Friction As a Feature, Not a Bug

Why the businesses that build for good friction are going to win.

Silicon Valley was built on the belief that friction is a problem to be solved. A key product design axiom holds that the best product is the one that requires the least from you — the least amount of waiting, effort, resistance — to get you from point A to point B as quickly as possible.

In fact, the history of consumer technology is largely a history of successful friction removal. The companies that did it best became the most valuable businesses in the world.

I have spent much of my career funding versions of this idea. But as I watched the AI world inherit that same logic — that friction is always something to be removed — the more I believe the premise is wrong.

For the wave of AI products now reshaping professional work, the pitch is some version of the same sentence: We are automating away the low-value work so you can focus on the high-value work.

In practice, the work being categorized as “low-value” and most susceptible to automation is often the work that contains friction: repetitive tasks, manual processing, procedural effort. In finance, that low-value work can mean eliminating manual reconciliation. In law, it can mean no longer drafting routine contracts from scratch. In medicine, it can mean AI handling documentation so physicians can spend more time with patients.

This pitch resonates — and has recently catapulted so many AI startups to billion-dollar valuations — because many industries are indeed burdened by repetitive work that consumes time without building skill, judgment, or understanding. Removing that type of work can be useful.

But the pitch also smuggles in a flawed assumption: that all friction is the same kind of friction. That whatever stands between a person and their desired outcome is, by definition, without value. I could not disagree more.

Purposeless Friction vs. Desirable Difficulty

There are two types of friction and they should not be conflated.

The first kind of friction — purposeless friction — might look like unnecessary extra steps or meaningless tasks that people don’t learn or grow from. It drains energy, capability, and interest and feels extractive instead of additive as people go through the process. Removing this kind of friction is objectively good because people feel relieved, or simply do not miss it, when it’s gone.

Alternatively, cognitive psychologists have a name for another kind of friction, one that is both productive and necessary. This is known as desirable difficulty (coined by UCLA professor and cognitive psychologist Robert Bjork). The principle is essentially that when learning is too easy, the brain processes information passively. In contrast, the right kind of struggle forces deeper encoding, stronger retention, and better transfer to new situations. It argues that difficulty (or friction) is intrinsic to learning.

On the surface, these two types of friction can look very similar. Both can involve the same amount of effort, the same amount of time, the same apparent inefficiency. But the resistance provided by desirable difficulty results in something very different. When you go through this type of friction, you are not merely going through the motions. Instead, the resistance is the point. You grow from it. You develop judgment, intuition, and understanding by facing it.

Many of the tasks now dismissed as “low-value” in the AI age may actually provide desirable difficulty. They are slow, repetitive, and easy to undervalue from the outside, but they often force the kind of close contact with the work through which judgment develops. Made to manually write hundreds of similar contracts, a junior lawyer learns to identify risk and ambiguity. The financial analyst building models by hand develops intuition for business mechanics, leverage, and value. The physician writing notes is often not merely documenting care, but processing, organizing, and clarifying their own thinking. What looks like inefficiency can be the training ground for expertise.

The problem is that AI often does not index for desirable difficulty. In indiscriminately working to remove purposeless friction, it often takes the good kind of friction with it. By collapsing the steps between ideation and execution, by shifting workflows from manual processing to agentic orchestration, AI produces a frictionlessness that can benefit people in meaningful ways. But it can also cultivate the bad habit of friction-avoidance — a disinclination to do the hard thing because an easier path is always available.

The Civic Version of the Problem

The pitfalls of frictionlessness impact not only us as individuals but also us as a collective.

Democracy depends on citizens doing certain kinds of cognitive and social work for themselves: Forming views through deliberation rather than passive consumption, encountering opposition and learning to respond to it, and weighing competing claims instead of receiving frictionless certainty on demand.

These processes are often slow, uncomfortable, and effortful, but they are also how democratic capability is formed.

AI offers a seductive alternative to this effort, which is a world where information, interpretation, and even judgment arrive pre-processed. Personalized AI systems can smooth away the friction of encountering disagreement, uncertainty, or ideas that require sustained engagement. They can make it easier to consume conclusions without participating in the harder process of arriving at them.

The philosopher Mark Coeckelbergh describes this as an erosion of epistemic agency: the capacity to form and revise beliefs through active engagement with the world and with other people.

Democracy is a system reliant on good friction. It asks citizens to endure processes that are often inefficient, adversarial, and frustrating because those processes are part of what teaches pluralistic societies how to function.

When every surrounding system optimizes for frictionless personalization, convenience, and ease, democratic participation itself can begin to feel unnecessarily demanding by comparison.

What This Means For Builders

None of this means friction is inherently virtuous. But we do need a more nuanced understanding of friction, because the tools being built right now are going to shape nearly every dimension of human life: how people learn, work, form opinions, make decisions, relate to one another, and participate in their communities.

And as AI removes more friction from daily life, people already seem to be searching for certain forms of it elsewhere. Part of what sits underneath the analog resurgence — younger generations gravitating toward vinyl records, film, photography, physical notebooks, running clubs, and record players — is not nostalgia so much as a desire for products and experiences that feel participatory rather than frictionless. With all of these products and experiences, effort is needed to uncover value.

What is built around shared difficulty is fundamentally different from what is built around passive consumption. The attachment is deeper because something is required of you — whether that’s time, investment, tenacity, or a willingness to fail — and you are changed in the process.

The AI products that win will be the ones that understand this. At Principal Venture Partners, it has become a lens we apply when evaluating early-stage companies: does this product understand the difference between friction that blocks people and friction that helps them grow? One of our portfolio companies, Constella, is built on the premise that organizing your own thinking is itself valuable. Its product helps users connect concepts, revisit prior thinking, and surface patterns over time in ways that deepen reflection and expand understanding. Another portfolio company, Laurel, automates the administrative burden of timekeeping for professionals while preserving the parts of the work where judgment and expertise are actually formed.

Both businesses are built around the same idea: that including the right kind of friction as a feature — rather than dismissing it as a bug — can be a competitive edge.

After all, the things that hold the most value are rarely the things that require the least from us.