” and it’s one of the best framings I’ve heard for why certain software companies are nearly impossible to displace.

His example is janitorial services. If you walk into the office of a CEO at a major corporation and tell them you can get their toilets 9% cleaner and save 1% on janitorial services spend, they won’t just say no. They won’t even exercise the mental energy to find the person who cares. The spend is too small relative to the business, the switching risk is too high relative to the reward, and nobody’s career is going to be made by optimizing the janitorial contract.

Read more

Not the Twitter version of Taleb where everyone throws around “antifragile” like it means “tough.” I mean the actual framework. The one that says some systems don’t just survive disorder, they actually gain from it. The mechanism is optionality: having more upside than downside from random events.

And I keep coming back to one company that I think is a near-perfect real-world implementation of this framework and it wasn’t even designed that way on purpose.

$CSU. Constellation Software.

Let me walk through why.

A Naturally Antifragile Structure

Constellation has 1,000+ independent operating businesses across 150+ verticals. I want you to sit with that number for a second because the implications are wild when you think about it through Taleb’s lens.

No single failure can cascade systemically. Each unit absorbs its own volatility locally. And a struggling subsidiary gets attention, capital and operational intervention from a portfolio that’s simultaneously benefiting from unrelated tailwinds elsewhere.

Subscribe now

That’s not just resilience (bouncing back). That’s antifragility. The system actually gets stronger from the disorder because the healthy parts fund the recovery of the weak parts and the whole organism keeps compounding.

Compare this to a single-product software company. One bad product cycle, one platform shift, one AI disruption and the whole thing is threatened. For Constellation, volatility in any one vertical is just noise at the portfolio level.

I think this is under-appreciated by the market.

The Hidden Barbell

Taleb’s barbell strategy is about combining extreme safety with extreme optionality and avoiding the “middle” which he considers the most dangerous zone.

Constellation runs a version of this without explicitly naming it.

On one end of the barbell:

• Incredibly stable, mission-critical VMS businesses

• 90%+ customer retention

• Negative churn

• Recurring revenue

These are the safe side. The boring, Lindy-tested, not-going-anywhere software that runs utilities and municipalities and dental offices.

On the other end:

• Constant small, asymmetric acquisition bets in obscure niches

• Downside is capped (small deal size, limited integration risk)

• Upside is uncapped over decades of compounding

And what they explicitly avoid is the dangerous middle. Large, consensus, well-understood deals where everyone is competing and the risk/reward is mediocre. Mark Leonard has written about actively avoiding “lottery ticket” large deals precisely because they violate this logic.

I would say this is one of the most disciplined capital allocation frameworks in public markets and it maps almost perfectly onto Taleb’s barbell. Whether Leonard read Taleb or just arrived at the same place through experience, the result is the same.

Black Swans: Defense and Offense

Here’s where it gets really interesting.

Most companies are negatively exposed to black swans. A pandemic, a credit crisis, a technology disruption hits them hard. Constellation has engineered itself to be positively exposed to black swans on the upside and robustly defended on the downside.

Think about the private credit scenario (i.e. Blue Owl Capital2). A tail event that destroys the leveraged PE acquirer model. For Constellation, that same black swan is an opportunity accelerant. Their FCF-funded, zero-leverage acquisition model means they don’t just survive the disruption. They feast on the distressed assets it creates.

Read more

Developers saying they are already tired of working with AI for their projects. “It’s like managing some interns, instead of crea…

Read more

But at the enterprise level, particularly what we talk about here on P&I Nerd, is there an asymptote? And if so, how do we arrive at a possible “ceiling” for it?

Here are some questions to ask:

1. Is there an asymptote for enterprise/business use of intelligence?

2. Is there an asymptote for humans using intelligence?

3. Is there an asymptote for intelligence applied to each Vertical?

4. What am I not imagining here because of the limitations of my own intelligence?

Let’s go one by one shall we?

1. Is there an asymptote for enterprise/business use of intelligence?

If we think about any business and to put on my Project Manager hat, there's a 3-vector model we can apply to thinking about improvements at the Enterprise level due to AI that are known today:

1. Time

2. Cost

3. Scope

Time

AI and Agentic AI in particular is everyone’s darling and what people have been calling a true time compressor. I can agree with that.

With more time, we can output more. Fair.

Cost

Does it save us money?

Well some seem to think so. You can vibe-code your way to better marketing software for your 50-branch mid-sized Dental Office instead of paying an agency. Done! Time saved and money saved.

In another context, money gets saved in other ways too. Like if you own an Accounting firm and you no longer need the services of your bookkeepers because all the journal entries are magically all aligned and perfect. Less labour.

Then there’s what happened to the trucking industry (I really mean trucking literally here) when SemiCab1 came along. SemiCab is a logistics AI company that built what they call a Collaborative Orchestration system. The idea is pretty simple but hard to do at scale: instead of trucks running half-empty on return trips or making inefficient routes, their AI coordinates across multiple shippers and carriers in real time to fill capacity and optimize routes.

Their white paper showed they can save both time and money by doing what no human dispatcher could do across that many variables simultaneously. Pretty cool if you ask me.

Btw, so can your competitors take up all these things. The question is, do people share these improved margins with their customers? Maybe. Dentists might have more affordable services and undercut their competition. Would they? I dunno.

It saves costs/money.

Subscribe now

Scope

If we define scope as how much you can do and how far you can go in your Jobs-To-Be-Done for your end customers, then I would say AI helps extend the depth and breadth. But as these things go, you have to know where you’re digging first.

AI can help your people find better answers faster and make more informed, timely decisions. It can think about a problem longer than any one person can, given that it never needs to sleep as long as there are tokens available for the taking.

And the outputs tend to be more robust when given the right requirements, which means product quality goes up without proportionally more effort.

So scope is well supported by AI.

2. Is there an asymptote for humans using intelligence?

Read more