What every domain calls 'friction' — cognitive struggle, verification checkpoints, energy buffers, governance architecture, temporal formation — is load-bearing infrastructure whose removal creates structural collapse.

The Tension

## THE TENSION

The tension is brutal because removing friction works. In the short term, it works spectacularly.

AI productivity research confirms this. Marginal Revolution's analysis shows AI gains concentrate in greenfield projects where there is no legacy structure to navigate. New code, new teams, clean slate. Gains are real. But legacy integration resists, because legacy systems have load-bearing friction baked into every interface. The outsourcing firms everyone predicted would die are not dying. They are the friction layer between AI output and production reality.

Energy markets show the same split. Oil majors are bidding record prices on Alaska petroleum leases, making 5-to-10 year bets at $112 oil. They are pricing in structural undersupply. The IEA calls a Hormuz closure "the greatest global energy security threat in history". Decades of lean inventory, just-in-time supply, and efficiency optimization eliminated the buffers that used to absorb shocks. The friction was the resilience.

Peter Attia's analysis of Mendelian randomization captures the epistemic version of this tension. Sophisticated models applied to flawed premises produce structurally misleading conclusions with high confidence. The method looks frictionless. The output looks precise. But the verification friction that would catch the bad premise was never applied.

Here is the trade-off every builder faces: speed to market versus structural resilience. Every friction you remove accelerates delivery and removes a load-bearing wall you will not notice until the earthquake. The question is not whether to have friction. The question is which friction is structural and which is genuinely waste.

What This Unlocks

## WHAT THIS UNLOCKS

a16z published an analysis of open agentic commerce killing the advertising model. When AI agents handle purchasing decisions, they route around the friction of human attention, search, and comparison. Stack Overflow traffic is down 75%. Tech news readership is down 60%. The friction those platforms monetized was the user's need to search, read, and decide. Remove the user from the loop, and every business model built on human attention friction collapses.

This is the second-order consequence. Removing friction does not just degrade the user. It degrades every system downstream that depended on friction-as-signal. Advertising, editorial, curation, review, certification. These are all friction-dependent business models.

Winners: companies that own verification infrastructure. Testing pipelines, safety certification, audit systems, compliance architecture. These are the new chokepoints. When generation becomes free, as we discussed in the March 19 issue, the value migrates to the verification layer. Today's evidence shows that migration is already happening in security, energy, and governance simultaneously.

Losers: anyone positioned as a friction-removal service without owning the verification that follows. If your product pitch is "we make X frictionless" and you have no answer for what carries the load after, you are building on a shrinking foundation.

The US war strategy in Iran demonstrates this at geopolitical scale. The same administration sent "winding down" and "obliterate" messages on the same day. No exit conditions defined. 50,000 troops deployed. The deliberation friction that normally precedes large-scale military commitment was removed. What remains is escalation without architecture.

For builders: if you are designing AI-augmented workflows, the competitive moat is not in the AI layer. It is in the verification, governance, and judgment layers that sit between AI output and consequential action. Build those. Own those.

Watching Next

## WATCHING NEXT

Three observables, all structural.

First, AI coding failure rates at module boundaries versus interior logic. The QCon data shows failures concentrate at interfaces, not inside modules. If formal verification tools start targeting inter-module contracts specifically, that is the market recognizing friction-as-infrastructure. Watch for startups raising on "AI interface verification" positioning in Q2.

Second, energy buffer rebuilding timelines. If OECD petroleum reserves do not begin rebuilding within 90 days of any ceasefire or de-escalation, the market has priced in permanent undersupply. That changes the capital allocation picture for every energy-dependent business.

Third, and this one you can check in your own company: count how many human review steps sit between AI-generated output and customer-facing delivery in your pipeline. If that number has decreased in the last six months without a corresponding increase in automated verification, you have removed load-bearing friction. The failure will not be gradual. It will be sudden and correlated with the first novel edge case your system encounters.

Underweighting

## WHAT I MIGHT BE UNDERWEIGHTING

I think the friction-as-infrastructure frame is correct, but I might be wrong about the time horizon.

It is possible that AI systems will develop internal verification capabilities fast enough to replace the human friction they are removing. The frictionless AI psychology study measured weeks-to-months of degradation. But AI capability curves move on weeks-to-months timescales too. If autonomous verification matures before human skill atrophy becomes irreversible, the net effect could be positive. The friction gets relocated, not eliminated.

I might also be imposing a unified frame on signals that are structurally independent. Energy buffer depletion and cognitive skill atrophy share a formal similarity. Both involve removing reserves that absorb shocks. But the causal mechanisms are completely different. Oil is a physical commodity with geological constraints. Cognition is biological with neuroplastic recovery potential. Treating them as instances of one pattern might feel elegant and be misleading.

Stephen Batchelor's argument that solitude is an acquired art requiring sustained practice reinforces the friction frame. But it also suggests that friction might be a cultural preference, not a structural necessity. Some traditions value effortlessness. Wu wei is not laziness. It is friction removed through mastery, not avoidance.

The strongest counter: prediction markets outperformed fact-checking during wartime disinformation. That is a case where removing editorial friction and replacing it with financial incentive produced better epistemic outcomes. Not all friction is load-bearing. Some is just slow.

Bottom Line

## BOTTOM LINE

Friction is not the enemy of performance. It is the skeleton of it. Every system that removed its verification layer, its energy buffer, its deliberation process, or its cognitive effort is now discovering what that friction was holding up. The question for builders is not how to move faster. It is which walls in your system are load-bearing.