When efficiency becomes a weapon: AI rewards cognition, not numbers.

Author: Naman Bhansali

Compiled by: TechFlow TechFlow

Original title: AI will not achieve technological equality, but will only reward the right people.

TechFlow Dive: In the early stages of new technology adoption, people often have the illusion of "technological equality": when photography, music creation, or software development become effortless, will competitive advantages disappear? Warp founder Naman Bhansali, drawing on his personal experience of moving from a small Indian town to MIT and his entrepreneurial practice in the AI-driven payroll field, profoundly reveals a counterintuitive truth: the lower the barrier to entry (floor) for technology, the higher the industry's ceiling (ceiling) actually rises.

In an era where execution has become cheap and can even be "vibecoded" by AI, the author argues that the true moat is no longer simply traffic distribution, but rather an unforgeable "taste," a deep understanding of the underlying logic of complex systems, and the patience to continuously compound interest over a decade. This article is not only a sober reflection on AI entrepreneurship, but also a powerful demonstration of the power law that "common technology leads to aristocratic results."

The full text is as follows:

Whenever a new technology lowers the barrier to entry, the same prediction inevitably follows: since everyone can do it now, no one has an advantage anymore. Camera phones made everyone a photographer; Spotify made everyone a musician; AI has made everyone a software developer.

These kinds of predictions are always half right: the floor has indeed risen. More people are involved in creation, more people are launching products, and more people are joining the competition. But these predictions always ignore the ceiling. The ceiling is rising even faster. And the gap between the floor and the ceiling—that is, between the median and the top—is not narrowing, but rather widening.

This is the characteristic of power laws: they don't care about your intentions. Techniques for equality always produce elitist results. Every single time.

AI will be no exception, and may even behave in more extreme ways.

Evolution of the market

When Spotify launched, it did something truly radical: it gave any musician on Earth access to distribution channels previously only attainable by record labels, marketing budgets, and exceptional luck. The result was an explosion in the music industry—millions of new artists emerged, and billions of new songs were released. The bottom line truly rose as promised.

But what happened next was that the top 1% of artists now capture a larger share of streams than they did in the CD era. It didn't shrink; it expanded. More music, more competition, and more ways to find quality content have led listeners, no longer limited by geographical location or shelf space, to flock to the best. Spotify hasn't achieved musical homogenization; it has merely intensified this competition.

The same story unfolds in writing, photography, and software. The internet has spawned the largest number of authors in history, but it has also created a more brutal attention economy. More participants, higher top stakes, and the same fundamental structure: a tiny minority captures the vast majority of the value.

We are surprised by this because we are used to thinking linearly—we expect productivity gains to be distributed as evenly as water being poured into a flat container. But most complex systems don't work that way; they never have. Power-law distribution is not a market quirk or a technological failure; it's nature's default setting. Technology didn't create it; it merely revealed it.

Consider Kleiber's Law. In all life on Earth—from bacteria to blue whales, spanning 27 orders of magnitude in body weight—metabolic rate is proportional to body weight raised to the power of 0.75. A whale's metabolism is not proportional to its size. This relationship is a power law, and it maintains extremely high precision in almost all life forms. No one designed this distribution; it is simply the form in which energy manifests itself as it follows its inherent logic within complex systems.

Markets are complex systems, and attention is a resource. When friction disappears—when geography, shelf space, and distribution costs no longer act as buffers—markets converge to their natural form. This form is not a normally distributed bell curve, but a power law. Stories of equality coexist with the consequences of elitism, which is why every new technology catches us off guard. We see the baseline rising and assume the ceiling is following at the same pace. This is not the case; the ceiling is accelerating away.

AI will accelerate this process faster and more forcefully than any previous technology. The baseline is rising in real time—anyone can release a product, design an interface, and write production code. But the ceiling is also rising, and faster. The crucial question is: what ultimately determines your position?

When execution becomes cheap, aesthetics become the signal.

In 1981, Steve Jobs insisted that the circuit boards inside the first Macintosh had to be aesthetically pleasing. Not the exterior, but the interior—the part customers would never see. His engineers thought he was crazy. But he wasn't. He understood something easily dismissed as perfectionism, but actually closer to a kind of proof: the way you do anything is the way you do everything. A person who can make the hidden parts aesthetically pleasing isn't performing quality; he simply has an inability to tolerate releasing any defective products.

This is important because trust is hard to build but easy to forge in a short time. We constantly run heuristics, trying to figure out who is truly excellent and who is just performing excellence. Credentials are helpful but can be manipulated; Pedigree is helpful but can be inherited. What's truly hard to forge is taste—a lasting, observable, and unquestionable adherence to a certain standard. Jobs didn't have to make the circuit boards so beautiful. The fact that he did tells you in itself what he would do behind the scenes.

For much of the past decade, this signal has been somewhat masked. During the heyday of SaaS (roughly 2012 to 2022), execution became so standardized that distribution became a truly scarce resource. If you could acquire customers efficiently, build a sales machine, and achieve the "Rule of 40"—the product itself was almost irrelevant. As long as your go-to-market strategy was strong enough, you could win with a mediocre product. The signals of aesthetics were drowned out by the noise of growth metrics.

AI has revolutionized the signal-to-noise ratio. When anyone can generate a functional product, a beautiful interface, and a working codebase in an afternoon, whether something is "easy to use" is no longer a differentiating factor. The question becomes: Is this thing truly excellent? Does the person know the difference between "good" and "insanely great"? Even without being forced, do they care enough to bridge that last little gap?

This is especially true for business-critical software—systems that handle payroll, compliance, and employee data. These aren't products you can casually try out and abandon next quarter. Switching costs are real, failure modes are severe, and the people who deploy the system are accountable for the consequences. This means they'll run all the trust heuristics before signing on. A beautiful product is one of the loudest signals it can send. It says: the people who built it put their heart into it. They care about what you can see, which means they're likely to care about what you can't see either.

In a world where execution is cheap, aesthetics become proof of work.

What are the rewards for the new phase?

This logic has always held true, but market conditions over the past decade have made it almost invisible. There was a time when the most important skills in the software industry were not even related to the software itself.

Between 2012 and 2022, the core architecture of SaaS was largely established. Cloud infrastructure was inexpensive and standardized, and development tools matured. Building a functional product was difficult, but it was a "solved difficulty"—you could hire people, follow established patterns, and reach a passing grade as long as you had sufficient resources. What was truly scarce, what distinguished winners from mediocre ones, was distribution capability. Can you acquire customers efficiently? Can you establish repeatable sales actions? Do you have a sufficient understanding of unit economics to fuel growth at the right time?

The founders who thrived in that environment mostly came from sales, consulting, or finance. They were intimately familiar with metrics that sounded like gibberish a decade ago: Net Retention Rate (NDR), Average Contract Value (ACV), the Magic Number, and the 40 Rule. They lived amidst spreadsheets and sales pipeline audits, and in that context, they were indeed right. The SaaS boom gave rise to the SaaS founders of the boom. It was a rational evolutionary adaptation.

But I felt suffocated.

I grew up in a small town in a state in India with a population of 250 million. Each year, only about three students from the entire country of India get into MIT. Without exception, they all come from expensive preparatory schools in Delhi, Mumbai, or Bangalore—institutions specifically built for this purpose. I was the first person in my state's history to get into MIT. I mention this not to boast, but because it's a microcosm of the argument in this paper: when entry barriers are high, prestige predicts the outcome; when entry barriers are high, deep people always prevail. In a room full of people from privileged backgrounds, I was a chip that won by depth. It was the only bet I knew how to make.

I studied physics, mathematics, and computer science, and in these fields, the most profound insights didn't come from process optimization, but from seeing truths that others missed. My master's thesis was about straggler mitigation in distributed machine learning training: how to optimize a constraint that falls behind when you're running a system at scale, without compromising the overall integrity.

When I was in my early twenties looking into the startup world, I saw a picture where these profound insights seemed irrelevant. The market premium went to "go-to-market," not the product itself. Building something technologically superior seemed naive—it was seen as a distraction from the "real game" (i.e., customer acquisition, retention, and sales speed).

Then, at the end of 2022, the environment changed.

What ChatGPT demonstrates—in a way that is more intuitive and impactful than years of research papers—is that the curve has bent. A new S-curve has begun. Phase transitions don't reward those who adapt best to the previous phase, but rather those who can foresee the infinite possibilities of the new phase before others even see the price.

So I quit my job and founded Warp.

This bet is very specific. The U.S. has over 800 tax agencies—federal, state, and local—each with its own filing requirements, deadlines, and compliance logic. There are no APIs, no programmatic access interfaces. For decades, every payroll provider has dealt with this in the same way: by stacking people. Thousands of compliance experts have manually navigated these systems that were never designed to operate at scale. The traditional giants—ADP, Paylocity, Paychex—have built entire business models around this complexity, not by addressing it, but by absorbing it into their workforce and passing the costs on to their clients.

In 2022, I could see that AI agents were still fragile. But I could also see a curve of improvement. Someone deeply involved in large-scale distributed systems, closely observing the trajectory of model evolution, could make a precise bet: the technology that was fragile then would become incredibly powerful within a few years. So we took the bet: to build an AI-native platform from first principles, starting with the most difficult workflow in this category—the workflow that traditional giants could never automate due to architectural limitations.

Now, that bet is paying off. But the bigger picture lies in pattern recognition. Technical founders in the AI ​​era possess not only engineering advantages but also insightful advantages. They can see different entry points and make different bets. They can examine a system that everyone assumes is "perpetually complex" and ask: What is needed to achieve true automation? And, crucially, they can build the answers themselves.

The dominant force in the peak SaaS era was the rational optimizer under constraints. AI is removing those constraints and installing new ones. In this new environment, scarce resources are no longer about distribution, but about the ability to discern possibilities—and the aesthetic and conviction to build them to their proper standards. But there is a third variable that determines everything, and this is precisely where most founders of the AI ​​era are making disastrous mistakes.

Long-term games at high speed

A popular meme in the startup world right now goes like this: You have two years to escape permanent poverty. Build quickly, raise funds quickly, or exit or die.

I understand where this mentality comes from. The rapid pace of AI development creates a sense of existential crisis, and the window of opportunity to ride the wave seems extremely narrow. Young people who see stories of overnight fame on Twitter naturally assume that the essence of the game is speed—the winner is the fastest runner in the shortest amount of time.

This is correct in a completely wrong dimension.

Speed ​​of execution is indeed crucial. I firmly believe this—it's even etched into my company name (Warp). But speed of execution is not the same as short-sightedness. The founders who build the most valuable companies in the AI ​​era are not those who cash out after two years, but those who strive for ten years and reap the benefits of compound interest.

The mistake of short-sightedness lies in the fact that the most valuable elements of software—private data, deep customer relationships, real switching costs, and regulatory expertise—require years of accumulation and cannot be quickly replicated, no matter how much capital or AI capabilities competitors bring. When Warp processes payroll for interstate companies, we are accumulating compliance data across thousands of jurisdictions. Every resolved tax notice, every borderline case handled, every completed state registration trains a system that becomes increasingly difficult to replicate over time. This isn't a feature point; it's a moat, existing because we've cultivated it with such high quality for such a long time that it achieves quality density.

This compound interest is invisible in the first year. It's barely visible in the second year. By the fifth year, it's the whole game.

Frank Slootman, former CEO of Snowflake, who built and scaled more software companies than anyone else in existence, put it simply: get used to the "uncomfortable" state. Not for a sprint, but as a permanent state. The "fog of war" of early-stage startups—the disorientation, incomplete information, and the pressure to make action decisions—doesn't disappear after two years. It simply evolves, with new uncertainties replacing the old. The founders who endure are not those who find certainty, but those who learn to navigate clearly through the fog.

Building a company is incredibly brutal, a brutality difficult to convey to those who haven't done it. You live in constant, mild fear, occasionally punctuated by even greater anxieties. You make thousands of decisions with incomplete information, knowing full well that a single string of wrong decisions could lead to ruin. Those "overnight successes" you see on Twitter aren't just outliers in a power-law distribution, they're extreme examples of outliers. Optimizing your strategy based on these cases is like training for a marathon by studying the performance of people who stumbled through a 5km run on the wrong track.

So why do it? Not because it's comfortable, not because the odds are high. It's because for some people, not doing it feels like they're not truly living. Because the only thing worse than the fear of "building something from nothing" is the silent suffocation of "never trying."

And—if you bet right, if you see a truth that others haven't yet priced, if you execute with aesthetics and conviction over a long enough period—the results will be more than just financial. You've built something that truly changes the way people work. You've created a product that people love to use. In the business you've built, you've hired and empowered the best people to excel in.

This is a ten-year project. AI cannot change that; it has never changed.

AI is changing the ceiling that founders can reach in the last decade for those who are able to stick around and see the bigger picture.

The ceiling that no one pays attention to

So, what will the software look like on the other side of all this?

Optimists say AI creates abundance—more products, more builders, more value distributed among more people. They're right. Pessimists say AI destroys the moat of software—anything can be copied in an afternoon, defenses are dead. They're partially right too. But both sides are focused on the floor, nobody's paying attention to the ceiling.

The future will see thousands of point solutions—tiny, functional, AI-generated tools capable of solving specific, narrow problems. Many of these won't even be built by companies, but rather by individuals or internal teams to address their own pain points. For certain low-barrier-to-entry, easily replaceable software categories, the market will truly democratize. The bottom line will be high, competition fierce, and profit margins extremely thin.

But for business-critical software—the systems that handle cash flow, compliance, employee data, and legal risks—the situation is entirely different. These are workflows with extremely low tolerance for error. When the payroll system fails, employees don't receive their paychecks; when tax returns are incorrect, the IRS comes knocking; when benefits payments lapse during open enrollment periods, real people lose their coverage. Those who choose the software must be held accountable for the consequences. This responsibility cannot be outsourced to an AI cobbled together in an afternoon through "vibecoding."

For these workflows, enterprises will continue to trust vendors. Among these vendors, the "winner-takes-all" dynamic will be even more extreme than in previous generations of software. This is not only because of stronger network effects (although that is indeed the case), but also because the compounding advantage of an AI-native platform that has accumulated private data through large-scale operations, millions of transactions, and thousands of compliance edge cases makes it virtually impossible for latecomers to catch up in a "leapfrog" fashion. The moat is no longer a set of features, but the quality accumulated through long-term, high-standard operations in an area that punishes mistakes.

This means the software market will be more consolidated than in the SaaS era. I predict that ten years from now, the HR and payroll sectors will not have 20 companies each holding single-digit market shares. I expect two or three platforms to capture the vast majority of value, while a long list of point-in-the-box solutions will barely get a slice. The same pattern will occur in every software category where compliance complexity, data accumulation, and switching costs all play a role.

The companies at the top of these distributions look remarkably similar: founded by tech-savvy individuals with a genuine product aesthetic; built on an AI-native architecture from day one; and operating in markets where current giants cannot respond structurally without dismantling their existing businesses. They made a unique insightful bet early on—seeing a certain unpriced truth about what AI creates—and then held on long enough for the compounding effect to become clear.

I've been describing this type of founder in an abstract way. But I know exactly who he is because I'm striving to become him.

I founded Warp in 2022 because I believe the entire stack of employee operations—payroll, tax compliance, benefits, onboarding, equipment management, HR processes—is built on manual labor and outdated architectures, which AI can completely replace. Not improve, but replace. Established giants built billion-dollar businesses by absorbing complexity into their workforce; we will build our business by eliminating complexity at its source.

Three years have proven this bet. Since launch, we've processed over $500 million in transactions, are growing rapidly, and are serving companies building the world's most important technologies. Every month, the compliance data we accumulate, the edge cases we handle, and the integrations we build make the platform harder to replicate and more valuable to our customers. The moat is still in its early stages, but it's already taking shape and accelerating.

I'm telling you this not because Warp's success was predetermined—in the world of power-law distributions, nothing is predetermined—but because the logic that led us here is exactly what I've described throughout the text: See the truth. Dig deeper than anyone else. Establish a high standard that can be maintained without external pressure. Stick to it long enough to see if you're right.

In the AI ​​era, exceptional companies will be built by those who understand the following: access has never been a scarce resource, insight is; execution has never been a moat, taste is; speed has never been an advantage, depth is.

The power law doesn't care about your intentions. But it rewards the right intentions.