The Human Bottleneck: Why AI Won’t Eliminate Friction, It Will Redirect It
It is not all doom and gloom
Sometimes you really have to wonder what life was like before the internet, when commerce meant physically going to a store, when discovery meant walking through a mall, and when information asymmetry between buyers and sellers was simply accepted as a cost of doing business. The last two decades were genuinely transformative, not because technology eliminated human behavior, but because it rerouted it.
The question on everyone’s mind now is whether AI represents another such inflection point. As models have advanced rapidly across language, reasoning, and computer vision, a debate has emerged about what AI will actually change, and more importantly, what it won’t. The sentiment has shifted noticeably in recent months, from skepticism toward something closer to fear. Everyone is fearing that agentic AI will disrupt knowledge work, compress labor markets, and dissolve the familiar structures of commerce and society.
I want to make the argument that AI will be at least as revolutionary as the internet, and perhaps more so, especially in the domain of knowledge creation and research, but this does not mean it will reduce all friction. The crucial insight that tends to get lost in this debate is this: what determines friction in any market or human system is ultimately the human being at the center of it. Technology does not decide how much friction exists. Humans do. And in some cases, AI will not reduce friction at all — it will deepen and enrich it.
The Internet’s Lesson: Technology Serves Human Preference, Not the Other Way Around
To understand where AI is headed, it helps to revisit what the internet actually disrupted. Amazon is the canonical example. Bezos started with books not because he loved literature but because books are uniform in size, making logistics predictable and cheap. But books were really just a proof of concept for a much bigger idea that you could rebuild the entire architecture of retail around a warehouse model rather than a storefront model.
Traditional retailers are constrained by physical space in expensive, high-footfall locations — a Macy’s on Fifth Avenue can only stock so much, and every square foot of floor space carries a cost. This forces a constant, high-stakes gamble of predicing what consumers will want months in advance, manufacture it, ship it to shelves, and hope you got it right. Get it wrong and you’re either sitting on unsold inventory that you have to discount or discard, or turning away demand you can’t fulfill.
Amazon’s insight was that warehouses located outside of cities, where land is cheap, fundamentally change this equation. You no longer need to guess with the same rigidity, because you’re not constrained by expensive retail floor space. You can hold more SKUs, stock up dynamically when demand signals suggest a surge is coming, and replenish far more fluidly than any traditional retailer ever could. Combined with the data flowing in from online behavior — what people are browsing, saving, and buying — you shift from a guessing game to something approaching a science. Supply begins to track demand rather than anticipate it blindly, waste shrinks, and the economics of the whole operation improve dramatically.
What the internet solved, at its core, was an information asymmetry problem. Retailers operating blind were producing deadweight loss — excess inventory, mismatched supply and demand, missed sales. Better data collection meant you could move from guesswork to something closer to science. The result was a seismic shift: fast fashion, e-commerce, and ultimately ultra-fast platforms like Shein that respond to consumer trends in near real-time. Traditional retailers like Macy’s and Nordstrom, unable to compete on price, discovery, or convenience, collapsed under the weight of their own friction.
The critical insight the age of internet revealed is that not all friction disappeared. The stock market embraced algorithmic trading and commission-free brokerage almost immediately — frictionless transactions made obvious sense for a standardized, information-dense market. Yet real estate, another information-rich market, remains stubbornly dominated by brokers, agents, and negotiation. Why? Because buying a home is an emotional act. People want a human layer of trust, guidance, and accountability precisely because the stakes are high and the decision is deeply personal. The friction isn’t a bug in that market. It’s a feature that humans have chosen to preserve.
This is the insight that gets buried in most conversations about AI. The human is the bottleneck, and that is not going away.
Agentic AI and the New Information Layer
What agentic AI promises is the internet’s core value proposition — closing information asymmetry between buyers and sellers — but at a dramatically higher resolution. Today’s e-commerce still operates with significant blind spots. A consumer browsing Abercrombie & Fitch leaves behavioral traces, but the brand still doesn’t truly know who is behind the screen, what they’re genuinely willing to pay, or whether they’re likely to respond to a complementary recommendation. The consumer, meanwhile, is limited by their own awareness — they only discover what they already know to look for.
An AI agent changes this by acting as a digital twin on behalf of the consumer. Tell your agent you want a cashmere sweater in neutral tones between $500 and $1,500, and it canvasses the entire market, surfaces brands you’ve never encountered, and returns a curated shortlist. On the seller’s side, the agent understands not just what you’ve bought before but what your preferences are now, enabling targeting that is genuinely personalized rather than statistically approximated. This exchange between agents — buyer’s agent and seller’s agent — creates a far more efficient marketplace, one where supply and demand can approach true alignment, where obscure brands can access new audiences, and where product quality competes on its own merits rather than on brand recognition or marketing spend alone.
The second-order effect of this is that opacity becomes harder to sustain. Products that rely on information asymmetry — marketing-driven premiums over actual quality, brand prestige masking generic manufacturing — become more vulnerable when consumers have agents doing rigorous comparison on their behalf. This dynamic extends beyond retail into healthcare, education, financial services, and anywhere that knowledge gaps between provider and consumer have historically generated inefficiency or exploitation.
But the Human Remains the Bottleneck
Where the standard AI narrative tends to go wrong is that it assumes that because AI can eliminate friction, it will — uniformly, inevitably, across every domain. The history of the internet tells a different story.
The stock market runs at speeds no human can perceive because humans collectively decided that speed and efficiency were the only things that mattered there. Real estate hasn’t been similarly disrupted because humans collectively decided that the emotional weight of the transaction requires friction — requires a trusted intermediary, a negotiation process, a human hand to hold through the uncertainty.
Luxury goods are perhaps the clearest example. A Hermès bag is not a product, it is an experience, a social signal, a relationship with a brand. The friction of acquiring it, the waitlist, the boutique visit, the relationship with a sales associate, that is not inefficiency to be engineered away. That is the product. An AI agent that could instantly source a comparable luxury item at optimal price would be solving a problem that luxury consumers don’t have, because the journey is the point. AI in this context doesn’t reduce friction — it could actually be deployed to enhance and curate that experience, making it richer, more personalized, more exclusive.
This suggests a more nuanced framework for thinking about where AI will and won’t be transformative. AI will eliminate friction wherever humans have decided friction is wasteful, and it will enrich and deepen friction wherever humans have decided friction is meaningful.
Routine, low-emotional-weight transactions — grocery shopping, comparing insurance plans, sourcing commodity products — will trend toward frictionless automation. The human bottleneck in these domains is already yielding, because people have little attachment to the process itself. But high-stakes, emotionally resonant decisions, like buying a home, choosing a school, navigating a health diagnosis, purchasing art or luxury goods, will retain their human texture, and AI will more likely serve as a sophisticated support layer than a replacement for human judgment and feeling.
What Kind of Friction Do We Want?
There is a risk in the frictionless ideal that doesn’t get discussed enough. A world in which your AI agent knows your preferences and curates everything to match them is also a world of narrowing, of algorithmic bubbles, reinforced tastes, and the gradual erosion of the productive discomfort that comes from encountering something unexpected. Discovery, in the truest sense, often requires friction. Stumbling into a bookstore and leaving with something you never would have searched for is a friction-dependent experience. Whether agentic AI can preserve that kind of serendipity while eliminating wasteful inefficiency is one of the genuinely open questions about the technology.
What is clear is that AI will not be a universal solvent for human friction. It will be a powerful tool that humans will use, shape, and selectively resist according to their own values, emotions, and cultural preferences. The businesses that will thrive are those that understand this distinction, that recognize where their customers want efficiency and where they want to be held, guided, and delighted by the process itself. Those who mistake the elimination of friction for a universal good will find themselves optimizing away the very things that made them worth choosing in the first place.
The human is the bottleneck. That, more than any technical capability, will determine what AI actually changes.